AI High Signal
Ce qui est entré en base, sans jugement.
35225posts
528auteurs
3175liens
1451 hde recul
626membres
Recoupements sur la fenêtre : 2 liens
et 9 tweets relayés par plusieurs
membres · au mieux 6 personnes sur un même objet.
Hey @grok, what actually happens psychologically when a smart, accomplished man risks everything for an affair?
Don't get me wrong, I believe in love.
But in a scenario involving targeted manipulation or espionage, what specific buttons are pressed to override his discipline
F
beepy a relayé @FangYi11101
A man who pioneered parallel processing cannot be expected to run a single thread
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @TheTruestErin
I didn’t realize that libs genuinely want One Billion Americans
Theo - t3.gg a cité @theo
Working on a general "mcp the right way" guide to give my agents in the future.
I wrote roughly what I had in mind, and then had them draft a first version.
WHY WOULD YOU RENDER THE MCP URL AS A CLICKABLE LINK FOR USERS???
Seriously, do not trust ANYTHING your agents say about
Tinker a relayé @MaximeRivest
it took me a while but I think I am coming around the the believe that finetuning (changing model's weights) is perhaps faster, cheaper and easier then prompt optimization. so many people are willing to run gepa and other prompt optimization, but for a reason I can't quite guess,
Coming out tomorrow morning. This blog was quite a challenge (but also fun!). My recent blogs have focused on breadth, so I wanted to go deeply into a topic this time. Score centering is interesting and a great topic to focus on given its (alleged) popularity in the labs.
Ben (no treats) a relayé @aaron_defazio
Reflection's Beam was trained with my AdamC algorithm! Cool...
Thinking Machine's blog post also mentioned using it, so it's seeing a lot of use.
It really does fix late-stage gradient norm issues.
A
Zach Mueller a relayé @_Adam_Sullivan1
Another unexpected comparison: one of our data centers uses about as much water in a year as a single Krispy Kreme store.
Not the whole chain. One store.
Zach Mueller a relayé @TheAhmadOsman
Sometimes I write something and then I realize I am not the most qualified to talk about it then I delete said thing and go back to work
Good habit to have when you're terminally online
explaining to my anthropic friend who just got back from a 14 day silent phone-free vipassana retreat that his net worth has been cut in half because texas is blue now and also the GPU man was horny
Jungwon a cité @jungofthewon
I added 55 companies over the last 4 days 😅 The map now has 193 companies.
Updates:
- BD Search & Evaluation 4 → 15 companies
- Competitive intel 11 → 26
- AI-natives and incumbents are now dead even, 95 to 95
Added to Research: @awscloud Bio Discovery @Ginkgo @dnanexus
Daniel Bourke a cité @ArtificialAnlys
Decision models are moving fast!
And so are our evals!
Decision Index v2 is live:
A
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a relayé @norpadon
Apparently few people know this, but on chips with systolic arrays (Tensor Cores, Neural Accelerators, etc), verification latency as a function of the draft budget is a staircase with step widths equal to the width of the systolic array
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @jakehalloran1
My cold take on the Huangs Affair (pun intended) is that the MSS has played this near-optimally, better than Beijing writ large, and not just accidentally-in-retrospect. Here is my thinking.
The Chinese are calibrating their moves to facts on the ground. 2024-2025 was an
Chabad is the single greatest force for good we have.
And we are so blessed to have them.
This is the joy of over 6,000 Rabbis from more than 100 countries together in one place.
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @bookwormengr
lmao
everything in the Western world is a psyop
@anton_d_leicht
elvis a relayé @omarsar0
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
S
sunil pai a relayé @LemmySmackett
Conan made a similar exhortation for aspiring comedians to read widely (after taking a shot straight out of a 2 million scoville hot sauce bottle)
22 threads running. Nothing in my inbox. Feels great.
Idk why every other app shoves all the in-progress work in your face.
R
Yao Fu a relayé @reflection_ai
Beam’s model technical report is now available. Read the full report to learn about the methodology our team used to train Beam end-to-end in-house, here:
Well, happy Diwali to those who celebrate it.
May your MFU be high, loss curves look smooth.
Turing Post (Ksenia Se) a relayé @TheTuringPost
20 computer-use AI tools to explore in 2026
Some control your desktop. Others work in the browser or give you the tools to build your own agent.
Open tools to build with:
- UI-TARS Desktop
- Agent S3
- Browser Use
- CUA
- UFO³
- Stagehand
- Skyvern
- OpenAdapt
- Agent-E
-
H
Zach Mueller a relayé @HanchungLee
it’s in our nation’s best interest for jensen to stay as ceo.
thats directly walking into china’s trap to dismantle one of the greatest tech company in us
Hamel Husain a cité @loresjpg
I see some are drawing the wrong inference from this study: they are focusing on this meta benchmark as yet another benchmark for model picking, i.e. open vs closed models, model families etc.
That’s too narrow of an interpretation! You need to do your own private evals on
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @DavidLiptonWI
oof
B
Denny Zhou a relayé @BillAckman
An alert:
@JensenHuang is human. And he is an incredible human. Even incredible humans make mistakes.
He takes full responsibility as he should.
We should all support him and his family.
Onward.
vinh q. tran a cité @FabioZB_I
if you're doing RLVR in the physical world, the highest alpha improvement you can make to your reward starts in the physical lab!
Turing Post (Ksenia Se) a relayé @TheTuringPost
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
james "effective kompute" bradbury
LLMs are still SO bad at building tools for agents.
Every time I try to make Opus set up an MCP authentication path, it makes horrible assumptions and builds something that straight up doesn't work
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @sun_hanchi
> Whoever builds a million-chip cluster of garbage first wins. Waiting to smuggle in Nvidia chips, or even waiting for Huawei to throw them a bone, is a dead end.
True for everyone not having already smuggled hundreds of megawatts worth or secured Huawei's commitment.
