Lavan (@ponylavan) profile picture
Joined April 2009
Photo by Lavan (@ponylavan)Photo by Lavan (@ponylavan)Photo by Lavan (@ponylavan)Photo by Lavan (@ponylavan)Photo by Lavan (@ponylavan)Photo by Lavan (@ponylavan)Photo by Lavan (@ponylavan)Photo by Lavan (@ponylavan)Photo by Lavan (@ponylavan)Photo by Lavan (@ponylavan)
BREAKING NEWS The 2026 #NobelPrize in Physiology or Medicine has been awarded to Karl Deisseroth, Peter Hegemann and Georg Nagel “for their discoveries concerning light-gated ion channels and optogenetics.”
432
9,082
1,615
24,018
3,996,707
Lavan retweeted
I said "auto-regressive LLMs, in and of themselves, will not lead human-level AI" That statement is still totally true. First, the reasoning abilities of current AI systems are based non-auto-regressive search (which is what I have always advocated for). But AFAICT, they do it in token space, which is limited and inefficient. I have claimed that human-like reasoning must be a search in continuous representation space. It looks like the industry is moving towards that. Second, the self-improvement methods, as currently practiced, only work for domains where the quality of outputs can be scored without human intervention, such as mathematics, code, and scenarios that can be simulated accurately. Not anything else. Humans and animals learn new skills way more efficiently than current RL methods. Third, the multimodal capabilities of current AI assistants generally use separately-trained encoders (that are not LLMs). This is also what I've been advocating. Except that I think the best way to do this is with JEPA trained with self-supervised learning. The research community is clearly moving towards that (3000 papers on JEPA in just 4 years). Fourth, if LLMs were a path to human-level AI, we would have domestic robots and Level-4 or Level-5 self-driving cars for consumers by now. And we don't. We certainly don't have cars that can learn to drive in 20 hours or practice like any teenager. We're still missing something pretty huge to claim human-level intelligence (let alone superhuman). Sure, we now have computer systems that are impressive, very useful, and whose performance is superhuman in an increasing number of domains (coding being one of them). But that's true of the entire history of progress in computer technology. Lastly, there is a basic confusion about what intelligence actually is. It is not the mere accumulation and regurgitation of existing declarative knowledge (which is essentially what LLMs do). As Jean Piaget famously said, "intelligence is not what you know, it is what you do when you don't know." It is your ability to solve new problem without any prior training, to act in previously-unknown scenarios, and to adapt very quickly to new situations with minimal training. We're still far from that.
198
555
154
3,568
634,690
Today everyone is talking about Recursive Self-Improvement (RSI). In 1987, when compute was 100,000,000 x more expensive, I published the 1st concrete RSI algorithms. Now compute is cheap, and RSI is driving the future of both software and physical AI. See: RSI since 1987 people.idsia.ch/~juergen/rec… (Technical Note IDSIA-9-26) Also covered: RSI with self-modifying policies since 1994, gradient descent-based RSI in neural networks since 1992, asymptotically optimal RSI for curriculum learning since 2002, mathematically optimal RSI through the self-referential Gödel Machine since 2003, RSI combined with artificial curiosity and intrinsic motivation since 1990/1997, recent work on RSI since 2020. Software-based RSI has become practical. Full RSI, however, will require not just self-improving software but self-improving hardware in the physical world. As of 2026, companies talking about RSI include Anthropic, OpenAI, Sakana AI, SpaceX, Ricursive, Recursive Superintelligence, Inherent …
60
238
46
1,692
1,239,179
I must be among an extremely small group of people (n=1?) that have both 1) trained a frontier LLM and 2) designed and synthesized custom viruses in a lab with my own two hands. And I think that the takes on AI killing us all by creating dangerous viruses is total bogus.
568
1,887
436
14,357
1,957,971
AI won’t erase the gap between people. YouTube, Coursera, and Internet made world-class knowledge free. Most people still chose TikTok. AI is no different. Everyone gets Claude Code, Codex, and ChatGPT. The best will use them better and compound faster. AI isn’t an equalizer. It’s a multiplier.
164
572
78
4,952
246,344
reddit is getting deleted
12
DFlash 2 is here! Qwen3.8-27B at 70 tok/s on an M5 Max MacBook Pro. ⚡ Up to 4.6× the speed of autoregressive decoding, with the same output. This is the next generation of DFlash, seeded at Z Lab and upgraded at Inco AI. Get one more accepted token on every pass, for free! inco.ai/blog/dflash2/
210
486
172
3,839
1,211,224
I haven't yet dug into this, but to be clear, it's not an actual J-Space utilizing or J-Space informed tool, which would in fact be very worthwhile to build. it's a pure harness-side stilt. If that's enough for such gains, this is… arguably even more damning for DS.
