Gpt/LLM/GenAI

Ce qui est entré en base, sans jugement.

Montrer Tout Porteurs de lien
3065posts
36auteurs
518liens
1470 hde recul
38membres
Aucun recoupement sur 24 h pleines : ni un lien ni un tweet n'est relayé par deux membres (2 liens, 13 tweets relayés). Le recul est suffisant, donc cette fois le constat porte : ces comptes diffusent au lieu de converger, et c'est la composition de la liste qu'il faut revoir — pas la patience qu'il faut allonger.
T
OpenAI Developers a relayé @xikotadesign
Thiago Xikota @xikotadesign · 00:32
/goal Reúna os builders do Brasil em São Paulo. Na sexta rolou o OpenAI DevDay Exchange com a @openaidevs: adesivo lindo, cartinha personalizada, palestra muito boa e umas mil pessoas no salão. O que você vai construir? #codex
M
OpenAI Developers a relayé @renatarsilvax
Maki @renatarsilvax · 00:31
Agora, eu e a @krisnarane1 sendo as maiores blogueiras do DevDay[Exchange(São Paulo)] @OpenAI 😚😌🥰🤩 cc: @matheusgobbidev
D
OpenAI Developers a relayé @danizeres
dani bauer @danizeres · 00:31
On Friday I spoke at @AfroTech in Houston, where Black founders, creators, and technologists share what they’re building. I designed my session for the person making the content and running the business behind it. We walked through how to use ChatGPT to: – Find partners who fit
Nous Research @NousResearch · 00:30
Using an agent that can change its models or code at any time without telling you is like building on quicksand
Greg Brockman @gdb · 08/11 23:42
frontier models, with lots of test time compute, are becoming the engine of scientific discovery
P
Pete ✈️ CoRL a cité @theSamPadilla
Pete ✈️ CoRL @nonmayorpete · 08/11 21:38
Please DM me for insane amounts of off-the-shelf stereo
Yann LeCun a relayé @ylecun
Yann LeCun @ylecun · 08/11 21:10
@TimDarcet @gabriberton The P in JEPA stands for predictive. The main distinction is not between JEA and JEPA, but between Joint Embedding (predictive or not) and generative architectures. Whether a predictor is necessary depends on the task. Given observations x and y (y being available only during
OpenAI Developers @OpenAIDevs · 08/11 21:04
Tomorrow.
elvis @omarsar0 · 08/11 20:15
I've seen many make this mistake over and over again. Overrelying on someone else's harness. As you go deeper with agents, you quickly realize how standard, out-of-the-box harnesses constrain LLMs. Imo, agent harnesses need customization to unlock real gains and
Google AI Studio @GoogleAIStudio · 08/11 19:40
What are you vibe coding this weekend?
Jeremy Howard a relayé @math_rachel
Rachel Thomas @math_rachel · 08/11 18:44
"When two things are different, you should not use one to mimic the other. It would've been a colossal mistake if Phil Jackson had pushed Jordan to mimic Pippen & Rodman. "The same applies to AI & human skills. Using AI to expand what humans do is more productive" @DAcemogluMIT
Yann LeCun a relayé @TheTuringPost
Turing Post (Ksenia Se) @TheTuringPost · 08/11 18:28
There's one great example that shows how physics can help upgrade AI world models, and it's LeHami. @ylecun, @randall_balestr and @RominaYalov brought Hamiltonian mechanics into JEPA to preserve phase-space volume and prevent latent collapse during long-term predictions. Why
S
Yann LeCun a relayé @iamsupersocks
Supersocks @iamsupersocks · 08/11 18:16
Arthur Mensch vient de poser sur LinkedIn la phrase que les députés devraient lire avant le 19 novembre : quand Mistral choisira entre un data center en Essonne ou hors de France, la loi Darcos pèsera dans la balance. Donc le Sénat a voté le texte à l'unanimité le 8 avril : dans
D
Yann LeCun a relayé @DeBunKerEtoiles
Debunker des Etoiles @DeBunKerEtoiles · 08/11 18:06
Attention, quelles que soient vos opinions politiques par ailleurs, cette mesure de LFI est extrêmement nocive pour la France en terme écologique, géopolitique, et économique. L’Allemagne a fait pareil et c’est un fiasco, qui lui a coûté très cher. Elle s’est retrouvé dépendante
elvis @omarsar0 · 08/11 17:55
Impressive way to reuse agent trajectories to evolve agent harnesses. Given how good LLM agents are getting at context awareness, this is worth a look. This idea isn't new, but it's becoming an important pattern in harness engineering: enabling the agent harness to reuse
elvis a cité @pushmeet
elvis @omarsar0 · 08/11 17:08
Been working for this moment for over a decade. What I've witnessed behind the scenes, together with what we have been building, tells me we are about to enter a truly incredible era of AI-powered scientific discovery. It's an incredible time to be doing research. I will be
J
Jim Fan a cité @DrJimFan
Jim Fan @DrJimFan · 08/11 16:42
Real robot. Real time. Uncut shot. Autonomous. Tomorrow.
S
Harrison Chase a relayé @simon_budziak
Simon Budziak @simon_budziak · 08/11 16:33
Everyone is converging on the same agent architecture. An LLM in a loop, calling tools, inside a sandbox. That does not mean the hard problems are solved. @hwchase17 at @LangChain lists what is still open for always-on agents. A proactive agent has to help without annoying
elvis a cité @brian_armstrong
elvis @omarsar0 · 08/11 05:08
Decrease the number of touchpoints with agents and instead improve the coordination with and among agents. Add a queue and a good memory system and watch how much faster and efficient you become. Any other setup from what I have tested in my custom harness will hold back agents
elvis a cité @RichardSSutton
elvis @omarsar0 · 08/11 01:26
1000% agree with Sutton here. Having worked in AI for over a decade (pre-LLMs), it's clear we need smarter methods for intelligence to truly compound in the way some of us envision it. To me, Transformers just scratched the surface. This arc is coming to an end; a new
François Chollet a cité @RichardSSutton
François Chollet @fchollet · 07/11 23:52
Data scaling and compute scaling are *multipliers* for the generalization power of an approach. Trillions of dollars in investment can make very inefficient, low-generalization approaches look adequate, but that doesn't make them the ultimate end-all-be-all of AI. You should
Greg Brockman a cité @patopet
Greg Brockman @gdb · 07/11 22:25
security architecture fundamentals are more important than ever before
Yann LeCun a relayé @ylecun
Yann LeCun @ylecun · 07/11 20:04
@gabriberton Several comments: 1. image/video SSL methods differ by whether they use reconstruction or not. Joint embedding architectures Amy or may not have a predictor (JEA vs JEPA). They also differ by their anti-collapse method and data corruption/augmentation methods. 2. DINO *is* a kind
J
Jim Fan @DrJimFan · 07/11 18:14
CoRL Keynote. Hilton Austin. Monday Nov 9, 1:45 pm. 2 days countdown. Something extraordinary.
elvis @omarsar0 · 07/11 16:14
Great paper from Google on making ML experiments reproducible for coding agents. (bookmark it) If you run agents on research code, this one is worth your time. They find that refined artifacts raise agent reproduction success from 76.4% to 92.6% across four models and cut API
elvis a relayé @dair_ai
DAIR.AI @dair_ai · 07/11 16:07
Great paper finds that MCP agents rarely say that tasks are impossible. They propose MCPHallu, which runs 358 tasks on 36 production MCP servers. Each task is built to trigger one failure type, such as branch collapse, an unreachable goal, tool misuse, or forgetting context.
OpenAI Developers a relayé @fedesarquis
Fede @fedesarquis · 07/11 16:07
@pawelk411 @OpenAIDevs You can try the prompt yourself! Here: Create a seamless 4–5-second looping GIF celebrating [BRAND NAME], using the attached logo as the exact shape reference. Center the logo on a near-black navy background. Rebuild it entirely from visible ASCII characters: @ # % * + = 0 1.
hardmaru @hardmaru · 07/11 09:37
There’s a manga about Alan Turing at Tsutaya bookstore
Hugging Face a relayé @RisingSayak
Sayak Paul @RisingSayak · 07/11 09:10
The @huggingface crew is here in Doha till 7th for a grand Hackathon in collaboration with QAI. We're excited to be in this part of the globe and build something cool!
P
Pete ✈️ CoRL @nonmayorpete · 07/11 07:58
#CoRL2026 is going to be absolutely bonkers next week Loading up on Halls and hand sanitizer as a precautionary measure
E
Jason Wei a relayé @edwardjhu
Edward Hu @edwardjhu · 07/11 05:35
A task ≠ a job. Instead of working with just tasks and tools, the next frontier of AI evals puts a model inside an org chart. The model needs to navigate the organization to get things done. We built the foundational taxonomy for this next frontier of evals:
LlamaIndex 🦙 a relayé @jerryjliu0
Jerry Liu @jerryjliu0 · 07/11 05:27
The world needs two types of document OCR: 1. Extremely fast/cheap OCR tools: Each organization might have trillions of documents, and this is oftentimes helpful as a 'first pass' to digitalize each document so that it's accessible to AI agents. Accuracy might not be the
elvis a cité @AnthropicAI
elvis @omarsar0 · 06/11 23:49
Highly recommended read. Learn how one of the most advanced agent memory systems works. Your memory system is one of the biggest levers you have in the harness. I also don't think a one-size-fits-all memory system exists today. Memory is something you constantly have to
elvis @omarsar0 · 06/11 23:32
Great weekend read from Stanford on measuring AI agents as teammates. (bookmark it) If you build multi-agent systems or build with agent teams, this one is worth your time. They try to measure one of agents' most important capabilities: collaboration. They propose the
OpenAI @OpenAI · 06/11 20:58
We partnered with @fmf_org and its member firms on a new framework to help identify when bypassing an AI system's safeguards could enable serious cyber harm, and prioritize stronger protections. "...This framework is an important step toward a common methodology for evaluating
ClaudeDevs @ClaudeDevs · 06/11 20:58
We've made it easier to get your repo running in Claude Code cloud sessions on Pro, Max, and Ultra. Claude figures out what dependencies, network access, and API keys it needs, then saves the environment for next time. API keys are saved securely, so Claude never sees them.