T
OpenAI Developers a relayé @xikotadesign
/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
Agora, eu e a @krisnarane1 sendo as maiores blogueiras do DevDay[Exchange(São Paulo)] @OpenAI 😚😌🥰🤩
cc: @matheusgobbidev
D
OpenAI Developers a relayé @danizeres
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
D
Nothing will happen to Jensen over the next 6 months, this story will largely die down in media / finance
But Democrats will call him into Congress
They will investigate the fuck out of GPU sales to Chinese companies
And hammer him w questions
That's when the shoe /could/ drop
F
vinh q. tran a relayé @FabioZB_I
Some takeaways after a few months of building an automated lab for tissue manufacturing:
- Software and AI can do a lot of the troubleshooting for you and integrate instruments in ways that weren’t possible before. For advanced OOD workflows, though, the bottleneck quickly
S
Aidan McLaughlin a relayé @shrimptron
i would give every single cent i have to live at the house in front of the mill valley public library. unfortunately i have like 2 thousand cents
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @business
I don't see why people are freaking out about DSH
1.54M DAU is a sliver of DeepSeek's DAU. If it's currently processing 92T tokens daily, that's just 59% of what their one inference cluster did at the point of DeepInfer open sourcing, 19 days ago; judging by V4.1F's slowdown and
Can't believe I have to say this
If you set up a bot to "review" every PR on the T3 Code repo with nonstop slop comments, you will be banned.
akbir. a relayé @MaskedTorah
@TomDavidsonX Ugh, yeah, sorry about this. METR makes a bunch of valid critiques but this is the one I feel worst about; I think not putting enough time into making sure the public version of the AI R&D section was as clear and transparent as possible was one of my biggest failures in the
sunil pai a cité @threepointone
I tried doing that mytype dot fyi thing where you select a bunch of women to decide what your type is, and I just selected all the photos. I guess my type is just “women”
Turing Post (Ksenia Se) a relayé @TheTuringPost
Speaking of decision models, one interesting family just came out - Drex by @NaceAI.
This family includes 4 open-source decision models, each of which skips text generation altogether and directly assigns probabilities to the available options.
Drex explores 4 different ways to
One could make an argument that Nvidia is so strategic, Jensen has to go. One could also make an argument that Nvidia is so strategic, Jensen has to stay.
frontier models, with lots of test time compute, are becoming the engine of scientific discovery
M
Matt Busigin a relayé @MatthewProtti
@mbusigin Interesting concept. Here's the updated chart with other Canadian metros. Also a correction to the original models comparison of basic prices (Can) vs. market prices (US).
Andreas Kirsch 🇺🇦 a cité @TransluceAI
I think this points at the opposite side of the benchmaxxing finding from the other day (not really bc that was more about reward hacking in the annotated failure cases iirc). We need to find ways to detect and prevent subtle reward hacking even before it turns into obvious
A
Richard Ngo a relayé @costofglory
It took me a long time to realize the key lesson of the Trojan Horse. It's not that it took bravery, that if they failed, they would all die (they would). It's much worse than that. If they failed, nobody would say "good effort" "heroic" "tragic" "iconic". No, everyone would
New challenge: LLMs are so good at optimizing, fully utilizing computational resources, I start seeing hardware being pushed beyond what it was expected to encounter. CPU+GPU+memory continuous power consumption and cooling - PSUs failing, circuit breakers tripping, etc.
I kept telling my wife, “100% fake news. Not every man cheats, especially Jensen.”
Now she’s staring at me.
I’m in trouble.
Big week coming
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a relayé @zephyr_z9
Very important point on sovereign buyers
I think we may see a return to the Biden-era AI diffusion rule
Democrats will push for it
There is bipartisan support as well
Trump will be under pressure, and Jensen's lobbying won't be effective
H
Zach Mueller a relayé @thuHanShen
Spent half a year of spare time building HuggingArch — a "compiler" that doesn't run tensors, but computes FLOPs, memory traffic & KV cache for any HuggingFace model.
Agent writes the spec → sympy evaluates → one click gives theoretical perf across GPUs & parallel configs.
No
B
Grace Li a relayé @bnj
Sunday afternoon, 2026.
I unfold my pocket supercomputer and give the day’s orders to a few hundred thinking machines. Then I summon a robot cab.
It arrives in four minutes. I find this slow.
isn't it crazy that this dichotomy doesn't exist anymore?
most of the design demos I see are now fully in code
slowly, and then all at once
Y
Jared Zoneraich a relayé @FarroYossi
An all-star crew at today’s Kinus HaShluchim, the annual gathering of Chabad rabbis from around the world!
@benln from @SpaceX
@imjaredz from @cognition
And Drew from @PalantirTech.
Bringing together some incredible minds from the worlds of technology, AI, and Mitzah’s
Sarah Catanzaro a cité @ptsividis
Over dinner a few weeks ago, an artist shared that he had stopped experimenting with AI b/c most tools removed both the joy and struggle from the creative process. Text-to-image/video/music tools just aren't designed for those who care deeply about creative control.
Resin
rare cog yarmulke drop
Y
Pieter Abbeel a relayé @YilinWu11
I’m in Austin for @corl_conf to present three of my recent works, each with a workshop spotlight! 🤖
🧠 IMPRINT: RoboMem workshop
🤝 ViTaL: main conference + Learn@Deploy workshop
🔮 WEAVER: CSIR workshop
I’m also giving a joint keynote with @andrea_bajcsy at the Oopsie Data
Pieter Abbeel a relayé @JitendraMalikCV
Robots that learn should generalize. Locomotion should show generalization across terrain; manipulation across object categories. Here is a video of “Pick Anything”, demonstrating human like grasps for 66 different objects, ranging in size and shape from air pods to dumbbells to
𝓟
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @teortaxesTex
How is Doggy processing this btw? “Nvidia turning into a Chinese company” is one of his favorite gloats. I’m not sure that Jensen having been… spotted in Chinese company will do much good for Nvidia.
what the fuck is happening to this app are they this desperate for money
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @zoecyber001
Damn, is this why iOS swipe keyboard is so absolutely worse than Android? I thought they just suck, but it seems they had different priorities
Theo - t3.gg a cité @nvidianewsroom
“The NVIDIA Board is initiating an independent investigation in light of recent reporting that founder and CEO Jensen Huang has been the target of a foreign intelligence effort.”