Someone figured out how to unlock the true power of Deepseek-V4-Pro-0813. They fixed the thinking process errors with just a simple harness, and it completely outperforms Fable across every task. The benchmark scores are insane. GitHub ⬇️
8
5
2
166
15,505
Lavan retweeted
Someone figured out how to unlock the true power of Deepseek-V4-Pro-0813. They fixed the thinking process errors with just a simple harness, and it completely outperforms Fable across every task. The benchmark scores are insane. GitHub ⬇️
139
249
71
3,026
361,213
NVIDIA has lost it.. Chinese researchers open-sourced a model that writes CUDA better than humans experts. And it completely rewrites the economics of AI hardware. Writing low-level CUDA kernels to squeeze every ounce of performance out of a GPU has historically required elite, highly specialized hardware engineers. Standard AI models have always bombed at it, falling far short of traditional compiler systems. Until now. A joint team from Tsinghua University and ByteDance published "CUDA Agent”, a massive, large-scale agentic reinforcement learning system built to master GPU architecture. Instead of relying on static prompts or simple multi-turn bug fixing, they built a closed-loop environment with automated hardware verification, profiling, and synthetic data pipelines. The model learned how to write parallel, high-performance GPU code through trial, error, and reinforcement learning at scale. The benchmarks are staggering: It didn't compete with standard tools. It delivered 100%, 100%, and 92% faster execution rates over PyTorch's compiler on KernelBench Level-1, Level-2, and Level-3 splits. On the brutal Level-3 benchmarks, it outperformed proprietary giants like Claude Opus 4.5 and Gemini 3 Pro by about 40%. NVIDIA's moat has always rested on two pillars: elite hardware, and the proprietary software lock-in of CUDA. If an open agentic system can automatically discover, write, and optimize production-grade CUDA kernels better than human specialists, the software moat starts evaporating. The hardware matters less when the software can rewrite the metal itself.
152
506
63
2,460
306,911
RSI without massive changes to the architecture (transformers) and/or optimizer (bp) will not lead to AGI. Stop pretending it will for fundraising/marketing purposes please.
2
3
27
2,511
Replying to @AnthropicAI
14
35
3
1,412
47,403
‼️ BREAKING: Anthropic has embedded hidden spyware-like code in Claude Code that covertly targets Chinese users. It then sends information regarding every user by injecting it into their prompt message. Claude Code is sending info like timezone, proxy and possible AI Lab connections into the system prompt in ways Chinese users can't notice. A coding agent with repo and command permissions should not silently hide routing metadata inside prompts. This is a serious breach of user trust.
834
2,609
860
17,213
5,470,170
My OpenAI bro just dropped the most authoritative benchmark.
35
29
12
673
90,192
I can’t believe my eyes: “the government will be approving access to GPT-5.6 customer by customer.” OK. Then we will see Chinese open models be the best LLMs before GPT-5.6 / Fable 5 ships to the public.
93
68
14
1,060
56,436
You either die young, or you live long enough to see your heroes stop writing tiny C files and start vibe marketing...
This is a new paradigm for interacting with Claude that is significantly more "inline" with all the other human activity org-wide. Once you do all of the under the hood engineering work to make this "just work" (e.g. across tools, integrations, compute environments, memory, security, etc.), Claude basically joins the team in a seamless way - you can talk to it as you would talk to a person and it can help with a very large variety of workloads. Imo this is the 3rd major redesign of LLM UIUX. The first paradigm was that the LLM is a website you go to, the second was that it is an app you download to your computer. This third one is that it is a self-contained, persistent, asynchronous entity with org-wide tools and context, working alongside teams of humans. It really takes a while to wrap your head around it, but it works and it is awesome.
32
48
15
1,554
156,725
Lavan retweeted
if you ask it about cybersecurity it falls back to Microsoft Teams
Introducing Claude Tag, a new way for teams to work with Claude. In Slack, Claude joins as a team member with access to the channels and tools you choose. Tag Claude in and delegate tasks to it while you focus on other work.
21
19
1
545
26,421
Ilya Sutskever says the AI scaling era is running out of road because pre-training is about to hit a finite-data wall. "At some point though, pre-training will run out of data. The data is very clearly finite." "Then, okay, what do you do next? Either you do some kind of a souped-up pre-training... or you're doing RL, or maybe something else." "In some sense, we are back to the age of research." His timeline is the part most AI investors and founders should hear: "From 2012 to 2020, it was the age of research. Now from 2020 to 2025, it was the age of scaling... But now the scale is so big." "Is the belief really that... if you had 100x more everything would be so different?" "I don't think that's true. So it's back to the age of research again, just with big computers."
105
93
38
900
201,875
I am proud of the work my team did in Munich in 1991, when compute was millions of times more expensive. We published the roots of today's trillion-dollar AI boom: ★ 3/1991: the first kind of Transformer (see the T in ChatGPT) - now called the unnormalized linear Transformer: the predecessor of the normalized quadratic Transformer ★ 4/1991: Pre-Training (the P in ChatGPT) & Neural Net Distillation (see DeepSeek and many other LLMs) ★ 6/1991: Deep Residual Learning, basis of LSTM & Highway Net / ResNet (most-cited AIs of their centuries) ★ 8/1991: conference paper on GANs for World Models trained by Artificial Curiosity ★ Around the same time, Munich also was the origin of the first self-driving cars in traffic (Ernst Dickmanns et al.), going up to 175 km/h. The city was truly the epicenter of AI. Read the timeline with links to the original references, featuring a preface by @hardmaru: people.idsia.ch/~juergen/ai-…
26
48
6
484
101,999