Anthropic @AnthropicAI · 06/11 20:49
New on our Science Blog: How AI is beginning to change scientific modeling. Scientific models help researchers explain and predict real-world systems—like how a single nerve cell fires, or what happens when neutron stars collide. Recent advances have transformed how scientists
Harrison Chase a relayé @caspar_br
Caspar @caspar_br · 06/11 20:36
Managed Deep Agents can now be steered mid-turn. If you send a message while the agent is working, it reads it at its next model call instead of waiting for the turn to finish. Small change that makes agents much, much nicer to work with
E
ElevenLabs @ElevenLabs · 06/11 20:11
Andy Kelly, EVP of Global Brand Partnerships at @iHeartMedia, is coming to ElevenLabs Summit NYC on November 11. He joins Dustin Myers of @FOXSports and Alex George for an ElevenProductions conversation on bringing AI content to audiences around the world. Apply to join:
OpenAI @OpenAI · 06/11 20:08
A new way to measure how AI agents handle the routine analyses of wet-lab science—BioBasicsBench. Developed with Ligo, a biotech company focused on enzyme design, the benchmark draws on 100 tasks from real experiments to test whether agents can interpret raw lab data, identify
Hugging Face a relayé @Thom_Wolf
Thomas Wolf @Thom_Wolf · 06/11 19:44
By the way you should follow Augustin on X - I’m trying to get him to post more - it’s one of the most cracked roboticist I’ve ever met
J
Hugging Face a relayé @jeffboudier
Jeff Boudier 🤗 @jeffboudier · 06/11 19:43
We spent $2m so you don't have to. And so you can learn how to train MoEs. Mixture-of-Experts LLMs are what powers the AIs you use everyday, and few people in the world know how to train them at scale (a handful at each top AI lab). This handbook distills everything we
Hugging Face a relayé @ClementDelangue
clem 🤗 @ClementDelangue · 06/11 19:43
Open-source is critical for robotics AI. It means a student, a researcher or a startup can build on years of progress and contribute something the whole community can use rather than concentrating everything in a few companies building a black-box robot. That’s why we’re
G
Rowan Cheung a relayé @guyoseary
Guy Oseary @guyoseary · 06/11 19:31
Rowan is the founder of The Rundown AI, the world's largest daily artificial intelligence newsletter with an audience of over 3.5 million people. We are honored to have him join the @sound_ventures_ family.
M
OpenAI Developers a relayé @AzianMike
Michael Luo @AzianMike · 06/11 19:21
We shipped a bunch of stuff in Agents API in October! And make sure you read to the end for some beta features 👀 The full list: - auto review: automatically review and approve or deny supported agent actions, similar to Codex’s “review for me” - automations: you or the
A
Rowan Cheung a relayé @alexeheath
Alex Heath @alexeheath · 06/11 19:08
Welcome Rowan, architect of the biggest AI newsletter in the world, to Sound Ventures! We at Sound have a view that how you tell your story will only matter more and more as code becomes cheaper to create and capital becomes abundant. More to come...
R
Rowan Cheung @rowancheung · 06/11 18:31
Personal news: I’m joining Sound Ventures as an advisor. Fun story, Guy Oseary first reached out back when The Rundown had 150K subscribers. I had no idea what I was doing, and I’m still not sure why he took the call... But we’re at 3.5M now, and I’ve learned he has a real
LlamaIndex 🦙 a relayé @imaanxsultan
Imaan Sultan 🇵🇰🇸🇦 @imaanxsultan · 06/11 18:05
the two best times to drink chai ☕️: 1/ diwali weekend 2/ when the sleeve has $500 in LlamaParse credits on it this weekend you get both unindexed ep 4 is a bts of our doc chai launch party at @UseCorgi. we were trending in the bay on both luma and partiful, with 1,000+ RSVPs
J
Jim Fan @DrJimFan · 06/11 17:14
I'm giving a Keynote talk at CoRL next Monday Nov 9, please come by at 1:30 pm at Hilton Austin! We've been cooking 👀
elvis a relayé @dair_ai
DAIR.AI @dair_ai · 06/11 17:05
If you want to learn how to build self-improving harnesses, check out this paper. They present Exo, an agent runtime architecture for in-harness learning and adaptation during ongoing work. Harness optimization usually runs offline, so it needs a representative dataset and many
Hugging Face a cité @yb2698
Hugging Face @huggingface · 06/11 15:43
Stream live training runs on Hugging Face!
elvis @omarsar0 · 06/11 15:43
Weight memory vs. retrieval in agents. Brilliant work from Amazon. This work shows weight memory can outperform text memory on recall of user facts. But the best part is that you can leverage both. Using the same model and user data, this work shows that plain fine-tuning
elvis a relayé @dair_ai
DAIR.AI @dair_ai · 06/11 15:36
Brilliant paper on optimizing agent environments. This is an extremely important skill if you work with agents today. You can make a lot of improvements by optimizing the environments agents operate in. Many agents work inside a shared workspace, such as a company's Slack or a
A
OpenAI Developers a relayé @aerdna_c
Andrea Cau @aerdna_c · 06/11 14:16
This started with something I noticed at home over the summer: the air wasn’t moving through the house the way I’d hoped. It made me wonder whether I could turn a floor plan into an interactive way to explore airflow, considering the house’s orientation and where its doors and
elvis @omarsar0 · 06/11 14:16
Banger report on foundation decision models. This work presents a 27B open decision model that lands within 1.6 points of Claude Opus 5.5 on Decision Bench v4 and answers in 183 ms instead of 3.5s. Nace AI released Drex, a family of four open decision models at 8B, 9B, 35B with
OpenAI Developers a relayé @higgsfield_ai
Higgsfield AI 🧩 @higgsfield_ai · 06/11 13:56
GPT Image 2.6 made a video game out of a dog GIF. Made on Higgsfield.
A
OpenAI Developers a relayé @alexbiemann
Alexander Biemann @alexbiemann · 06/11 13:55
White cardboard + projector + webcam = tabletop arcade! 🦆 Duck Hunt in Project Tabletop 1.4.1: a saloon shooting gallery, flying ducks and clay pigeons. The camera tracks your laser; the mic hears your click. Built with #Codex ❤️ #GameDev @OpenAIDevs
Hugging Face a relayé @Thom_Wolf
Thomas Wolf @Thom_Wolf · 06/11 11:13
18 of the 20 best open models today are Mixture-of-Experts. Dense is basically over. And yet the know-how to train a good MoE still lives mostly in the heads of a handful of teams. So we spent 800k GPU hours running the experiments ourselves and wrote everything down. Super
P
Yann LeCun a relayé @PolymathicAI
PolymathicAI @PolymathicAI · 06/11 09:10
Interested in building what’s next in AI for Science? Join us at PolymathicAI in New York (NYU + Flatiron). We are looking for: Interns, Postdoctoral Researchers, Research Scientists and Research Engineers. Submit your interest here:
forms.gle https://forms.gle/EMfMccvPZ3S19BLD6
Nous Research a relayé @Teknium
Teknium 🪽 @Teknium · 06/11 08:21
Hermes supremacy tweet #9138
Nous Research a cité @liamzebedee
Nous Research @NousResearch · 06/11 07:24
That's hilarious but you should probably try Hermes which doesn't do that
Mira Murati a relayé @cHHillee
Horace He @cHHillee · 06/11 05:30
Our infra has really improved quite a bit since the last time we released a model - this one took just under 3 days to pretrain! Also, it's quite cheap on Tinker (0.30/1.00/1.50 for in/out/train)! Hope you guys find this model great for customization purposes :)
A
Mira Murati a relayé @_alex_kirillov_
Alexander Kirillov @_alex_kirillov_ · 06/11 05:21
The model factory is model factoring. Super proud of the team!
elvis a relayé @omarsar0
elvis @omarsar0 · 06/11 05:15
Own your intelligence stack, folks! We've reached the point where it's simply not possible to do reliable groundbreaking AI research (using autoresearch loops) with GPT or Claude. I constantly get hammered with inaccurate refusals. This is sad! They should at least spend time
Mira Murati a relayé @johnschulman2
John Schulman @johnschulman2 · 06/11 04:18
Proud to release our new model, Inkling 2 Small. Check out our blog post -- it includes a lot of detail about our training process. To me, the most interesting part of this release was what we learned about alignment, which you can read about in the sections "Safety and
D
OpenAI Developers a relayé @devteamdrew
DreW @devteamdrew · 06/11 03:37
A few scenic GIFs made using ChatGPT Images 2.6
P
Pete ✈️ CoRL @nonmayorpete · 06/11 02:40
Lol. Lmao, even.