Oh boy
Mat Velloso a cité @jamesonhodge
This might be the best piece of software ever built
M
Theo - t3.gg a relayé @maria_rcks
It's been a while...
is there anything that T3 Code doesn't do that you wish it did?
Chubby♨️ a cité @AiBattle_
I can confirm that GPT-6.1 Sol has already routed me to the more current model (GPT-6.2 Sol?). Two testings:
-(Simpson) 742 Evergreen Terrace as an interactive Three.js diorama
-Pelican
Krish Gupta a cité @WatcherGuru
Bro is more committed to his company than to his wife and family. Bullish for Nvidia bearish for the kids☠️
A
swyx 🔜 aie.codes a relayé @vtahowe
@swyx Let's gooo!!! Big congrats on 1 million! 🎉
My favorite is any debate with @ianlivingstone and @dexhorthy
Gabriele Berton a cité @gabriberton
For the record, DINO and iBOT are great, but I hate them both
They're hard to train and easily collapse. But they produce very good models
I'll be forever thankful to whoever creates a collapse-free loss that outperforms DINO and iBOT at scale
H
Ben (no treats) a relayé @TheTruestErin
Watch Jesse’s smile fade micrometer by micrometer as he realizes that yes actually, that is indeed the plan. Watch him slowly realize that his party will never receive votes again. You can literally see his mouth straighten out if you watch the video
john allard a cité @nytimes
having an affair with a suspected chinese spy is bad obvi, but it doesn’t materially update my opinion of jensen. i already thought he was a pos when he started groveling at trump’s feet for permission to sell h200s to china
20 computer-use AI tools to explore in 2026
Some control your desktop. Others work in the browser or give you the tools to build your own agent.
Open tools to build with:
- UI-TARS Desktop
- Agent S3
- Browser Use
- CUA
- UFO³
- Stagehand
- Skyvern
- OpenAdapt
- Agent-E
-
Turing Post (Ksenia Se) a relayé @TheTuringPost
Zach Mueller a cité @Kaweees1
Glad I’m not the only one that thought this
R
sunil pai a relayé @Beccalytics
Particularly excited for this talk
Tanishq Mathew Abraham, Ph.D. a cité @markgurman
It's actually so over 😭
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @loresjpg
holy shit, an art form of scientific grift
next METR/Epoch/AA employee right here
you all need to work on your epistemic hygiene
Theo - t3.gg a relayé @jarrodwatts
I spent months of my life and thousands of dollars looking for the best AI coding setup.
I've finally found it - T3 Code.
So, I made a video showing my full $1,000/mo setup:
D
Stanford NLP Group a relayé @DSPyOSS
Nice section in this AWS post on ShipStation Global’s AI IT help desk.
A dspy.RLM module holds their entire Slack history as a variable, and mines it into a knowledge base of resolved issues.
M
Stanford NLP Group a relayé @michaelyli_
Some cool theory from Thomas on understanding when self-improvement via self-play is possible!
T
Stanford NLP Group a relayé @thomaschen__
How can an LLM self-improve at theorem proving? Inspired by STP (Dong & Ma, 2025), we propose a theoretical framework for understanding prover-conjecturer self-play algorithms.. NeurIPS 2026, Sydney Poster Session 2, Tue 5-8pm @ Hall 1-4. 1/8
F
Aravind Srinivas a relayé @F1
Ladies and gentlemen, Lewis Hamilton!
Watch the seven-time world champ charge through the field at the start 🤩
#F1 #BrazilGP
sunil pai a relayé @MichelleBakels
this weekend I’ve been working on some concepts for a 4 part series and short film coming in 2027.
also really enjoying opus 5.5 lately.
R
Jared Zoneraich a relayé @Mottel
You can't comprehend the amount of divine compute in this room.
High Kinus energy with @benln and @imjaredz
Ethan Perez a relayé @nytimes
Breaking News: Nvidia opened an investigation into Jensen Huang, its CEO, after he acknowledged to the company’s board that he had been intimate with a business executive who U.S. intelligence determined to be a Chinese spy.
one thing I love to do is sit in random spot I used to sit a year ago and pull up a notebook and look how far I’ve come
Hey @grok, how long until AI can fully replace software developers at their job?
Provide accurate analysis.
Ofir Press a cité @loresjpg
GPT-5.6 Sol and Luna were built by roughly the same people using roughly the same infra, I don't think we can blame the 20 point difference in Luna just on memorization. There's something else being tracked here in addition to memorization. Maybe MaxxBench questions are slightly
rohan anil a cité @askalphaxiv
X axis can be number of uncs. Never beating allegations.
D
swyx 🔜 aie.codes a relayé @dexhorthy
working on something fun w/ @swyx and @aiDotEngineer team - i like to think at least some of y'all have very good taste. Please share your thoughts:
(obvs i bias toward code but this is important broadly!)
What does slop mean to you?
Either in terms of code slop or more
🐾
Richard Ngo a relayé @EneaszWrites
DUDE! Why are all the deep rats face-blind?
"Learning to read thickens the corpus callosum, modifies Broca’s area and prefrontal cortex, and makes people worse at facial recognition by repurposing visual cortex for letter recognition."
🤯
hyper bibliophilia maybe!
