Andrew

Was just thinking the best thing for Ant and Amazon would be to go all in together. Guess they already thought of it. They don't even have to actually merge companies. Provide enough compute and it's a de facto merge.

@SaidAitmbarek In terms of efficiency (accuracy per compute), it's not so logarithmic. Efficiency is way up compared to e.g. gpt-4. New models advance SOTA via efficiency. That's why all the labs appear equal. Because we don't know their margins. They can control raw accuracy however they want

@ai_ops_lead interesting angle, agreed about the positive correlation except that accuracy per unit-comp is logarithmic, price is linear another simple counter-point would be openai cutting prices ÷2 as well.

if anthropic dropped prices by ÷2, nobody’d use openai anymore

@Dimillian @krzyzanowskim Good time to announce an Ubuntu implementation?

@krzyzanowskim Bro

prediction: Apple to block background window clicking on macOS for security reasons

@CtrlAltDwayne Anthropic literally can't release Mythos. No GPUs for it. If they could have, they would have, because now they're like a limping gazelle on the Savannah.

Everyone's focused on whether Spud is GPT-5.5 or GPT-6. I'm more interested in what happens if Spud really is a step change. Does Anthropic release Mythos in response, does OpenAI release Spud in an intentionally diluted form while keeping the strong stuff locked up?

@jakehalloran1 For the work I do, GPT-5.4 was the biggest leap we've had since o3.

GPT 5.5 (or whatever they name it) in pro mode is clearly the best model anyone has ever publicly released but at the same time it kinda feels like we are past step change moments with models. Not that it isn’t great. Just that for most things that no longer matters

@paraschopra No, it's about a single ongoing context with a heartbeat. Has nothing to do with integrations.

I don’t get the OpenClaw hype Connecting Claude with Telegram / WhatsApp is trivially easy, you can literally ask it to help you do this and it’ll guide you. Same story with recurring jobs. I just did this - now Claude send me local bangalore news summary at 12pm IST daily on Telegram. Took me 15 mins to build. If the argument is that Claw lets nontech users do this, imagine the security implications when users let an LLM take over their system while having no idea what’s happening under the hood. Making custom scripts and workflows with Claude lets you at least know what you’re configuring on your system.

@AndrewCurran_ Mythos is "too powerful to release to the public", and Anthropic doesn't trust the government to make good choices... but they can give Mythos to the NSA.

The NSA has access to and is currently using Mythos, reporting by Axios. Dario's meeting on Friday, to which Treasury Secretary Scott Bessent was a last minute addition, was reportedly about exactly this; granting certain US Government agencies early Mythos access.

I think we'll look back on Mythos as the moment Anthropic lost. They gave OpenAI a clear target and said "here's the mark to beat. We can't serve it". Now all OAI has to do is actually serve a comparable model. Ant will continue to train models for enterprise, but that's about it

@patrickc You might want to mention the risks of leaking your genome in a world where identifying as yourself is increasingly difficult.

I'm lucky enough to have a great doctor and access to excellent Bay Area medical care. I've taken lots of standard screening tests over the years and have tried lots of "health tech" devices and tools. With all this said, by far the most useful preventative medical advice that I've ever received has come from unleashing coding agents on my genome, having them investigate my specific mutations, and having them recommend specific follow-on tests and treatments. Population averages are population averages, but we ourselves are not averages. For example, it turns out that I probably have a 30x(!) higher-than-average predisposition to melanoma. Fortunately, there are both specific supplements that help counteract the particular mutations I have, and of course I can significantly dial up my screening frequency. So, this is very useful to know. I don't know exactly how much the analysis cost, but probably less than $100. Sequencing my genome cost a few hundred dollars. (One often sees papers and articles claiming that models aren't very good at medical reasoning. These analyses are usually based on employing several-year-old models, which is a kind of ludicrous malpractice. It is true that you still have to carefully monitor the agents' reasoning, and they do on occasion jump to conclusions or skip steps, requiring some nudging and re-steering. But, overall, they are almost literally infinitely better for this kind of work than what one can otherwise obtain today.) There are still lots of questions about how this will diffuse and get adopted, but it seems very clear that medical practice is about to improve enormously. Exciting times!

@rendererr I don't use openclaw, but agents with heartbeats are still clearly the future. Openclaw is the latest checkpoint telling us "getting closer, but not quite there". Requires cheaper, better AI.

Feels like every few years something gets yanked out of niche tech circles and anointed as the current speculative mania. I remember scrolling on here when OpenClaw started gaining serious traction and nobody could name a single truly interesting use case for it. Meanwhile I watched as my local MicroCenter sold 45 Mac Minis in a day. I seriously almost bought one out of pure fomo. Then when I was in China last month, there were free events to help grandmas get their OpenClaws set up. That's when I knew 100% it was a fad. "But it texts you first!" Why the fk do you want your AI to text you? Put the mac mini in the bag bro

Projecting forward, it begins to make sense why OpenAI cancelled Sora, if world modeling is just coding. Jeez is Google going to be totally wrong yet again?

@thdxr Love all the stock footage casually scrolling on 2/3 of the screen. "I know you can listen to someone talk for 30 seconds, so here's some beep boop bedoop."

so yeah he obviously needs to shut up with this talking point but also why is this literally the only thing he ever says. like every few weeks he just says the same thing again with new made up numbers if only claude could make new material

@agrossule @pmddomingos Interesting, what's the reverse stat?

@pmddomingos Even funnier is that 79 percent of OpenAI's paying customers also pay for Anthropic.

Funny that Anthropic has the demand and OpenAI the compute.

@NicolasZu @dkundel Looks amazing

Ok this is a truly insane NEW way to design for me, WITHIN Codex If you are doing dev on a web technology, watch below In that video: - I am playing my game within Codex (yes) - I use a codex-made tool to design buildings (see some tweets below it's really powerful) - I can ask Codex to iterate and the game changes WITHOUT refreshing - I can point at UI elements - I can take a screenshot All WITHIN Codex It makes iterating on design incredible

@Srini_Pa Tell that to my Codex agent that has been running for two weeks, learning as it goes, compacting when it needs to. OpenAI teaches the model weights and we get updates every month. We teach the context and it learns from every message.

LLMs cannot learn. The frozen core never updates from experience. So the AI industry piles on prompts, RAG, memory, agents, and orchestration to preserve the illusion of learning. That illusion shatters when reality changes. Yet they think enterprise adoption is a PR problem.

@mweinbach Codex CLI runs on it all

If you don’t have a Mac and are trying to keep up with cutting edge AI, you literally can’t. Everything is Mac only or Mac first. This is a huge deal. Leading AI products are essentially built for Mac, with everything else being an afterthought. This is significant

@markchen90 @SebastienBubeck @ahelkky Tell her bout how Codex is the AGI harness for science

Not true. Science is more important than ever. The future can’t be about dumping results on the community en masse. We need to work with scientists to use AI to accelerate discovery without stripping away the artistry. Excited for @SebastienBubeck and @ahelkky (who are both amazing scientists) to take on this mandate!

The next time you're tempted to make fun of bird watchers, just remember they descend from a long line of dinosaur watchers.

@jshchnz @github When GitHub starts charging for hosting repos, we'll know who forced their hand. This is why we can't have nice things.

With my codemaxxed project surpassing 353,000,000 lines of code (not a typo) I actually got a @Github cease & desist 🪦 "We've noticed that the repository is growing fast while committing very frequently. This looks like some sort of automated activity that serves no purpose."