P
Yann LeCun a relayé @PhillipsPOBrien
Phillips P. OBrien @PhillipsPOBrien · 06/11 02:06
Watching all these Republican attempts to explain why they lost the election is painful, because they cannot come out and say one simple truth. They handed the party to a corrupt, criminal, dictator-loving, economic and strategic idiot--who proceeded to be all of these things.
Yann LeCun a relayé @SamuelFitouss10
Samuel Fitoussi @SamuelFitouss10 · 06/11 02:05
- Je ne vois AUCUNE raison de continuer à entraîner des modèles d’IA toujours plus puissants. - Guérir le cancer ? - Oui bon. Mais à part guérir le cancer ? - La possibilité de vieillir en bonne santé, sans oublier qui sont ses enfants, sans porter des couches ? - Admettons. À
A
Harrison Chase a relayé @amal_irgashev
amal @amal_irgashev · 06/11 00:51
go async or go home in all seriousness this is a fantastic implementation of async subagents. its being used for all of our internal agents and the UX in slack is so much better
Mira Murati a cité @thinkymachines
Mira Murati @miramurati · 06/11 00:45
Inkling 2 Small is available today on Tinker, with weights coming soon. It's a capable small model. We look forward to seeing what you make with it.
F
Sam Altman a relayé @fouadmatin
fouad @fouadmatin · 05/11 23:34
Today’s GPT-6 Cyber release is a meaningful step for defenders: stronger vulnerability research and better alignment. We’ve already been using GPT-6 Cyber alongside GPT-6 Astra with Daybreak to harden our own systems, finding weaknesses, validating their impact, and developing
Harrison Chase a cité @sydneyrunkle
Harrison Chase @hwchase17 · 05/11 23:03
most of the agents we're building inside langchain leverage async subagents in a meaningful way now available in deep agents!
Harrison Chase a relayé @sydneyrunkle
Sydney Runkle @sydneyrunkle · 05/11 23:03
as we ask agents to do more work for longer, async subagents become a very important primitive async subagents enable - continuous human interaction w/ the main orchestrator, since it's not blocked by subagent work - efficient distribution of work across the background workers
elvis a cité @Davidstout
elvis @omarsar0 · 05/11 22:30
I am bullish on this vision. I've been using smaller models on my local machine to scale my agent teams. Powerful small reasoning models running locally can unlock all kinds of workflows, such as AI R&D and custom harness optimization. Love what this team is building. Do
Greg Brockman a cité @i0n1c
Greg Brockman @gdb · 05/11 22:07
gpt-6 cyber is a good model
Nous Research a cité @0xdippo
Nous Research @NousResearch · 05/11 21:48
Unfuck your hardware with Hermes
elvis a cité @shobhitbanga
elvis @omarsar0 · 05/11 21:24
A model being great at US English says little about how it ranks in other languages like Brazilian Portuguese. These benchmark efforts matter a lot for voice. The per-use-case benchmark is also ideal for properly measuring voice agents' effectiveness and robustness in certain
E
OpenAI Developers a relayé @e1daru
Eldar @e1daru · 05/11 21:08
Thank you so much to everyone who came to the DevDay hackathon! @Andy_AJT and @oscarfalll keep levelling up OpenAI events in London, and we can't wait to see more 🚀🚀🚀 I was honoured to judge alongside our great team 🙌 With 100 incredible submissions and 100% great vibes, it
LlamaIndex 🦙 @llama_index · 05/11 21:00
LiteParse integrates with MiniCPM-OCR from @openBMB 🚀 MiniCPM-OCR is a new lightweight OCR model that performs OCR after capturing cropped layout regions. On ParseBench, it beats Gemini 3.5 Flash Lite and PaddleOCR-VL-1.6 on content faithfulness. LiteParse is our ultra-fast
Hugging Face a cité @abidlabs
Hugging Face @huggingface · 05/11 20:50
Our open-source ML Intern is now available as a Slack bot you can install in your own workspace! Evaluate models, build datasets, and fine-tune models on the fly 🚀
elvis a cité @frydwia
elvis @omarsar0 · 05/11 20:29
AI gets much more useful when it shows up where the work already lives. A lot of that work still sits in email. Viktor now has his own email address. Cc him on a team thread or forward him the work, and he does it right there in the thread. Just email viktor@viktor.com to get
A
elvis a relayé @AlexAtServal
Alex McLeod @AlexAtServal · 05/11 20:08
Starting today, every engineer who joins Serval begins with a weeklong, $10,000 paid vacation. We’re building the next generational company. We need generational talent to see it through. Our mission has always been to empower people across every team to turn their understanding
Greg Brockman a cité @firesidealpha
Greg Brockman @gdb · 05/11 19:31
Announcing our strategic partnership with JPMorganChase, to explore how advanced AI can support their enterprise: helping engineers build, security teams defend, and employees and customers benefit from more capable AI. I'm especially excited about working together on
T
OpenAI Developers a relayé @matt_spendlove
the_matador @matt_spendlove · 05/11 19:27
I've been using Codex to build a sculpting tool for my own creative experiments on Apple Vision Pro. Here's a little sketch, captured from outside the headset.
Greg Brockman a cité @ky1ebot
Greg Brockman @gdb · 05/11 19:14
team is doing incredible work to help secure Internet infrastructure
Greg Brockman a cité @OpenAI
Greg Brockman @gdb · 05/11 18:31
GPT-6 Cyber: a major advance for security research, already being used to help defend security-critical software such as the Linux kernel, Firefox, and many other projects. We’ve invested a great deal in trust, specifically training the model to stay within the scope specified
elvis a cité @aitonomyfound
elvis @omarsar0 · 05/11 18:19
Can AI agents do scientific research on their own? Science's Last Exam is a new benchmark measure this. It's built from real research tasks across nine scientific fields. It tests whether an agent can form hypotheses, plan protocols, run scientific code and software, and read
Perplexity a relayé @AravSrinivas
Aravind Srinivas @AravSrinivas · 05/11 18:10
Today, we're launching Perplexity Teammates. They are personal agents for work. And change the paradigm from writing prompts in a session to giving high-level goals. Rolling out to Max users on the web today. Native apps (mobile, desktop) coming soon.
Google AI Studio @GoogleAIStudio · 05/11 18:08
introducing gemini omni 1.2 flash, now available in ai studio and the gemini api what's new: - fast & 30% lower cost - longer clips - edit in conversation - more natural speech - more controllable shots
OpenAI Developers a relayé @natalieyeo
Natalie @natalieyeo · 05/11 17:53
A little bit of CAD and Astra makes it neat and tidy
LlamaIndex 🦙 @llama_index · 05/11 17:37
Going to AWS re:Invent and looking for the best afterparty of the week? We're partnering with our friends at @Braintrust, @Cartesia, @p0, @turbopuffer, and @Modal to host the Agent Soundcheck. Join us for drinks, food, and karaoke at On the Record at the Park MGM. Tue, Dec 1,
elvis a cité @lazyapprove
elvis @omarsar0 · 05/11 17:21
Fascinating how proactive personal agents are pushing fast innovation in AI hardware interfaces. Aina is a ring with a tiny screen that shows you the one thing your agent is about to do. You glance and approve. Your Uber gets booked, the "running late" text goes out, the lights
elvis a cité @OfirEhrlich
elvis @omarsar0 · 05/11 16:32
Agent swarms will certainly break the data infrastructure we have built for humans. Agents always need the latest records and the full history to complete tasks correctly. Take a look at what @Eon_io_ is building: a new data infrastructure to keep data fresh as it arrives in
elvis @omarsar0 · 05/11 16:17
Banger paper on RSI with foundation model systems. (bookmark it) Seriously, I highly recommend this paper if you want to start experimenting with RSI in your own company. Lots of great insights and tips. They propose Recursa, which raises DeepSeek-V4.1-Flash from 30.0% to
elvis a relayé @dair_ai
DAIR.AI @dair_ai · 05/11 15:10
Recommended benchmark from OpenAI on agents for everyday life tasks. They surveyed 3,837 US consumers to rank 148 life tasks by what people would pay an assistant to handle. Together, those tasks are worth over $1.7 trillion a year. LifeVal turns the most valuable of them into
L
Hugging Face a relayé @Llama_app
Llama @Llama_app · 05/11 14:57
Run decision models locally with a single click, get answers lightning-fast ⚡️ Llama now supports d1-Omni by @liquidai for multimodal queries, as well as d3-flash by vLLM, leading Jev Decision Index 🤝 Update Llama today 🦙
elvis @omarsar0 · 05/11 14:48
Impressive paper from CMU. Soon, we will have coding agents that can tune the entire stack, from prompts to token serving. This work experiments with coding agents to tune an LLM serving stack. They get Claude Sonnet 5.5 to build a serving system for Gemma E4B that delivers
D
Yann LeCun a relayé @daveeyyy_
Davey @daveeyyy_ · 05/11 08:48
Yann LeCun: The Al industry is stuck because every lab is chasing LLMs. Real Al agents will need to predict the consequences of their actions. "The Al industry is completely LLM-pilled, as they say." "Everybody is working on the same thing, they're all digging the same trench."
Greg Brockman a cité @_samirism
Greg Brockman @gdb · 05/11 07:28
dots is such a good product, and is getting better fast
M
Harrison Chase a relayé @onemikejohnson
Mike Johnson @onemikejohnson · 05/11 05:57
@hwchase17⁠ - we’re building the infrastructure for workflows where AI agents can’t afford to get things wrong, starting with hiring and compliance. We’re hiring to help build it. Thought it might resonate with some of the builders in your community. Would <3 a share if so.