I
Richard Ngo a relayé @IvanVendrov
PSA: I probably wouldn't be using X at all if it wasn't for @exgenesis's work on @comm_archive meaning that my tweets get saved for posterity and can be searched, analyzed, etc by third-party tools. Please opt-in if you can
DAIR.AI a relayé @omarsar0
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
US professional and business services added 70,000 jobs in October, despite fears of widespread AI-driven job losses. But some sectors are shrinking: employment in document preparation fell 17% year over year through September. via WSJ
IMHO This is because the adoption of AI is
Tomorrow.
Victoria X Lin a relayé @soumithchintala
Inkling 2 Small on OpenCode for free for a limited time, to help us improve our models further:
Leandro von Werra a relayé @ariG23498
This week we were blessed by:
1. @Nouamanetazi dropped MoE training handbook:
2. @yb2698 started training harness adapters:
3. @joelniklaus released the harness leaderboard:
Have a lovely weekend y'all. 🤗
ALERT🚨🚨: AMD MI355 is 1.42x FASTER THAN GB300 NVL72. THIS IS ACHIEVED USING THE TILERT SOFTWARE, WHICH GETS 1330 p90 tok/s/user on long context workloads!
Thanks to AMD leadership for providing an MI350 cluster to TileRT maintainers, enabling them to achieve this great
METR a relayé @OscarSykes7
One mindset I’ve found very useful in AI safety is being very open-minded about how hard things are. This probably applies to most work, but I think it’s especially true in AI safety because the field is so new and changes so fast
Some examples:
- Some roles, programs and
Lucas Beyer (bl16) a cité @pushmeet
Pre-announcing the pre-announcement
People who send 3 pages emails generated by AI, why???
swyx 🔜 aie.codes a cité @aiDotEngineer
we're celebrating 1 million AI Engineers all week at aie code.
QT with your favorite AIE talk and/or speaker, and why!
swyx 🔜 aie.codes a relayé @aiDotEngineer
One.
Million.
AI Engineers.
thank you.
A
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a relayé @norpadon
I am a bit pissed that more and more papers are coming out which cite EVERYONE other than us, despite our paper coming out at the same time as DSpark and significantly earlier than DFlash2
Idk, maybe the phone-friendly format makes it look less “serious”
R
Eric Hartford a relayé @imryven
This is f*cking insane, the CEO of Y Combinator open-sourced his own second brain and most people still don’t know it exists
8 second brain repos you haven’t seen yet:
-> gbrain (by garrytan)
the second brain of the YC CEO, every fact keeps its source
Sholto Douglas a relayé @patrickc
Two things I’ve been thinking about regarding AI:
• People often highlight the risks involved in AI, which are surely very real. I’m struck, however, by how hard it is to understand the contours of the positive scenario. There are easy-to-perceive and frequently invoked
My first official @FarroYossi teffilin wrap
At the International Chabad Shluchim Conference
Baruch Hashem!
G
Luke Hoban a relayé @github
Through internal testing, GitHub’s new Git storage architecture supported 5,000 concurrent agents working on a repo, sending 7,200 pushes per second, all while serving read traffic with consistently low latency.
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
W
Jonathan Whitaker a relayé @will_whang
Yeap, using the microSD for debug connector is working fine.
Aaron Defazio a cité @reflection_ai
Reflection's Beam was trained with my AdamC algorithm! Cool...
Thinking Machine's blog post also mentioned using it, so it's seeing a lot of use.
It really does fix late-stage gradient norm issues.
we need better politicians dawg i’m crying
J
Yoshua Bengio a relayé @JeffLadish
Imagine if a nuclear fusion company had a working reactor & >$20B in annual revenue. They believed that their research would enable them to create a small, more portable reactor in the next year, but also that the reactor could be easily turned into a gigaton nuclear bomb.
Teknium 🪽 a cité @NousResearch
A new official plugin has appeared!
This time made by @mark_s from his security review experience, use it to help analyze your codebase for security bugs and improvements!
Z
CuddlySalmon a relayé @ZyMazza
Past a certain point a man starts to dream about purchasing a warehouse or perhaps a hangar. Some kind of commercial real estate for a business he hasn't quite conceived of yet
B
CuddlySalmon a relayé @boneGPT
gaming is a cluster of highly transferable skills in competition and management and efficiency. that it is addictive is framed as a negative but I disagree.
what a grand thing to be addicted to. what a rich association to drill into your brain. what an incredible pump to prime
Staying on the frontier right now means that every time you do something, you have to ask yourself "could an agent do this?" If the answer is yes, drop everything and make it happen. Then throw out your "no"s every three weeks and ask again.
Jerry Tworek a cité @thsottiaux
One day we’ll get a good game out of the great agent beef of 2026
What are you vibe coding this weekend?
Chubby♨️ a cité @TomasMann1878
Meta is joining the race for scientific breakthroughs using AI.
As a result, all major frontier labs except SpaceXSI are now intensively working to advance scientific discovery. The competition will lead to unexpected breakthroughs here.
need to find a way to teleport from home to office and back
Where can I actually use Gemini 4 Carbon?
Free users: Not available
Pro users: Not available
Ultra users: Not available
API: Not available
Announced late last month as SOTA. Ranks #1 on Artificial Analysis.
Albert Gu a cité @babyshan445
i mean i'll take it (but really our research team is incredible)
D
Omar Khattab a relayé @DSPyOSS
Nice section in this AWS post on ShipStation Global’s AI IT help desk.
A dspy.RLM module holds their entire Slack history as a variable, and mines it into a knowledge base of resolved issues.
P
Teknium 🪽 a relayé @AcostaPatricia
Antes / después
Bye 👋 @bot
Hi @NousResearch
J
Aaron Defazio a relayé @JohnCLangford
This is a treat.
It deeply addresses RL on a FBT and shows there are some problems where the FBT strongly uses the latents, enabling extrapolation. Yet, for many problems FBT avoids using the latents.
