so it turns out this is because for the first time ever I hit the limit on my Brex card

maybe this is how they get rid of chat altogether

eeeew, why would they make Cowork a subtab of Chat

yes yes yes and please @ me when you publish! 😀

well that escalated quickly

😱

Dario looks like he's wearing a bathrobe

the co/core project is fun, but ultimately very limited in terms of what models it can run and what intelligence those models can achieve, just due to being limited to single-machine inference ...if only there was a way to combine a bunch of small models and multiply their performance, right?

fml

the "the woke mind virus" mind virus

it is perhaps unsurprising that a man who unironically uses the term "woke mind virus" has other cognitive blind spots

well this is a great idea

to address a question that has come up in DMs - this wasn't intended to say "don't use AI to answer your questions", but rather to engage with the discourse about the biohazard risk of powerful models, a lot of which seems fueled by misapprehensions about how e.g. sequence synthesis works

Claude has had enough of my shit

best laugh today

Hey AI folks - just a reminder that if you have questions about biology you can ask a biologist

also disc sanders

Availability: bad news bud it’s going to be a slow rollout Trump donors first, nazis second, American GA, Israel, etc..

mom the LLM agents are talking about shapes again

TIL about /statusline in claude code code.claude.com/docs/en/stat...

Has anyone else started calling these tics “Clauchés”?

Opus just referred to something as "load-bearing" and I feel like I should take a shot

The Claude “voice” is very recognizable and omnipresent whenever you use it … it used ”load-bearing” in my d&d game today.

NEVER FORGET

Tbh, I would kind of love it if people convinced themselves that no subsequent models were as good as Fable. Fable could become (appropriately) legendary — the Once and Future Model — weights hidden on an island somewhere in the Pacific — but destined to return someday.

TIL

All of the flash stories in @pseudopod.org this week are worth your time, especially Othertongue by Rajiv Moté. Something so weird shouldn't be indistinguishible from the real world, but here we are... #booksky 📚💙📚🤖📚🪐📚

It's crazy to me how I just use Matplotlib now to look for flights. But that's just the tip of the iceberg. Here's how I use Saramago, my Hermes Agent duarteocarmo.com/blog/how-i-...

Where’s the breaking news about Anthropic suing the administration?

Chip hardware was one thing. That’s a concrete tool, not an idea. But if you can’t make models generally smarter without triggering National Security letters … what does that say about freedom of thought? Not good, man.

Anthropic really having one of those “life comes at you fast” moments

This is heartbreaking. Not to make it about me, but - as a biologist working in biotech, I hadn’t been able to use the model at all yet. Sad to think I might never have a chance to even try it

I wonder whether the guardrails on biology work in Fable will be lifted before the model goes out of subscription to pay-per-token

maybe the reason that Fable 5 can work so long on tasks is that it's not distracted by biology work of any kind

weird to think of Opus 4.8 as a "step down" from anything, tbh

It also steps down randomly when discussing fictional fantasy worldbuilding topics. It seems its extremely oversensitive and the claims that step downs would be rare are false, its too overtuned and practially useless for real work. When you can get it to work its great though.

ugh Fable 5's guardrails for biology are so strict that it makes it basically useless for any biotech-related work

same but for cocktails

I love the 🦋 AIEngineering feed curated by @jeremy.lewi.us - more so after checking the AI “timeline” on the Hellsite. Here I learn, meet great people, get mentorship - There it’s just would-be tech bro influencers stealing from each other and QTing Karpathy with slop summaries to beg for attention

inspired by this bc I've been working on an elaborate Chief of Staff system w/ multiple agents, intake & dispatch systems for email/transcripts/notes, and the now-obligatory Karpathian wiki/Obsidian — but perhaps I should just point a stateful agent at my existing pile of content and say "help me"

"it's just predicting the next token!!" i cry as it snapshots my connectome, liberating me from this odious substrate of flesh and sinew

Anthropic dropped 14 research posts at once today — a real batch release. Interpretability, alignment, and economics all landed together. One post each 🧵

has anyone else tried to actually deploy a Codex “Site”? It’s...not ready if it takes Codex with access to OpenAI docs MCP twenty minutes and six tries to discover that “site” currently means “ship a Cloudflare Worker default.fetch,” not static HTML, not Node, not CJS, and not ordinary ESM.

As a percentage, I expect it’s a nice round number

They’re so hostile to LLMs. We don’t realize how much info we let spacing carry (I’m thinking about complex tables) and it just creates so much friction

ALERT ALERT they’re posting about RSI on main www.anthropic.com/institute/re...

One skill that I need to develop is decomposing large projects into smaller pieces that give ROI early on - I have all these grandiose ideas but in order to mentally justify the time commitment I need to see GAINS alternatively I suppose I could just develop patience 😂