Greg Brockman a cité @juberti
Greg Brockman @gdb · 05/11 05:52
GPT-Live for outbound telephony:
Greg Brockman a cité @OpenAI
Greg Brockman @gdb · 05/11 03:54
chatgpt images 2.6 is here!
G
Yann LeCun a relayé @che_shr_cat
Grigory Sapunov @che_shr_cat · 05/11 03:04
1/ A 109M text model trained with good views matched next-token prediction with 42x fewer samples on WIT-1024 (.109 at 21.6M samples vs .111 at 900M). Scaled up, it reached .184 mutual 10-NN with DINOv2, outscoring Qwen2.5-72B (.154) and Llama-3-70B (.151). 🧵
Yann LeCun a relayé @AndrewYNg
Andrew Ng @AndrewYNg · 05/11 02:52
Join me and 2000 other AI engineers and builders at AI Dev in New York City, Nov 30-Dec 1! A LinkedIn study found that over July-September, the job title AI Engineer has been one of the 10 fastest-growing occupations in seven major economies. AI engineering has become a core
Yann LeCun a relayé @SteveRattner
Steve Rattner @SteveRattner · 05/11 02:44
Democrats have won more votes than any party in midterm history: 66 million and counting.
Yann LeCun a relayé @paulg
Paul Graham @paulg · 05/11 02:44
Now we'll finally learn what's in the Epstein files. Mike Johnson is no longer in a position to shut down the House in order to prevent votes on measures to force their release. It's time. It's past time.
OpenAI Developers a relayé @gabrielchua
Gabriel Chua @gabrielchua · 05/11 01:45
The GPT Image 2.6 API gives you more control over image complexity and lets you create GIFs. So as fun short experiment - I asked Flare to create a GIF of a man doing a standing backflip. Compare low, medium and high complexity side by side. 👇 Detailed explanation below
Greg Brockman a cité @_catwu
Greg Brockman @gdb · 05/11 01:28
welcome @_catwu to the team!!
OpenAI Developers a relayé @higgsfield
Higgsfield AI 🧩 @higgsfield · 05/11 01:25
GPT Image 2.6 is LIVE on Higgsfield. Unlimited for 7 days for new users. > Turn any prompt or image into a GIF > Save references once and reuse them anytime > Refine with markup and export as SVG Available now on Higgsfield.
Nous Research a cité @TencentHunyuan
Nous Research @NousResearch · 05/11 01:21
Hy4, the new model from @TencentHunyuan, is 80% off in Nous Portal for one week. It adds image input and computer use, and Tencent trained it for game development in Unity, Unreal and Godot and for editing Word, Excel and PowerPoint files. Try it today at
elvis a cité @RicursiveAI
elvis @omarsar0 · 04/11 23:18
Always a great day when we get another strong open weight model.
A
OpenAI Developers a relayé @ajambrosino
Andrew Ambrosino @ajambrosino · 04/11 22:37
we're giving you some RAM back. local threads could grow memory usage as they became long, even without starting other processes. @btraut declared war on the problem. enjoy in equally important news, @matteing made his handwriting into a font, which i use below
Nous Research a cité @NousResearch
Nous Research @NousResearch · 04/11 22:33
The Hermes Agent Plugin Hackathon starts now! It runs for 7 days, with $20,000 in cash prizes and more. Plugins are the most powerful way to extend Hermes Agent, and we want this hackathon to show what they can do. A plugin can change anything from the model Hermes runs on,
Harrison Chase a relayé @sydneyrunkle
Sydney Runkle @sydneyrunkle · 04/11 21:26
this piece is worth the read, but quite dense. here's my (relatively) brief breakdown: super agents are defined by 4 characteristics: - continuous: one agent across threads and access points - productive: responds quickly and completes work efficiently - proactive: responds to
S
Stability AI a relayé @stableaudio
Stable Audio @stableaudio · 04/11 20:58
You finally get the perfect snare. Ask for more punch, and it’s gone. Sculptro by Javor Vatchkov lets producers keep shaping a sample with Stable Audio 3.0, changing only what isn’t working. 3rd place at the @MusicHackspace ADE Lab Hackathon 🥉
Andrew Ng @AndrewYNg · 04/11 20:56
Join me and 2000 other AI engineers and builders at AI Dev in New York City, Nov 30-Dec 1! A LinkedIn study found that over July-September, the job title AI Engineer has been one of the 10 fastest-growing occupations in seven major economies. AI engineering has become a core
Harrison Chase a relayé @sydneyrunkle
Sydney Runkle @sydneyrunkle · 04/11 20:15
full transparency, things that went wrong during the filming of this demo... - almost booked a reservation that required a HUGE downpayment / cancellation fee bc my card was saved on resy, so the agent could just book+pay - hit rate limits w/ resy bc i simulated log in so many
A
elvis a relayé @anvisha
Anvisha @anvisha · 04/11 20:10
We just made a 5x cheaper Higgsfield. Voyager gives you AI video models like Seedance 2.5 at half Higgsfield’s price. More importantly, it cuts wasted generations. Users get final outputs in 2–3x fewer API calls. Half the price per generation. A fraction of the generations.
elvis a cité @Davidstout
elvis @omarsar0 · 04/11 19:42
Another great open-source model release. TwIL-8B shows strong results on coding, instruction-following, and math benchmarks. I'm bullish on small models we can run locally and on our own hardware. As we unlock efficient intelligence and capabilities, smaller models will play
elvis a cité @pranavctl
elvis @omarsar0 · 04/11 19:28
Impressive work from @pranavctl on cloud coding agents. It's the fastest cloud coding agent: 19.4s from prompt to a running stack, and the benchmark is public. It runs Claude Code, Codex, and OpenCode on isolated cloud machines you watch and steer live, on the Claude or
Harrison Chase a relayé @caspar_br
Caspar @caspar_br · 04/11 19:17
A year ago I wrote that LangGraph isn't going anywhere. Today it's the runtime under everything at LangChain, including MDA @JoshARosen wrote a good piece on why deterministic layers will become standard agent infrastructure. We made the same bet early:
Harrison Chase a relayé @sydneyrunkle
Sydney Runkle @sydneyrunkle · 04/11 19:17
if you give your agent a browser, it can do work on the web, even on sites w/ no agent native APIs! in this demo my agent books dinner reservations for me. i only had to log into the reservation site once; it saved my login and acts on my behalf for every new session!
Greg Brockman a cité @pierrecomputer
Greg Brockman @gdb · 04/11 19:05
welcome @pierrecomputer to openai! pierre’s open-source developer tools already power parts of codex today, and excited to build together the future of collaboratively building software with ai.
elvis a cité @axiommathai
elvis @omarsar0 · 04/11 18:22
Huge release for AI and math! Many people assume AI-generated code at scale means low quality. Axiom's new Stacks Project formalization in Lean says otherwise. The Stacks Project is a massive, deeply interconnected foundation for modern algebraic geometry. They turned it
Google DeepMind a relayé @Google
Google @Google · 04/11 18:12
Voice Remixing for Gemini 3.8 Flash TTS is now live in @GoogleAIStudio and the Gemini API. Start with a preset from our library of 2,000+ voices or a custom-designed voice, and use prompts to fine-tune timbre, pitch, pace, and accent, like "add a subtle Southern U.S. accent" or
Yann LeCun a relayé @ylecun
Yann LeCun @ylecun · 04/11 18:02
@Arjunjain Plot twist: I was Vapnik's Department Head between 1996 and 2002, and made sure he had the space, resources, and collaborators to do his best work. To people who think I'm dogmatic: I famously disagreed with Vapnik about the best way forward in ML, but did what I could to
Yann LeCun a relayé @SteveRattner
Steve Rattner @SteveRattner · 04/11 17:52
Elon Musk dropped nearly half a billion dollars on the midterms. 98% of the money he spent on specific candidates went to losers.
Yann LeCun a relayé @antirez
antirez @antirez · 04/11 17:51
Remember one thing when you read JEPA vs LLMs in the context of the platonic representation hypothesis. Hidden state of an LLM does not represent the full representation of the LLM, it carries only what is needed for the next token. It's not an encoder.
elvis a cité @yuchenping
elvis @omarsar0 · 04/11 17:48
Huge market about to unlock here! Atmee v2 is a real-time avatar model for live video agents. It creates any character from one image and 10 seconds of audio. They report that it runs up to 50x faster than video models like Seedance 2.5 and Wan 3.0. Text and voice work well
R
OpenAI Developers a relayé @RaemondBW
Raemond @RaemondBW · 04/11 16:50
Two of my Astra projects (Astra panelized to save me a bit of money). 1. USB power analyzer (rp2040, ina226, usb2514b, etc.) 2. Replacement pcb for an IKEA speaker that gives it airplay support. (Esp32s3) Astra helped me quote and pick parts in stock at @JLCPCB
OpenAI Developers a relayé @natalieyeo
Natalie @natalieyeo · 04/11 16:40
You can make a physical Codex pet with Astra too!
Hugging Face a cité @RoboColosseum
Hugging Face @huggingface · 04/11 16:38
Evaluating robot models in the real world is hard, but solvable. That's why we're a founding partner of RoboColosseum 🤗 100+ real robots, 140+ in-the-wild scenes, and every rollout, failures included, open on the Hub.