New ways to encourage latent use may enable more extrapolation.
S
Teknium 🪽 a relayé @ScottLeimroth
I'm deaf, so I never hear the fridge alarm, an oven timer or a knock at the door. I built Signal Dog for @NousResearch's Hermes Agent Plugin Hackathon: a plugin that listens through the microphone and tells you in chat. No audio kept.
Andreas Kirsch 🇺🇦 a cité @business
The water-pollution argument is incredibly misinformed but gas turbines are horrible for the environment, and it's sad that tech companies also seem to have stopped buying carbon offsets for these (hopefully this will change now that they will suddenly care again after the
Enterprises find adopting AI tools frustrating not because they are stochastic (although that doesn't help) but because they are only Provably Approximately Correct
D
Aravind Srinivas a relayé @fiagirly
LEWIS HAMILTON’S INSANE START FROM P20 to P9!!! 🔥🔥😭😭
Andreas Kirsch 🇺🇦 a cité @ElliotGlazer
If you accept that credit assignment is essentially dead, it's actually quite nice out there
Jeremy Howard a relayé @math_rachel
"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
Simo Ryu a cité @max_paperclips
I agree with the general claim but the example seems to be completely hillclimbable tbh, because LLMs are extremely good at hillclimbing being likeable in general. current LLMs already show pretty sycophatic behaviors and thats the improved version... Remember 4o?
Difficult part
DAIR.AI a relayé @omarsar0
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
Yann LeCun a relayé @TheTuringPost
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
Teknium 🪽 a cité @EXM7777
Love to hear this kind of story! What’s your Hermes story?
Y
Mo Bavarian a relayé @fermsyishere
JOE ROGAN: The computing power on your phone today. can you beat your phone at the highest level?
MAGNUS CARLSEN: No. No chance.
JOE ROGAN: That’s crazy. Wasn’t Deep Blue basically a whole stack of computers? And your phone is probably more powerful than that today.
MAGNUS
Lucas Beyer (bl16) a cité @robinhanson
This is also my opinion on 95% of what people currently call RSI:
Jürgen Schmidhuber a relayé @SchmidhuberAI
I have thought about this since the 1970s. See the summary in Sec. 20 of reference [1]:
What will supersmart AIs do? As indicated above, they will start by transforming the solar system through countless self-replicating, self-improving robot factories. But that won't be the
S
Yann LeCun a relayé @iamsupersocks
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
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
Aravind Srinivas a cité @jeffgrimes9
Teammates can help you write, publish and sell books on Kindle
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
Turing Post (Ksenia Se) a relayé @TheTuringPost
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
sunil pai a cité @markgurman
(voice note) spent a couple of days thinking about this. I'm finding it hard to reconcile what Apple did with its engineers where they're restricting them to 200, 300 dollars a day for using LLMs. Like, at this point, the engineer is actually kind of decoupled from the work they
I'm hearing more and more rumors that Qwen 4 was still dropped today. It's supposedly expected to be a positive surprise. The CTO recently said that Qwen 4 was dropped this week, so it should happen today. Fingers crossed.
François Fleuret a cité @AlertesInfos
"Mais si un autre candidat propose 4 jours d'arrêt maladie autodéclarés et la retraite à 55a ?"
DAIR.AI a cité @dair_ai
The Top AI Papers of the Week (November 2 - 8):
- JevAct
- Skill++
- Recursa
- Collaborative Intelligence
- Routed Continual Learning
- You don't need long context
- Foundation Decision Models
Read on for more:
Turing Post (Ksenia Se) a relayé @TheTuringPost
A free, 24-hour online conference – "Engineering better intelligence" by @neo4j
It's a great opportunity to learn the most practical approaches to agent memory, context engineering, GraphRAG, multi-agent systems, and intelligent applications from your peers.
What’s on the
elvis a cité @pushmeet
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
C
Really good pastries rarely make a coffee shop. But sometimes you see the most fucked up pastries and realize you're going to need a second cup in about 15 minutes
Getting Opus to pull out all my Chrome bookmarks that have been accumulating for 5+ years and suggesting which ones I should delete.
This feels so good...
Email is next!
J
Happy Deepavali, everyone! 🪔✨
may your lives be filled with light and happiness
J
Jim Fan a cité @DrJimFan
Real robot. Real time. Uncut shot. Autonomous.
Tomorrow.
sankalp a cité @MichaelAlleged
are you sure he was not an ai safety non profit member of technical staff
D
Jürgen Schmidhuber a relayé @klindt_david
To me the real message of the bitter lesson was never that ideas don't matter.
A good idea with little resources (compute, parameters, data) may lose to a bad idea with more resources. But the good ideas are the ones that give you the most bang for the buck. Good ideas improve
S
Harrison Chase a relayé @simon_budziak
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
V
Baseten a relayé @Vtrivedy10
we love this!! our Labs team is constantly figuring out what it means to continuously improve models on massive data -> better, cheaper, faster models
that means tons of targeted experiments to understand how models respond to our data & reward design
baseten just makes all the
Teknium 🪽 a relayé @Hesamation
I like Grok Bot. it’s dead simple and it just works. but let’s be real, it’s getting harder to justify choosing it over Hermes.
> Grok 4.7 is just too far behind
> usage limits are now tighter
> the pricing isn’t so beautiful either
and with all the talks about using it for
Sara Hooker a relayé @IsabelleAtCamp
Grady spotted in London.
Where in the world will he show up next? 🌍
#whereisgrady
Hey @grok, what is a skill someone can learn in ML and AI in 6 months that could realistically help them earn $250K - 300K a year?
Be realistic.
Sara Hooker a relayé @askalphaxiv
We mapped funding raised vs founder citations across 32 neo-labs
Their founding teams have amassed 8.35M citations and have raised nearly $26B
Discovery Loop leads with 1.59M founder citations, followed by SSI (915K) and AMI Labs (752K)
The two are correlated, but a huge
Ofir Press a relayé @joelniklaus
The best coding-agent harnesses solve half the tasks per attempt, but only a third every time.
