Yeah, the speed of progress now will depend on the compressibility of the relevant circuits. I wouldn't even be completely certain that GPT-6 will use more training compute than GPT-5. The main reason is that we have multiple discoveries in model efficiency that require testing,…
Yeah, the speed of progress now will depend on the compressibility of the relevant circuits. I wouldn't even be completely certain that GPT-6 will use more training compute than GPT-5. The main reason is that we have multiple discoveries in model efficiency that require testing,…
Yep, I constantly see people wrongly assume that the things they can buy now with $20 or $200 subscriptions represent the current SOTA. It's partially true at the raw LLM level, but far from true for complex assemblies. There are IMO, IOI, ICPC gold level systems, for example,…
Yep, I constantly see people wrongly assume that the things they can buy now with $20 or $200 subscriptions represent the current SOTA. It's partially true at the raw LLM level, but far from true for complex assemblies. There are IMO, IOI, ICPC gold level systems, for example,…
I don't understand why anyone uses SWE bench verified to track the capabilities of models that far into the future. It's like this report has zero predictive value and was created for people with zero knowledge of forecasting in this area. I would speculate more about the…
I don't understand why anyone uses SWE bench verified to track the capabilities of models that far into the future. It's like this report has zero predictive value and was created for people with zero knowledge of forecasting in this area. I would speculate more about the…
Gary Marcus is a master at confusing balloons with birds, and he distracts from the fact that we don't have any way to be sure that balloons aren't early versions of rockets. Even if a rocket is completely different on every level from a balloon, a balloon shows that flight is…
Gary Marcus is a master at confusing balloons with birds, and he distracts from the fact that we don't have any way to be sure that balloons aren't early versions of rockets. Even if a rocket is completely different on every level from a balloon, a balloon shows that flight is…
The battle for the enterprise market continues. The revenue potential is much larger, obviously, especially when models get serious. At this moment, it’s obvious to think about Anthropic, what will they do? They raised a lot of money recently. If they believe they are able to…
The battle for the enterprise market continues. The revenue potential is much larger, obviously, especially when models get serious. At this moment, it’s obvious to think about Anthropic, what will they do? They raised a lot of money recently. If they believe they are able to…
I will reiterate my timeline prediction. Before the end of 2027, there will be technology that will be able to replace 95% of white collar jobs as they are done today. That doesn’t mean it will replace them, because of the typical friction in human systems and a lack of GPUs.…
I believe this is correct, and I need to admit I am not sure how long it will be sustained. My first instinct, based on hardware, suggests that Moore’s Law has already stopped and that there are only two years left of cluster size increases. But why do I think this way when we…
There will be tradeoffs for sure, but IQ, and especially knowledge per joule, will be much higher on average. I’m still bullish on maximizing every type of sparsity.
There will be tradeoffs for sure, but IQ, and especially knowledge per joule, will be much higher on average. I’m still bullish on maximizing every type of sparsity.
28.7% on HLE with web search, fine-tuned from GPT-OSS-20B. This result doesn’t make sense to me. I assume that a model of this size obviously cannot store the relevant bits of this knowledge because it’s too small. This looks like serious generalization (at least in certain…
28.7% on HLE with web search, fine-tuned from GPT-OSS-20B. This result doesn’t make sense to me. I assume that a model of this size obviously cannot store the relevant bits of this knowledge because it’s too small. This looks like serious generalization (at least in certain…
1K Followers 5K Following"Fame is a form, perhaps the worst form, of incomprehension."
-Borg(es)
Notes to self(?)
Some intent matter of re-presentation🪞
Harmony, Melody, Rhythm by the
20K Followers 9K FollowingProgramme Director @ARIA_research | accelerate mathematical modelling with AI and categorical systems theory » build safe transformative AI » cancel heat death
3K Followers 911 FollowingStartups, emotions & intuition, AI.
Learn this communication skill to be a better partner/friend/collaborator: https://t.co/HCagOrKpud
6K Followers 5K FollowingFOLLOWS 🫵 https://t.co/F7MzDOTC1k
ML/AI R&D eng quant trade noisy ASR TTS
Open AGI, ASI…—Open AI compute4 */acc 😇
It's more fun to compute 🥰 I feel understood 🤯
3K Followers 17 FollowingHigh-volume account of @ESYudkowsky, the original AI alignment guy. If it's missing punctuation, it's humor. If you can't tell, it's probably also humor.
7K Followers 369 FollowingDedicated to the protection and thriving of sentient beings. PhD in evo bio.🔸
Executive Director of @PauseAIUS. Opinions not necessarily those of the org.
11K Followers 461 FollowingHelping the world prepare for extremely powerful AI @open_phil (views my own), writer and editor of Planned Obsolescence newsletter.
1K Followers 5K Following"Fame is a form, perhaps the worst form, of incomprehension."
-Borg(es)
Notes to self(?)
Some intent matter of re-presentation🪞
Harmony, Melody, Rhythm by the
20K Followers 916 Following@hashmatrixmax CEO |@Columbia | Stoyteller of top Asian-built, high-potential AI products making their mark in the U.S. 📮[email protected]
3K Followers 334 FollowingMeditation︱Fractal Epistemology︱Conscience
Meditation/Phenomenology tutoring for (de)objectifying the mind. Be a bodhisattva to all of yourself
393 Followers 683 FollowingPsychologist (with PhD in bioinformatics) meets AI. I build cognitive interfaces that steer AI and reshape human thinking. Prev: Biotech → VR → Nootropics → AI
25K Followers 5K FollowingTurnaround CTO, advisor, and startup vagabond. Former head of AI @NASA CAS and tech wonk for (Obama) @WhiteHouse, DOD, and DOJ. Tweets are my own.
495K Followers 152 FollowingNobel Laureate. Co-Founder & CEO @GoogleDeepMind - working on AGI. Solving disease @IsomorphicLabs. Trying to understand the fundamental nature of reality.
31K Followers 130 FollowingI call shake-ups before the Street trades them | InsiderEdge LIVE M–F 9am PST (link below👇) | Yahoo Finance | Mom of 9 | Trade like you have the CEO’s playbook
22K Followers 1K FollowingI make games with AI. Product Designer, Co-founder at https://t.co/C6xWXxJrsL & https://t.co/5jKwzHKucU. Founding Designer at RedMart, acquired by Alibaba.
19K Followers 1K Followingferal awareness researcher ᐧ strangely earnest ᐧ I teach Alexander Technique, which helps you get out of your own way → https://t.co/e0qcOYJBfG
10K Followers 138 FollowingBuilding super-tiny AI models that (hopefully) think • YC S25 • Made @DiffusionBee • Previously: AI research @Microsoft, @CarnegieMellon, @Meta
3K Followers 911 FollowingStartups, emotions & intuition, AI.
Learn this communication skill to be a better partner/friend/collaborator: https://t.co/HCagOrKpud
57K Followers 2K FollowingHead of Design @Cursor_ai. Early @NotionHQ, @Stripe, built startups. I make a world where anyone can make software. Aspiring k-pop idol.
6K Followers 5K FollowingFOLLOWS 🫵 https://t.co/F7MzDOTC1k
ML/AI R&D eng quant trade noisy ASR TTS
Open AGI, ASI…—Open AI compute4 */acc 😇
It's more fun to compute 🥰 I feel understood 🤯