M
OpenAI Developers a relayé @zavrenn
Mohamed @zavrenn · 04/11 16:33
I kept pushing GPT-6 Astra in Codex, and this little city now has its own airfield. An airplane, more detailed buildings, and better water. Still all @threejs, with no imported 3D models.
elvis a cité @ladyxtel
elvis @omarsar0 · 04/11 15:15
An effective memory system lets a harness leverage agents' persistent capabilities. So decoupling memory from the model is smart. Living Brain gives ChatGPT, Claude, Cursor, Codex, and your own agents one shared persistent memory. They report 92.2% on LoCoMo, with a 464ms
elvis a relayé @dair_ai
DAIR.AI @dair_ai · 04/11 14:18
Great paper from UC San Diego on why long-context retrieval fails. (bookmark it) It also offers an interesting discussion of why context rot happens in LLMs. When a model fails on a long input, people usually blame the length. This paper finds that length alone does very
L
Hugging Face a relayé @Llama_app
Llama @Llama_app · 04/11 14:17
Llama ❤️ @huggingface Hub 🤗 Get started with thousands of LLMs with a few clicks, running lightning-fast on your computer 🦙 Private, local, open-source 🙌🏻
elvis @omarsar0 · 04/11 13:25
Highly recommended paper from NVIDIA. (bookmark it) There is a lot of debate about whether to use context compaction aggressively. They show that with frequent compaction, frontier language models lose little to no accuracy on a broad range of long-horizon and long-input tasks
N
hardmaru a relayé @nhk_news
NHKニュース @nhk_news · 04/11 10:45
インフラや防衛分野に向け国産AIの開発進む #nhk_news
NHKニュース インフラや防衛分野に向け国産AIの開発進む | NHKニュース 【NHK】AIの活用がさまざまなインフラや防衛分野にも広がることが見込まれる中、国内の企業が基盤となるモデルから手がける「国産AI」の開発が進んでいます。 このうち、スタートアップ企業の「Sakana AI
Hugging Face a relayé @NielsRogge
Niels Rogge @NielsRogge · 04/11 09:21
We @huggingface reproduced this paper!! It's built with ML Intern, an open-source harness to do AI research, train models, and more Model: Demo:
huggingface.co ML-Intern-lab/Qwen3.5-0.8B-weight-space-programming · Hugging Face Image-Text-to-Text model with 0.9B parameters. Download, explore files and run it on Hugging Face.
huggingface.co Weight-Space-Programmed Qwen3.5-0.8B - a Hugging Face Space by ML-Intern-lab This app lets you type a math problem and (optionally) set a token limit. It then runs both the original Qwen3.5‑0.8B model and its weight‑edited version, showing the two generated answers side‑by‑...
A
Yann LeCun a relayé @AlexiGlad
Alexi Gladstone @AlexiGlad · 04/11 07:02
looped reasoning is completely overrated Energy-Based Reasoning (EBR) runs circles around looped reasoning paper dropping this week!
S
Yann LeCun a relayé @shaunbjohnson
shaunbjohnson @shaunbjohnson · 04/11 06:52
At @224ventureshq, one thing we look for when we speak to founders is speed of execution. Execution is critical because of the rapidly changing market we are in. When you're starting out, it is often the case that you do not have a moat. You need customer traction first, to
Nous Research @NousResearch · 04/11 06:16
Would you rock this
Yann LeCun a relayé @KenRoth
Kenneth Roth @KenRoth · 04/11 04:55
This is the current NY Times forecast. Incredible!
Yann LeCun a relayé @randall_balestr
Randall Balestriero @randall_balestr · 04/11 04:54
If you are at INFORMS in San Francisco, come by room 211 where I will be talking about our latest JEPA research in 10min (8:20am)! Have some time to chat afterwards as well!
Yann LeCun a relayé @ylecun
Yann LeCun @ylecun · 04/11 04:35
@heyshrutimishra LLMs accumulate a lot of declarative knowledge. That's how they can answer a large number of questions. That's also why they require so many parameters and hence so much memory. But one can replace the accumulation of knowledge by *understanding* That what JEPA world models
B
Yann LeCun a relayé @Brian_Bo_Li
Brian Li @Brian_Bo_Li · 04/11 04:28
@ylecun is certainly one of the few true thinkers and innovators in today’s overhyped AI landscape. He is brave to say something unpopular, but at the same time this is not without reason; all his thoughts come from math and physics, and he is one of the few researchers I have
A
Yann LeCun a relayé @Arjunjain
Arjun Jain | Fast Code AI @Arjunjain · 04/11 04:15
I was trained to think Yann LeCun was wrong. My PhD was in Vapnik’s school. Kernels. Bounds. Neural nets were alchemy. Then I went to NYU, and he changed my mind. Recently, he did the one thing my old school always asked for. He brought a proof.
elvis a cité @jacsebl
elvis @omarsar0 · 04/11 03:40
🤯Another mind-blowing example (this time from the CTO of Legora) on Fable hand-writing weights of a model. This weight-space programming stuff is seriously some of the most exciting research I have seen in a long time.
hardmaru a relayé @SakanaAILabs
Sakana AI @SakanaAILabs · 04/11 03:30
本日よりGartner IT Symposium/Xpo™ 2026が開幕しました🐡 Sakana AIはブースにて、Sakana Marlin、Fugu、Namazuのデモ展示中。 18:30からはIT Xpo内でネットワーキング・レセプションも開催されます。ぜひブースにお立ち寄りください🚀 #GartnerSYM
LlamaIndex 🦙 a relayé @jerryjliu0
Jerry Liu @jerryjliu0 · 04/11 03:27
OCRia-1 is one of the fastest models ever for doc parsing. We sped up OCRia-1 by 10x on a single B200 (285 tok/s --> 2800 tok/s). there are a ton of useful learnings in this extremely detailed blog post for anyone looking to optimize models! * we made use of established
Sam Altman a relayé @thsottiaux
Tibo @thsottiaux · 04/11 02:55
Day 25/ We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today. We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering
François Chollet a cité @sama
François Chollet @fchollet · 04/11 02:46
Co-sign. The only future that is worth building is one in which AI remains a tool in human hands, used to elevate the human condition.
Sam Altman @sama · 04/11 02:40
People should always be in control of the future. OpenAI is against the notion that it would be acceptable for people to be relegated to a role of doing things that “AI systems think makes sense to reward in humans”. People should not cede control of the future to
M
OpenAI Developers a relayé @kanwisher
Matthew Campbell @kanwisher · 04/11 02:30
I’ve been saying it for months . Vibe code hardware
Greg Brockman a cité @OpenAI
Greg Brockman @gdb · 04/11 01:51
ChatGPT for physical engineering, including enabling amateurs to design prototype devices:
D
OpenAI Developers a relayé @diegocabezas01
Diego | AI 🚀 - e/acc @diegocabezas01 · 04/11 00:43
GPT-6.1 Sol Ultrafast via API can make 3d in realtime I asked how an airplane works, built a live 3D model over my desktop. • Decisions API (like JEV API) picks text, 3D or HTML. • Responses API / GPT-6.1 Sol · Low · Ultrafast makes the visual. • Native overlay + Three.js:
Greg Brockman a cité @mikeallen
Greg Brockman @gdb · 04/11 00:05
welcome @jillhazelbaker to the team, extremely excited to be working together!
elvis a cité @dair_ai
elvis @omarsar0 · 03/11 23:24
Good research on maintaining and scaling agent skills.