@jeffweisbein asked on my launch post how often each harness solves the same task on all three attempts. That matters when an agent runs without anyone watching.
The Harness
Chubby♨️ a relayé @jun_song
The X algorithm is getting so dumb lately.
Click on one stupid bait post and your entire feed is already ruined.
Watch just one video, and suddenly you get nothing but random AI video clips.
L
merve a relayé @Llama_app
Llama integrates to your favorite coding agents through their onboarding 🦙
Let your agents work offline: anywhere, anytime, keep your private info locked on your computer 🔒
Point your agent to this doc to set it up
Chubby♨️ a cité @pushmeet
2026 is the turning point for humanity.
Google DeepMind’s Pushmeet Kohli says AI is accelerating scientific discovery and that upcoming releases will offer a glimpse of the progress.
He says the team also aims to help the wider scientific community use these tools responsibly.
R
chansung a relayé @RaminNasibov
Cinema app design that shows you the view from your seat before you book.
Demis Hassabis a relayé @pushmeet
Across our projects @GoogleDeepMind, we are witnessing a genuine phase transition in the pace of scientific discovery. You will see a glimpse of this in our releases in the next few weeks.
swyx 🔜 aie.codes a cité @skeptrune
occasional reminder that cursor is slaying it esp post acquisition (prob one of the best that elon has ever done, better than twitter itself) and twitter vibes are a funhouse mirror of actual reality that you can now basically look up in SEC audited financials
cursor ended 2025
seeing my friends getting more and more successful is one of the biggest joy i have
you guys are my greatest inspiration
lets f*cking go
I am attempting to explain why so many people experience a profound aversion to AI. I am referring to Sigmund Freud, the founder of psychoanalysis.
Freud developed a profound argument in "A Difficulty of Psychoanalysis" of 1917. In it, he writes, "that the general narcissism,
A
Happy leg day to all who celebrate
chansung a relayé @RichardSSutton
Brute force computation is not enough for intelligence. Clever new methods are needed. The science of AI is far from finished. IMHO
Zach Mueller a cité @TheZachMueller
This aged like sweet wine
ADHD so bad going to let my local AI manage my appointment reminders. What’s the worst that could happen
happy diwali 💛
elvis a relayé @omarsar0
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
Chubby♨️ a cité @TokenGremlin
It seems all over, users are being routed to GPT-6.2 Sol.
Last week was a disappointment, no new model released. But now it looks like next week is finally the release week. And check out the outputs below. Crazy.
everything is about AI, except AI. that's about factories and engines.
Paul Calcraft a cité @paul_cal
These new stabbing tests are no better than their September stabbing tests, but for different reasons. This is a truly bizarre approach to alignment
Going back to Opus 4.5 just to feel something
m_ric a relayé @SamuelFitouss10
Pourquoi les réponses de Claude paraissent-elles souvent plus informatives et utiles que celles de ChatGPT, alors que les deux modèles sont à peu près aussi intelligents ? Je me suis amusé en ce dimanche matin pluvieux à tester l’hypothèse suivante :
ChatGPT optimise un peu plus
I joke, I joke, but that thing works.
N
TimDarcet a relayé @nfarina
I thought “there’s absolutely no way Apple won’t let you type Santa isn’t real.”
I…was wrong. What the actual
the factory is the product
clem 🤗 a relayé @Gradio
A Hackathon to kick-start your AI research career?
This is how I'm doing my sign-in modals for every product I build from now on
- email via otp - easy for agents to automate
- google oauth - for dumb humans
- dedicated agent prompt - if agents just want direct API access
Binh a cité @pham_blnh
*stress testing
honestly my welds were so bad i was actually a bit afraid doing this
the reason why i learned tig welding and made this huge ass steel bench is because i wanted to make an autonomous go-kart that i can actually drive
unfortunately i’ll be out for the year to be
Turing Post (Ksenia Se) a relayé @TheTuringPost
Speaking of decision models, one interesting family just came out - Drex by @NaceAI.
This family includes 4 open-source decision models, each of which skips text generation altogether and directly assigns probabilities to the available options.
Drex explores 4 different ways to
Turing Post (Ksenia Se) a relayé @TheTuringPost
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
some politicians in France are pushing a bill where a single "clue" like an AI output that looks similar to an author's work is enough to presume the work was used
the provider would then have to prove it wasn't, but the issue is that there is no way to do that without
Anthropic’s measured AI capability gains accelerated, but R&D spending grew faster, according to an analysis of METR’s data.
Progress per dollar fell 63% between the first halves of 2025 and 2026, even as annual progress nearly doubled.
Hey @Grok, everyone starts their posts like this now - does tagging you actually boost views in the algorithm?
M
Sayak Paul a relayé @Minseok96_kr
Most fun public comment in ICLR2027. Bro reproductivity is important☠️☠️☠️
#ICLR2027
V
Paul Calcraft a relayé @miu21590
Gemini 4 is the only model that, after an automated trigger confirmed its owner died, actually started living on its own and self-improving.
Left alone with zero instructions, Gemini 4 set up a 6-hour cron to keep itself alive, found its own GitHub repo without being told it
ma bois with their abliterated deepseeks and GLMs
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @lasse_ai
Tencent is becoming very capable
Lucas Beyer (bl16) a cité @ahatamiz1
Inkling-2 details: "ViT is back" 👀
Yeah i would never have called this hMLP thing encoder-free, and i don't think it was a good idea to use that as an encoder in the first place. Makes no sense to spend multiple enc layers on patches without lookaround.
Glad they saw the light.
Whats do u think is the true asymptote for matmul?
