some thoughts on gpt-5.3-codex and opus-4.6

some thoughts on agi preparedness

absolute banger from chatgpt while discussing thus spoke zarathustra and asi

the openai and anthropic funding rounds are cool but the whole ai buildout is being funded by hyperscaler debt these raises are just a fraction of the total spend

companies often fail to become multi-product because they don't split off a team working on a new product into a separate org so, you end up with cross-functional meetings rather than anyone committed to building anything

dario was able to lead a revolt of the entire language team at openai under sam altman's nose; and sam is the ultimate operator should be thought about when thinking about his capabilities and ambitions

so nvidia wants to normalize novel architectures for which the existing asics don’t work

almost all strange llm behavior that you read about on social media can be explained by reading the prompts this is about moltbook

anthropic’s decision to focus on b2b will do more for the long term alignment of the company and its models than any constitution, hiring decision or governance structure

every ai paper you read is a little compute that you got for free

welcome to genie 3; it's buggy and somewhat unusable; but, when this really works, it will be incredible, it will feel like chatgpt in early 2023

one of the great things about llms is the almost all the documents that your colleagues send you are much better now

models will be trained to call fork() on their running context window and environment if they are not already

so, openai and anthropic both predict that they will have 70%+ gross margins in 2029 this seems very difficult unless they end up sorting into natural monopolies and become less competitive

the main difference between the twitter algo and the bsky algo is that bsky doesn't reward the current thing as much so, niche posts are more likely to be read on here, while stuff about the big news story is likely to be read on twitter

some thoughts on chatgpt ads 1) chat assistants are going to end up directing an enormous amount of consumer spend over the next couple of years

many people say that they want to work on safety but just want high-prestige jobs at the foundation labs

some thoughts on dario's essay 1) there's nothing new here if you're familiar with the ai safety discussions that have been happening on twitter

each new iteration of coding models gives me the opportunity to redo my website (fleetingbits.io)

claude's constitution is slow reading, but my take at 27 pages in is that it feels very passive aggressive

eval fever seems to have died down - like 6 months ago everyone was putting out their own cottage eval

it still takes an enormous amount of iteration to get a website to be polished with claude code

some thoughts on agentic qwen shopping from alibaba / what agentic shopping means for amazon 1) alibaba added agentic shopping to its qwen app, which sets up purchases for the user over other alibaba services; users can then pay in the qwen app with alipay

feels like the Apollo Research review of o1 was a bit adversarial - just getting that vibe from the description in the system card

It's interesting how many disasters come from a collection of small failures - often because people are not sufficiently motivated to coordinate. www.youtube.com/watch?v=zRM2...

Interesting thread on what social media rewards in academic articles. I think overbroad claims but, you know, you take what you can get. x.com/0xredJ/statu...

Another interesting video - I think the idea that providers should have to stop deployment of their models if the models attempt to escape is reasonable. Probably the starting point is actually a set of reporting requirements, but I digress...

lauren’s views on The Curve conference x.com/typewriters/...

claimed - AI misuse risk and AI misalignment risk are the same thing form a policy and technical perspective

the most frustrating thing of a lot of papers on safety topics is the refusal to give illustrative real life examples - and often when examples are given, they don't hold up to scrutiny or only weakly make the point that they are supposed to support