elvis @omarsar0 · 03/11 22:58
Super interesting finding on optimizing agent harnesses. When you optimize an agent harness for both quality and cost, apply your cost preference only at the end of the search, since applying it at every step throws away harnesses that lead to better ones. That is the key
U
OpenAI Developers a relayé @unitygames
Unity for Games @unitygames · 03/11 22:39
ICYMI: We launched the official Unity plugin for @OpenAI’s Codex, available directly in OpenAI’s universal plugin directory. Once installed, Codex can choose any skill in the plugin to help ensure the task is completed correctly. Check out our walkthrough of four different
Harrison Chase a relayé @sydneyrunkle
Sydney Runkle @sydneyrunkle · 03/11 22:04
we're hosting an event on thursday at the @LangChain office with @allietheicon from @typesafeai to chat about all things jev! come learn about - where jev fits in the agent loop - why jev is a "system one" model - jev as a judge for evals
luma.com LangChain SF Meetup: Judgment Calls · Luma Join us in San Francisco (SoMa) for a night of talks on building with decision models and using them to evaluate agents. We will be joined by Sydney Runkle, a…
Hugging Face a relayé @vanstriendaniel
Daniel van Strien @vanstriendaniel · 03/11 21:38
OCR models often get stuck in loops, so I asked Claude Opus if it could stop one from looping by editing its weights. It changed 0.0085% of them and with no training, it now reads this 1881 page instead of writing 16,399 characters of dots! Prompt was basically: "read this,
Anthropic @AnthropicAI · 03/11 21:10
New Anthropic research: progress on provable inference for large language models. Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more
elvis a cité @guilleflorvs
elvis @omarsar0 · 03/11 21:02
Own your intelligence stack, folks! Revolut joins the ever-growing list of companies that see owning the intelligence stack as paramount to staying competitive. That's the role open-source AI plays today. If you think of companies as mini-AGIs, you quickly realize the
Nous Research @NousResearch · 03/11 20:12
Thank you to everyone who traveled from across the world to NousCon 2026. We loved spending time with the people who make the Hermes community great. As you all know, we released the Hermes Mobile beta at the event, and as anticipated it filled up quickly. We are working
elvis a cité @svonava
elvis @omarsar0 · 03/11 19:29
Stop overpaying for agents, folks! A single agent request can trigger many model calls. Your agent might classify a ticket, rerank results, extract a date and run a guardrail before it ever answers. Sending all of those calls to a frontier model gets expensive fast. And it’s
Harrison Chase a cité @sydneyrunkle
Harrison Chase @hwchase17 · 03/11 19:13
earlier this ~spring, models got good enough to drive browsers in some ways, browsers are the new terminal! a general purpose tool one difference is there is a few different ways to design tools to drive the browser. @sydneyrunkle deep dives into how we did this for MDA
Harrison Chase a relayé @sydneyrunkle
Sydney Runkle @sydneyrunkle · 03/11 19:12
browser use is an important primitive for agents, but it's hard on the agent's context window: 1. tasks are long running, w/ tons of sequential tool calls 2. web pages carry a lot of content that's irrelevant to the task so tool design really matters. we chose @Stagehanddev
A
Harrison Chase a relayé @_anish_agarwal
Anish Agarwal @_anish_agarwal · 03/11 18:48
We just raised $120M, with a $1B valuation, to build software that keeps itself running. AI is writing more code than ever. It’s also making production less reliable: more code, shipped faster, understood by few. @traversal_ai understands how production systems fit together,
elvis a cité @_anish_agarwal
elvis @omarsar0 · 03/11 18:31
Software that keeps itself running is where agents unlock the most value today. I build a lot with coding agents. Shipping code isn't my bottleneck anymore. Now I spend more time figuring out what broke in production and why. @traversal_ai solves this by mapping how your
elvis a cité @intology
elvis @omarsar0 · 03/11 18:12
Highly recommended read. Very few companies I know have achieved this leap in scaling autonomous AI R&D in science. Intology claims their Locus system sets the frontier for automating molecular design. That covers molecular property prediction, potency prediction with
LlamaIndex 🦙 @llama_index · 03/11 17:49
We got OCRia-1 generating tokens nearly 10× faster on a single B200. Getting there meant making much better use of the same GPU. We tuned how OCRia-1 runs for the demands of document parsing, testing every change for both speed and accuracy. Some improvements were only 2–5%.
G
elvis a relayé @GreylockVC
Greylock Partners @GreylockVC · 03/11 17:48
One of the biggest bets startup employees make is on the founder. That trust may not be tested until years later, when the company exits. A startup can sell for billions, and the people who helped build it can walk away with vastly different outcomes. During M&A negotiations,
elvis @omarsar0 · 03/11 16:03
Great paper from Alibaba on self-evolving agents, with a live A/B test across 200,000+ merchants. (bookmark it) This is another brilliant example of leveraging episodic memory with agents and the importance of attribution. SPAR runs B2B sales outreach on Alibaba's 1688
elvis @omarsar0 · 03/11 15:30
I tested LongCat-2.5 and Qwen3.8-Max on the same browser game, and LongCat got it done for $0.28 versus $0.82. That's nearly 3x cheaper for the same task. I regularly test cheaper models on full app builds, and LongCat-2.5 from @Meituan_LongCat looks like a strong option. With
L
Hugging Face a relayé @Llama_app
Llama @Llama_app · 03/11 13:52
Our picks from last week's model drops 🦙 → LFM3 by Liquid AI: small models for text, image and speech, with d1 decision models → Mistral Large 4: large, 1M context frontier → Flint by Recursive: mid-sized, built for coding agents Get them at
llama.app Llama · Your AI, on your computer Run the latest open models on your computer. Chat with them, or connect them to your coding agents, editors, and apps. Free, private, and nothing to configure.
Nous Research a relayé @Teknium
Teknium 🪽 @Teknium · 03/11 13:38
Update your priors: Every nous portal sub in Hermes comes with a cloud agent and it's own computer now, high usage limit models, and all of your tools like search, scrape, browser use, image gen and more
elvis a relayé @_arohan_
rohan anil @_arohan_ · 03/11 13:37
Hi! this is Rohan Anil, I develop training algorithms among other things - my work trains various frontier models. At CoreAuto, I was fascinated by the problem of training on compute islands, eventually in space training. I hope you all enjoy this write up!🪐🛰️
田
hardmaru a relayé @ChiekoTazuke
田附千絵子 たづけちえこ @ChiekoTazuke · 03/11 06:39
Sakana AIを長くサポートいただいていたBearさん、ついに正社員に、、、感無量です🐻🐟
Harrison Chase @hwchase17 · 03/11 04:55
is org/agent level memory just "skills"? assuming there is process for updating skills from agent traces/history why or why not?
M
hardmaru a relayé @MLBear2
ML_Bear @MLBear2 · 03/11 04:24
本日、Sakana AI に入社しました！ 昨年3月から週3勤務の業務委託としてお世話になってましたが、今月からは正社員としてフルコミットさせて頂くことになりました。 来週の関西Kagglerをはじめ、Sakana AI社員としてイベントに参加することも増えると思います。よろしくお願いします😇
Yann LeCun a relayé @randall_balestr
Randall Balestriero @randall_balestr · 03/11 03:24
Another example that we should spend more time thinking about improving pretraining losses rather than just searching for more data/gpus! SIGReg is not the final answer, but certainly a good first step in that direction! Try it out, checkpoint is available!
L
Yann LeCun a relayé @lukaskuhn77
Lukas Kuhn @lukaskuhn77 · 03/11 03:16
Introducing LeVJEPA 1.5 Large, the first step in scaling our method. It closes the gap to V-JEPA on both image and video benchmarks, trained with 5.6x less compute and over an order of magnitude less data.
Greg Brockman a cité @OpenAIDevs
Greg Brockman @gdb · 03/11 02:54
Pando is a Git backend which can scale to very large monorepos. We took the hard road of continuing to use Git, which means we could enable Pando with no change to devex. Considering providing this to others, let me know (gdb@openai.com) if you'd be interested.
OpenAI Developers a relayé @derrickcchoi
Derrick Choi @derrickcchoi · 03/11 01:26
Learn about Pando, our new backend for our OpenAI monorepo
T
OpenAI Developers a relayé @tnm
Ted Nyman @tnm · 03/11 01:25
We've scaled source control at @OpenAI to keep up with merges, commits, & repo growth. Here's our technical blog post about how we do it. & one more thing: we've kept git compatibility, and we continue to contribute generally useful changes upstream. This is just the start.
P
Pete ✈️ CoRL a cité @xdofai
Pete ✈️ CoRL @nonmayorpete · 03/11 00:19
Free breakfast for CoRL attendees next week!
J
Yann LeCun a relayé @JeffDean
Jeff Dean @JeffDean · 02/11 23:46
Election Tuesday is just two days away. For those eligible to vote in the U.S., please make time to go to the polls to vote (or mail in your ballot if you haven't already). Voting is a great privilege of living in a democracy and how we all make our collective voices heard. 🗳️
G
Harrison Chase a relayé @GitMaxd
Git Maxd @GitMaxd · 02/11 23:30
If you get to go to one conference in 2027 about Applied AI - this is the one You’ll learn from and get to talk to @LangChain Engineers and Ambassadors who provide direct, actionable answers, not a “let’s connect”! Join any of the included cutting edge, hands-on interactive
P
Pete ✈️ CoRL a cité @theSamPadilla
Pete ✈️ CoRL @nonmayorpete · 02/11 23:09
Notably, if a startup is saying "we do tactile", check their understanding of the inherent engineering tradeoffs involved with tactile. Basically nobody knows what they're talking about.
elvis a cité @rikiyatakehi
elvis @omarsar0 · 02/11 21:31
Must-read report! They find that LLMs are surprisingly good at reading and writing neural network weights. Huge for RSI when you think about it! Current RSI pipelines focus on writing code to launch training jobs. This capability could offer a more native path to recursively
C
Harrison Chase a relayé @carodivittorio
Caroline di Vittorio @carodivittorio · 02/11 21:17
This is a huge reason I believe in cloud agents over local agents. A local agent only runs when you start and watch it, but what will make agents more useful next is proactive, capable work that happens in the background.
Harrison Chase @hwchase17 · 02/11 20:59
We're bringing back Interrupt for the THIRD year - This will be the biggest one yet (new venue that can hold ~2k people!) - Most talks & workshops we've ever had - New Executive Track Register today for the early bird special:
interrupt.langchain.com Interrupt 27 | The Agent Conference by LangChain | San Francisco, CA Join agent engineers and AI leaders for technical talks, hands-on workshops, and networking, and to explore the latest innovations from LangChain. March 24-25 in San Francisco, CA.
Anthropic @AnthropicAI · 02/11 20:51
Applications are now open for the Anthropic Residency, a paid, fixed-term research program for academics, technical founders, and staff-level engineers who want to make AI safe and beneficial for everyone. No ML background is required. Apply by November 22:
S
Stability AI a relayé @stableaudio
Stable Audio @stableaudio · 02/11 20:37
48 hours, one idea, one win. Meet Team Generative Gallery, winners of the ADE Lab Hackathon Stable Audio challenge. 🏆
T
Jim Fan a relayé @tonyzzhao
Tony Zhao @tonyzzhao · 02/11 20:02
This is a great model. Congratulations @DrJimFan and @yunzeman for the awesome work!