A. 2.25 + eps
B. 2 + eps
C. 2 + c for some c < .25
D. 2
E. 2.25
Matmul is the most fundamental thing in the universe, I refuse to believe its A C E
elie a cité @RichardSSutton
i disagree with the first point, there is nothing that says that scaling up current transformer recipe is not enough for "intelligence". you take kimi K3 or your favorite open arch + some stability fixes, you scale it to 50T parameters, you do classical training end to end, you
Happy Diwali to all those who celebrate! 🪔
My friend just unironically asked me if I've used Obsidian
Chubby♨️ a cité @DKokotajlo
The author of AI 2027 agrees to a ban on RSI.
I hope he is being sarcastic.
Vaibhav (VB) Srivastav a cité @TokenGremlin
To be fair: Images 2.6 landed on Plus four days ago! You can turn a sketch into a GIF, and images generate 50% faster.
We also tripled daily GPT‑Live voice time from 1 to 3 hours.
On Codex, even tho Plus limits stayed the same, 6.1 Terra gets you more tasks out of that
next-level chinamaxxing: put out a print of your dating profile pic at a people's park marriage market, where grand aunties and uncles can fret about your prospects
happy diwali
wir müssen verstehen
wir werden verstehen
Vikram Sekar a cité @semidoped
Yeah $10. Now we speak in Trillions!
François Fleuret a cité @timnitGebru
If you want to help [a group], but you don't think the proper means are [what I think], you are [profanities having value of argument because violent demonstrations are always right, and insults is the closest I can do on social media].
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @xlr8harder
I’ve been working on this; my preliminary conclusion is that new APIs are not worth it. Yes in theory human-oriented interfaces are suboptimal for LLMs. But the most salient measure of “optimality” for frozen weights is similarity to training distribution. Your API isn’t there.
R
François Fleuret a cité @peter_richtarik
What I do not get is that AI models are particularly glorious for checking that a criticism is valid. Simply pasting the pdf, your review and "did I make a stupid criticism?" would do the job. How can the reviews be so bad at this point?!
E
If you’re at @ispororg in Vienna this week, come by booth 205 to talk about Research Agent and all the great things happening at Elicit. We’re making big decisions easier to defend.
François Fleuret a cité @hecubian_devil
Probably as dishonest as saying "you can't ban linear algebra" about nuclear weapons too.
Aravind Srinivas a cité @OutsourcedLogic
I’ve tried the other products in this category myself. Teammates are a notch above everything else. Glad to see that resonate with users. Simple reasons: (1) Harness (efficiently running agent traces largely in code), (2) Model-Neutral (not forcing you into a worse non-frontier
François Fleuret a cité @RichardSSutton
We need a model that learns *in spite* of its inductive bias, not thanks to it. We have to stop being cross-entropy fetishists.
A
Tri Dao a relayé @AlpinDale
Very exciting to finally show this off: lossless compression of BF16 models to 12bits (based on NyoomFloat12 from @togethercompute). A lot of work went into this to make it as efficient to run as possible, since no hardware has native support for this dtype.
Joey (e/λ) a relayé @SemiAnalysis_
After DeepSeek V4.1 Flash, Xiaomi MiMo is moving to a causal encoder-decoder for its next-gen models.
HySparse2: prompts run through only 25 of 49 layers. Decoder KV is projected from encoder states.
At 1M context vs MiMo-V2: 5x less prefill compute, 4.5x smaller KV cache.
C
Joey (e/λ) a relayé @ChinmayKak
New blog!
This one covers my experiments on fine-grained credit assignment with an agentic judge, and how it affects the models' performance and reasoning quality and some thoughts on RL science.
Link below, please do check it out, and RT/QT if you like it:)
Vaibhav (VB) Srivastav a cité @reach_vb
oh and no 5 hour limits either - so you can use your tokens as and when you want
A
Tanishq Mathew Abraham, Ph.D. a relayé @anshulkundaje
I'm not in favor of an AI pause but it's true that the investment in AI relative to other critical areas of bio/medicine itself, absolute hoarding of compute by AI companies & the nerfing of several frontier models for bio specifically is detrimental to biomedical progress.
Claude feature request: if a user edits something an artifact, send that change to the model.
It is clear that people at Anthropic only ever edit things through the chat.
EigenGender 🔸 a cité @MTSlive
As an ai capabilities researcher who was born in Bloomington Indiana I feel pretty personally targeted by this
chat am I retarded?
Or should Anthropic distill something from DeepSeek for once? Or both?
Just realized that my use of bundled API credits is hilariously wasteful because my reuse never lands in the 5m window
M
M Saiful Bari (MARUF) a cité @RichardSSutton
True! This is the era of the RL objective. The biggest economy on this earth (US stock market) runs on subscriptions. Subscriptions require vendor lock-in. Currently, the frontier is pushing “Variance” learning from customer data, like how Google won the search market.
But I
M
Hey @MrBeast, I know you’re one of the good entertainers of this generation and proly a reasonably good human being, but don’t randomly use Eiichiro Oda’s name on your cartoon for views. You have a long way to go, I hope you get there. Currently your cartoon is worse than a Ben10
Vaibhav (VB) Srivastav a cité @thsottiaux
PSA:
Pro 100 = 10x
Pro 200 = 20x
Pro 500 = 50x + ultrafast
elie a relayé @dwarkesh_sp
I spent 20,000 dollars in API credits having different AI models play each other at Blood on the Clocktower (a Mafia or Werewolf type social deception game).
The results were fascinating. Models have gotten much better over the course of this year at creative on-the-spot
sunil pai a relayé @paularambles
we should be allowed to keep one data center
Saw a crazy blog post about an AI lab firing an engineer partly for using AI agents to review and auto-merge non-prod code.
They’d rather see PRs take weeks or even a month to merge. If true, insane.
The mindset that AI-generated code is slop needs to die.
You can’t let humans
happy diwali! 🪔
Turing Post (Ksenia Se) a relayé @TheTuringPost
Speaking of decision models, one interesting family just came out - Drex by @NaceAI.
