Alex

Now that I've used @OpenAI Codex a bit more, it's quite evident they have superior software quality, which makes it a lot more fun to use than Claude Code currently. Just feels more robust and responsive, which makes a big difference. @sama

@interaction Let's go, congrats on the launch!!

Starting today, personal superintelligence is just one tap away. No download, no signup. Text Poke for free now: Poke.com 🌴 — 0:00 – What's Poke? 0:50 – Introducing Poke Recipes 1:25 –  Create a Recipe in 10 seconds 1:43 – Earn on Poke 2:44 – Build with npx poke 12:58 – Recap 13:36 – Parisian Love

In case of agentic coding, the winter break acted as a synchronization barrier, where enough people suddenly caught on to something that had been possible for months before in really similar form, although Opus 4.5 is indeed really good 3/3

A few days into January 2020 I got a text from someone with a news story about a respiratory disease in Wuhan, China. Then, more and more news came out how bad the situation really was. But it still took more than two months before the lockdowns and market crash happened. 2/3

One learning from the Clawdbot hype and generally the past few years for me has been how public consensus about the state of the world evolves. Often the underlying state changes gradually but there is a sudden synchronization barrier where everyone suddenly catches on 1/n

@s_bielmeier thanks :)

@avrecum congrats!!

Got a paper accepted at #ICLR2026 🥳🥳🥳 Will share more details soon

More and more convinced that if we stay within the current paradigm, the AI era of the coming few years could be described as “chasing the 9’s” where you keep improving robustness by meticulously curating data to eventually get to a level of reliability that is acceptable

I love how there’s literally a name for everything, e.g. recently I found out “L2 phonological transfer” is when someone uses phonemes from their primary language (L1) to speak their secondary language (L2)

If you automate all the jobs with enterprise AI you have evaporated the consumer AI market, since no consumer has any money left to pay for anything anymore. So if you’re truly AGI-pilled consumer AI is a bet on UBI

Wow, 77 minutes of thinking on a single problem with GPT 5.2 Pro

This is the longest one yet

The name “OpenReview” should have been a giveaway Just staying true to its name 😭

@nonieengel @xai Congrats 🥳🥳

just joined @xai this week! so excited to work with this team and make grok imagine more creative, tasteful, and fun together :)

@eddybuild @iamRezaSayar did you try upload_large_folder()? I had issues with rate limits and stuff when uploading huge datasets but just syncing the entire folder with that function worked well

@iamRezaSayar huggingface unfortunately does not scale

today, we're releasing the largest egocentric dataset of physical jobs - 400k action labels - 2.5k clips - 2x'd open source dataset size (download below)

I can't believe the name "ChatGPT" stuck around, but it turned out it's much more memorable than any of the other deliberately crafted names, despite being a four syllables long technical acronym Perhaps since AI was very novel, it made sense to have a new unique name style

anyone got experience in suing tech giants for anticompetitive/monopolistic behavior? we could really use some help 🙏🏼 @dhh @TimSweeneyEpic

That's really interesting! Did you somehow also analyze the concepts that the SEE and HEAR cues amplified in the latent space, to try to understand what these amplified directions/features which move the embeddings closer to a vision embedding likely represent? E.g. by using an SAE and interpreting the semantic meaning of some neurons of it which tend to be more active on the cues vs. with no cue, or some other technique

In arxiv.org/abs/2510.02425, we find if you ask an LLM to “imagine seeing,” then how it processes text becomes more like how a vision system would represent that same scene. If you ask it to “imagine hearing,” its representation becomes more like that of an auditory model. 3/9

Over the past year, my lab has been working on fleshing out theory/applications of the Platonic Representation Hypothesis. Today I want to share two new works on this topic: Eliciting higher alignment: arxiv.org/abs/2510.02425 Unpaired rep learning: arxiv.org/abs/2510.08492 1/9

Blaming AI for increased energy prices is a popular take and might be true regionally, but general electricity price increase mainly comes from Ukraine war which caused negative NatGas supply shock, and overall inflation. If you factor these out, the chart will be mostly flat.