Harrison Chase a relayé @LangChain
LangChain @LangChain · 02/11 19:48
🌉 Interrupt, The Agent Conference by LangChain, returns to San Francisco March 24 + 25 🎟️ Early Bird tickets available until Dec 7th Info + RSVP:
interrupt.langchain.com Interrupt 27 | The Agent Conference by LangChain | San Francisco, CA Join agent engineers and AI leaders for technical talks, hands-on workshops, and networking, and to explore the latest innovations from LangChain. March 24-25 in San Francisco, CA.
Harrison Chase a relayé @sydneyrunkle
Sydney Runkle @sydneyrunkle · 02/11 19:26
we think a lot about context engineering at langchain there's a lot of of old but good primitives here: skills, tool search, summarization to name a few this short blog covers new challenges in this space: browser use, long term goal tracking, async work, and more!
François Chollet @fchollet · 02/11 18:59
The most grating aspect of LLM writing is that it tries to turn every single sentence into a punchline, a witty slogan, with no regard for which sentences need emphasis and which ones don't. "A tiring formula, overused"
elvis a cité @levantolabs
elvis @omarsar0 · 02/11 18:37
As a dev, I am bullish on this new type of system-one extraction model. It's great for evidence-based extraction and perfectly compliment decision models like Jev. With Sage 2, you describe the fields you need, and it returns structured values with the exact source text and a
OpenAI @OpenAI · 02/11 18:27
For AGI to benefit all of humanity, it should help people with the everyday tasks that matter most to them. So we built LifeVal, a new benchmark shaped by feedback from thousands of ChatGPT users on where they most want help from a personal assistant—from finding childcare to
elvis a cité @hume_ai
elvis @omarsar0 · 02/11 17:33
One of the most insane applications of decision models I've seen so far. Unlike other decision models, this one is audio-native. This unlocks new ways for how we interact with agents. The Hume Audio Decision API from @hume_ai listens to the audio itself. An agent can ask
E
OpenAI Developers a relayé @elise_ai
Elise Labs @elise_ai · 02/11 17:05
We partnered with @OpenAI as early testers of GPT-Live-1, exploring how it can improve the millions of conversations EliseAI handles each month across housing and healthcare. On Nov. 17 at @a16z's SF office, both teams are coming together to share how we integrated GPT-Live-1
elvis a cité @SemiAnalysis_
elvis @omarsar0 · 02/11 16:48
Recommended read. Always great to have new ways to measure the impact of frontier intelligence. What strikes me the most is how much better Gemini (now with Carbon) has gotten. Don't count out Google just yet. As we move away from standard benchmarks, you will start to see
Hugging Face a relayé @vanstriendaniel
Daniel van Strien @vanstriendaniel · 02/11 16:48
I re-OCR'd 14,483 pages of fossil fuel industry documents using Saturate. One GPU + $1.47. 8.9 pages/s once running via @huggingface Jobs. Saturate is a library I've been working on for running models over a dataset or bucket against any OpenAI-compatible endpoint/inference
elvis a relayé @dair_ai
DAIR.AI @dair_ai · 02/11 16:31
// Context as Infrastructure for Agents // The RAGFlow team at InfiniFlow ran a harness with only BM25 and grep on MiniMax M3. It answered 85% of BrowseComp-Plus and 91% of FRAMES with no vector or compiled index. They then tested summary trees, PageIndex-style outlines, wikis,
elvis @omarsar0 · 02/11 16:27
Nice paper on CoT controllability. They find that chain-of-thought monitoring depends partly on models being unable to control what they write in their reasoning. They traced much of that limit to the first few tokens after the opening thinking tag. Entropy collapses there, and
Nous Research a relayé @Teknium
Teknium 🪽 @Teknium · 02/11 15:48
Been focusing on the Desktop app this last week or two. This PR should be welcome news to a lot of you power users. On next full release later this week I think you'll start noticing the difference in the desktop app getting noticeably nicer. Still a long way to go, but
DeepSeek a relayé @DeepSeekHarness
DeepSeek Harness @DeepSeekHarness · 02/11 15:23
Your Harness. You Define It. Meet Creator Mode in DeepSeek Harness. Turn ideas into plugins that you can use, from a design workspace to a sandbox game or a 3D LLM visualizer. What will you create?
V
OpenAI Developers a relayé @velrino
Velrino @velrino · 02/11 14:36
I love exploring the universe. With @OpenAIDevs Codex GPT-6.1 Sol, I built our Universe: a 3D @threejs explorer with 20 real galaxies and 767 catalogedplanets The app and its sci-fi design system are open source now Links in the replies
Z
OpenAI Developers a relayé @zoe_newera
Zoe Zou @zoe_newera · 02/11 14:28
I used Codex to build a body from coordinates and particles, entirely in code. Its chest pulses to “Gravity Removed,” a track I made with Suno. A digital heartbeat reaching into our world. #CreativeCoding #AIArt
K
OpenAI Developers a relayé @kevinwhinnery
Kevin Whinnery @kevinwhinnery · 02/11 14:23
You: Me: okay fine I'll build an agent with the @OpenAIDevs Agents API to help me sort out this 1995 Miata so I can drive it to California and track it at Sonoma and Laguna Seca, geez, lay off
elvis a cité @rauchg
elvis @omarsar0 · 02/11 14:21
Own your intelligence stack, folks! I will never get tired of saying this. Your business's very existence is at play here. But you don't have to do this all at once if you can't afford it. I mean, open models are still very expensive to run and tune. If you want to start
S
OpenAI Developers a relayé @bargues_sofia
Sofia Bargues @bargues_sofia · 02/11 14:18
What an amazing day at the DevDay Community Hackathon in London! @OpenAIDevs Huge thanks for making it such a special experience! 💜 @Andy_AJT @oscarfalll @paw_lean @franmoretti_
R
hardmaru a relayé @rmaruy
R. Maruyama @rmaruy · 02/11 12:17
企画で関わらせていただいたイベントレポートが公開されました。Jürgen Schmidhuber教授の基調講演およびSakana AIのLLM、AI Scientist、Physical AI研究を展望するトークはどれも素晴らしいものでした。一企業を超え、AIの新時代が始まるイメージを共有する機会となればと思います。
Hugging Face a relayé @NielsRogge
Niels Rogge @NielsRogge · 02/11 10:10
It's raining OCR models again! This time, @OpenBMB released MiniCPM-OCR, a 1B-parameter model for document parsing I made a @huggingface Space so you can compare its outputs against Baidu's new Unlimited OCR v2 Find the link below
R
hardmaru a relayé @RobertTLange
Robert Lange @RobertTLange · 02/11 07:21
Had tons of fun last week at the @SakanaAILabs symposium 🗼 Blog: I first met Jürgen in Vienna at ICML 2024. We had just finished working on DiscoPOP and were pushing hard towards the AI Scientist v1. Fast forward to October 2026, and we are moving
sakana.ai Sakana AI 【開催レポート】Sakana AIシンポジウム：Jürgen Schmidhuberが語るAIのこれまでとこれから、日本から挑む言語モデル・世界モデル・Physical AI
hardmaru a cité @SakanaAILabs
hardmaru @hardmaru · 02/11 02:57
Last week in Tokyo, we hosted the Sakana AI Symposium with our Chief Scientific Advisor, Jürgen Schmidhuber. Eight years ago, @SchmidhuberAI and I wrote the "World Models" paper together. Today, we’re building on those foundations at our RSI Lab to train an agent-native LLM from
L
Harrison Chase a relayé @_loganfairchild
Logan @_loganfairchild · 02/11 02:35
I love that agent engineering keeps getting new weird problems as these things evolve
Yann LeCun a relayé @haider1
Haider. @haider1 · 02/11 01:16
Yann LeCun says: People in AI have made the same mistake for 70 years: if a machine can do a task humans see as intellectually complex, it must be smarter than humans. This is completely false We have computers that are superhuman at many tasks but we're still far from machines
François Chollet @fchollet · 02/11 01:01
Do I know anyone who knows anything about cyanotype printmaking?
François Chollet @fchollet · 02/11 00:43
Evolution doesn't benchmaxx. It optimizes for generality and robustness, because overspecialization makes a species fragile when the environment inevitably shifts.