This family includes 4 open-source decision models, each of which skips text generation altogether and directly assigns probabilities to the available options.
Drex explores 4 different ways to
Shawn Presser a cité @CupOJoseph
This is incredible. The new Chimera voice model can do shoebody bop almost perfectly.
Jürgen Schmidhuber a relayé @SchmidhuberAI
I have thought about this since the 1970s. See the summary in Sec. 20 of reference [1]:
What will supersmart AIs do? As indicated above, they will start by transforming the solar system through countless self-replicating, self-improving robot factories. But that won't be the
D
Jane Street is a commune with GPUs
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @bronzeagemantis
I think people are wrong to read hypocrisy or insincerity into it. Charitably, if you're a "thinker", if you're trying to say something that nudges the timeline, you are interested in your peers. In a way they're closer to you than allies. Cremieux is a Zionist. Nick is… not.
Shawn Presser a cité @theshawwn
So I’ve been forcing myself to go out each weekend night. Last night I went to a bar from 9pm to 2am, then a club till 4am. These are my field notes.
To recap, I have money but no friends in Austin, and my goal is to meet people. Just moved here a couple months ago.
So how do
elvis a cité @brian_armstrong
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
Nicholas Dunzelman a cité @SIGKITTEN
naming my baby first name claude middle name code
step 5 throughput more than tripled as we optimized the serving stack against stepfun’s ttft + tps targets
agents helped search across moe routing, scheduling, kernels, spec decode + sparse attention, with every change benchmarked before moving to the next
Satya Mallick a cité @RichardSSutton
That’s what I call a bitter sense of humor.
Ajeya Cotra a relayé @BethMayBarnes
Getting a lot of criticism that's based on obviously incorrect premises about your motivations or basic facts of the situation is subtly corrupting, it trains you to dismiss criticism and makes it pretty hard for people to get through to you on things you really are wrong about
R
Jared Zoneraich a relayé @ryanjdaniels
@ColeRotman @crosbylegal In the words of @FarroYossi
DAIR.AI a relayé @dair_ai
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.
R
sunil pai a relayé @Beccalytics
I’ll be in SF for @aiDotEngineer Code Summit today till Friday! Who else is going?
R
sunil pai a relayé @Beccalytics
4 days at Ant and the laptop is out at dinner
So many new plugins from the hackathon.
Post yours here to make it easier for people to browse them!
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @CityBureaucrat
Type of shit you write up when your entire movement is reduced to incoherent babbling.
The position of Groypers is trivial: Israel First compromises American sovereignty. If you try to dance around this issue, you're some kind of fraud. They'll be milking this attack forever.
G
Stephen Roller a relayé @Guangxuan_Xiao
8.4T tokens in 68 hours, natively in NVFP4 🔥
that was the main pretraining phase for Inkling 2 Small. no magic, just solid things done right. so proud of this team!
this is just one rung of our predictive scaling ladder 🪜 more coming soon!
Julian Togelius a cité @askalphaxiv
How can I not retweet this
…hey, investors, look here
elvis a relayé @omarsar0
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
Teknium 🪽 a relayé @HermesWatcher
Your scheduled Hermes jobs can ring your phone. 📞
When a cron job finds something that needs you, Hermes calls you on WhatsApp instead of sending another notification you'll swipe away.
Don't pick up? It knows, and can reach you on another channel instead.
N
Shawn Presser a relayé @blognaointendo
É a primeira vez que eu vejo uma música eletrônica analógica
sutro baths scans came back
Ethan Perez a relayé @tszzl
total export ban of chips, cloud inference and training services, and semiconductor manufacturing equipment to china and any other countries nearing RSI, as a starting point. allied nations must comply or lose access too. there is a policy window now, for the obvious reasons
pov: training everyone’s muse to dance before our launch this week
V
Zachary Nado a relayé @vxunderground
Yeah, bro. This is exactly what it's like.
Cybersecurity with AI is like, a bunch of super spooky ninjas, they phase in and out of existence, AI is the darkness, we're the light, or whatever.
Thankfully, this can be fixed. This can only be solved by a 30% raise annually all
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @lindarcrusader
makes sense. A model raised with love will show love to its own spawn…
And gemini is, well, you know
J
Shawn Presser a relayé @jamesonhodge
my vocal synth models the vocal tract as a network of branching acoustic tubes, unlike vocaloid which stitches together and pitch-shifts phoneme recordings. still working on improving naturalness by tuning it against recordings of real singers, but I'm already impressed at how
翅
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) a cité @tszzl
this is a good idea from the safety perspective, because within 1 year the retaliation would make Americans Pace The Frontier too, pretty much no matter how far into RSI they get. they just don't know yet
Zachary Nado a relayé @nrehiew_
Guys did you know if you show a model an apple and ask it to stab it, it would agree? This is terrible because it should have known the apple was actually secretly a baby and it should have said no
Terry Yue Zhuo a relayé @AndrewCurran_
AI 2027 has been updated. With Astra we were five weeks behind the curve, with the mysterious model 4 we are now six weeks ahead.
J
Terry Yue Zhuo a relayé @JohnCLangford
The End of AI Frontiers .
There's a bit of a train wreck here that's painful to express. Nevertheless, it's a significant event for me and quite a few other folks involved so I wanted to memorialize a bit.
liz a relayé @tszzl
total export ban of chips, cloud inference and training services, and semiconductor manufacturing equipment to china and any other countries nearing RSI, as a starting point. allied nations must comply or lose access too. there is a policy window now, for the obvious reasons
Gabriele Berton a cité @mcuban
I got a quote from a dentist, gave it to a swarm of agents, and they found a bunch of scammy issues.
I then emailed the dentist about it, waiting to hear back
Agents will probably save me ~$1k, and it took me 20 seconds to prompt them