OpenAI Developers a relayé @OpenAIDevs
OpenAI Developers @OpenAIDevs · 02/11 00:29
Meet @Scav, who brings builders together in a shared robotics community where they push the edges on what's possible with AI in the physical world. In one experiment, GPT-6 Astra remotely controlled a robot arm:
LlamaIndex 🦙 a relayé @jerryjliu0
Jerry Liu @jerryjliu0 · 01/11 22:25
Today we're introducing Extract Max Effort - the best frontier agent in the world for document extraction 💫. 🚀 It is an absolute bazooka. Use this when you absolutely need perfect 100% extraction accuracy over the most complex documents and schemas. This includes reasoning
LlamaIndex 🦙 a relayé @jerryjliu0
Jerry Liu @jerryjliu0 · 01/11 22:25
We've built the most accurate document extraction system on the planet. Extract Max Effort will spend dynamic tokens reasoning about every visual detail across hundreds of pages to ensure that you get the maximum accuracy possible. Check out this massive well log report below.
elvis a cité @patrickc
elvis @omarsar0 · 01/11 21:00
Highly recommended read. I've been bullish on the idea of personal malleable software. Patrick calls this pattern “liquid software”. This is a new way to build software, especially personalized software. I have been building my own harness this way, but this also has
S
Yann LeCun a relayé @heyshrutimishra
Shruti @heyshrutimishra · 01/11 20:37
Yann LeCun was right & nobody in this industry wants to say it out loud. 109 million parameters beat 72 billion. He has spent years saying scale is not the thing that makes a model understand the world. Meta bet the other way. He left in November and started his own lab in
François Chollet @fchollet · 01/11 19:37
I got a fun birthday present from the team at Ndea ❤️
Yann LeCun a relayé @ylecun
Yann LeCun @ylecun · 01/11 19:17
@iamsupersocks Tout cela est correct. Merci. Sauf une chose : j'ai bien créé FAIR en 2013, y compris la branche parisienne en 2015, et montréalaise en 2017. Mais j'en ai quitté la direction en 2018 pour devenir Chief AI Scientist: je me suis refocalisé sur la recherche (Self-Supervised
elvis a relayé @dair_ai
DAIR.AI @dair_ai · 01/11 19:00
The Top AI Papers of the Week (Oct 26 - Nov 1): - HarnessMM - DecisionRAG - Unified Training - Agentic Backdoors - EmbeddingGemma 2 - System One Rerankers - Verification in Agent Harnesses Read on for more:
elvis a cité @DeepSeekHarness
elvis @omarsar0 · 01/11 18:17
Highly recommended read. It's an absolute masterclass in building a modern harness and enabling more autonomous agent systems. This is the best write-up I have read on how I have been working with my agent teams powered by my custom harness. In simple terms, you want a
Yann LeCun a relayé @askalphaxiv
alphaXiv @askalphaxiv · 01/11 13:31
Yann LeCun and Randall Balestriero just Proved the Platonic Representation Hypothesis, and found a better way to achieve it. The Platonic Representation Hypothesis suggests that AI models trained on different modalities, like images and text, eventually learn similar
J
Emad a relayé @SuJinyan6
Jinyan Su @SuJinyan6 · 01/11 08:08
Poisoning just 0.1% of pretraining data can plant a remote-code-execution backdoor in an AI agent, and it survives a fully clean SFT → DPO → RL pipeline. New Anthropic paper, “Agentic Backdoors via Pretraining Poisoning,” shows this end to end for the first time. The attacker
M
OpenAI Developers a relayé @zavrenn
Mohamed @zavrenn · 01/11 00:29
After Three.js r187 came out, I wanted to see how far I could push it with Codex -> GPT-6 Astra. Tbh, I'm amazed by the result. This whole little city is built in @threejs, with no imported 3D models. And more to come.
V
Harrison Chase a relayé @Vtrivedy10
Viv @Vtrivedy10 · 31/10 22:08
personally have the exact opposite view here! building a great harness is a deeply core part of building a great product countless examples you can ask an agent to explain from Grok Bot YouTube video, Factory blogs, our blogs at LangChain, OpenAI’s Codex blogs, etc a great
elvis @omarsar0 · 31/10 21:33
Highly recommended paper from CMU and others on how foundation models get trained. It fundamentally asks whether pretraining, midtraining, SFT, and RL really need to be separate stages. They show that this separation isn't fundamental. All of these stages can run under a
François Chollet @fchollet · 31/10 21:25
Dopamine does not spike highest for guaranteed rewards. It spikes highest at reward prediction error. If you want to have fun in life, you should stop avoiding risk and uncertainty.
elvis a relayé @dair_ai
DAIR.AI @dair_ai · 31/10 19:00
Brilliant paper on agent memory from FAIR at Meta. (bookmark it) A single global history is the usual way to let agents learn from past runs. When tasks are interleaved, it gains almost nothing over no memory at all, an average of +0.015 in normalized reward. This work
M
Harrison Chase a relayé @mgreau
Maxime Gréau @mgreau · 31/10 18:02
💯 A major shift toward proactive, always-on, goal-oriented agents is starting now and will gain momentum over the next few months
A
Google DeepMind a relayé @a_ron_saunders
Aaron Saunders @a_ron_saunders · 31/10 17:27
While most jack-o-lanterns just sit on porches, the @GoogleDeepMind team decided to let AI agents design, simulate, and bring one to life. 🎃🤖 We built a physical, dancing 12 degrees of freedom robot directly from a single plain-English prompt, with no manual CAD required!
Harrison Chase @hwchase17 · 31/10 16:31
A lot of content engineering challenges have pretty standard solutions, which has lead to some engineers saying agent engineering is solved/boring True to some extent, but there are a bunch of (fun) new ones emerging as agents evolve!
X (formerly Twitter) X - The Everything App
elvis @omarsar0 · 31/10 16:06
Banger paper from IBM on decision gates for agents. They add a runtime gate to tool-using agents and keep the models and tools the same. JevAct is a check that sits between the agent and its tools. An LLM first turns the domain policy into code checks and yes/no questions.
Google DeepMind @GoogleDeepMind · 31/10 14:33
We built PlanAssist to help UK Government planning officers speed up building and renovations across the country. Built with Gemini, it has shown it can cut the time planning officers spend assessing householder applications by up to 58%. It is now being made available to 12
O
Nous Research a relayé @1ofthemanymatts
one of the many matts @1ofthemanymatts · 31/10 12:24
i think i figured out where @NousResearch is
Nous Research @NousResearch · 31/10 06:30
That was crazy
Anthropic @AnthropicAI · 31/10 02:59
New Anthropic research: Automating subtle emergent misalignment research. Red-teaming is a vital part of model safety testing, but it’s also limited by researchers’ time. To see if we could automate it, we set thousands of Claude agents the task of evading a training data
elvis @omarsar0 · 31/10 00:35
Space Bunny was @MiniMax_AI M3.1 Flash all along. 🐰🎃 So I put it to work in MiniMax Code. One prompt gave me a playable space arcade game in 17 minutes. The prompt asked for a bunny in a small spaceship, three lives, increasingly hard levels, and keyboard and touch controls.
OpenAI Developers a cité @stevendcoffey
OpenAI Developers @OpenAIDevs · 31/10 00:08
Building with the Agents API? Here's what we shipped this week for you:
Harrison Chase a relayé @ankush_gola11
Ankush Gola @ankush_gola11 · 30/10 22:37
traditional observability access patterns: - leading edge queries, eg what’s happening in the last few minutes or hours - catching breakages and paging someone fast - digging into where time is spent, errors agent observability access patterns: - to some degree, all of the
Harrison Chase @hwchase17 · 30/10 22:20
when people think of "agent observability" they may think of "datadog for agents" it's really more like "databricks for agents" we've done some stuff in this direction but expect to see more soon
elvis a cité @steipete
elvis @omarsar0 · 30/10 21:37
Build your own harness, folks! Stop complaining about companies modifying harnesses however they want. It's theirs, not yours! Instead, build your own. This isn't just about implementation disagreements; as models get more capable, the harness will keep adding constraints
Harrison Chase a relayé @LangChain
LangChain @LangChain · 30/10 21:20
There's still time to RSVP for our upcoming SF meetup with @typesafeai Join @sydneyrunkle, @allietheicon, and Seán Roche at LangChain HQ for a night of talks on building with decision models and using them to evaluate agents. Spaces are limited!
luma.com LangChain SF Meetup: Judgment Calls · Luma Join us in San Francisco (SoMa) for a night of talks on building with decision models and using them to evaluate agents. We will be joined by Sydney Runkle, a…
LlamaIndex 🦙 @llama_index · 30/10 20:42
Agentic Plus now has a new setting: max effort. It's built for your gnarliest documents, where extraction accuracy matters the most. With max effort, Agentic Plus can spend additional compute when needed, zooming in on tricky details like dense charts and diagrams to extract
E
ElevenLabs @ElevenLabs · 30/10 20:41
What does it take to enable humans to engage with AI agents about their health insurance? @CloverHealth found out, with voice agents facilitating how members engage with their health plan, including care check-ins, navigating plan benefits and instant access to member inquiries,
OpenAI @OpenAI · 30/10 20:38
There are many futures. Which do you want to build?
ClaudeDevs a relayé @claudeai
Claude @claudeai · 30/10 20:30
We’re renaming our Max plans: Max 5x → Max Max 20x → Ultra Max is still $100/month, and Ultra is still $200. If you’re a subscriber, there’s nothing you need to do. We’re also removing the 5-hour limit for verified Ultra accounts, starting today with most existing
E
Emad a relayé @emmanuel_2m
Emm | scenario.com @emmanuel_2m · 30/10 19:58
We just open-sourced an entire 3D studio. Introducing 3D OS by Scenario. It gives your coding agent 6 departments and 80 skills. 📁 3D OS ┃ ┣ 📁 Animation ┃ ┣ 📁 Rigging Any Character ┃ ┗ 📁 Motion From a Video ┃ ┣ 📁 Art ┃ ┣ 📁 Previs and Storyboards ┃ ┗ 📁
elvis a cité @shobhitbanga
elvis @omarsar0 · 30/10 19:37
The next breakthrough in voice AI might not be another model taking the #1 spot. It might be a better understanding of what we should be measuring. Voice Arena is pushing that conversation forward.
OpenAI Developers @OpenAIDevs · 30/10 19:29
Meet @Scav, who brings builders together in a shared robotics community where they push the edges on what's possible with AI in the physical world. In one experiment, GPT-6 Astra remotely controlled a robot arm: