will be in Mumbai for GFF from tomorrow till Friday.
if you want to chat ethereum, tokenisation & devcon, let’s catch up!
Satyaki
Satyaki's post on X, curated on eChai
-
Satyaki
More Founder Notes
-
Meta launched Muse on 8 September: a standalone agent app that keeps working after you close it and comes back when it needs an approval. Meta's own examples of what it does are sending an email, booking travel, opening a browser and filling out forms, negotiating on someone's behalf, selling a car for more, lowering a bill, adjusting a training plan as the rest of someone's life shifts, and turning a recipe reel saved on Instagram into a grocery list. An agent that browses the open web can be instructed by the page it is reading, which is what has kept agents away from cards and passwords. Meta's answer is to give each person's agent its own dedicated machine in the cloud, contained so no one else's agent can reach it, and to put a second agent called Sentinel on that same machine. Meta's security write-up calls Sentinel "the sole permission authority" both for actions on connected services and for all outbound traffic: Muse holds surrogate tokens that Sentinel swaps for the real credential at the network boundary, so the agent "never sees real tokens", and an approval is put to the person in the app itself rather than inside the conversation, so nothing Muse is reading can ask on its own behalf. That write-up is also where Meta says what the launch post does not: "Muse isn't immune to attack. Prompt injection remains an open problem in the industry", and Muse will sometimes get things wrong. It runs on Muse Spark 1.3, which Meta calls its most capable model to date. It has a file system and a terminal of its own, so it can write code and build the tools a task needs, and a full browser to search, fill out forms and complete transactions. Every account gets an audit trail of what the agent has done and what it plans to do next. It pays through Link built by Stripe, and Meta says it is the first AI agent covered by Link's purchase protections; Shop Pay is coming, along with 1Password support so it can use logins a person already has. Free to start, with paid plans at $20 and $100 a month, and it is the United States only for now, on iOS, Android, the web and inside WhatsApp.
Muse
1h
Introducing Muse, your personal AI agent from Meta that gets things done across every part of life.
Download the Muse app and get started:
-
The co-author, at Anthropic, quoting OpenAI's post back at itself. He had been silent all day: he was on neither of the calls, and everything said about him until now had been said by other people. Three things in it. He picks out the same hedge as the sentence worth reading and credits them for it. He has started reading the proof, which is the first public word from anybody outside OpenAI on what is actually in it, and his read is that it looks closer to a different Euler blowup proof he and Buckmaster already had than to the work they published. And he adds a detail about the calls that was not in Buckmaster's statement, which is his own account and nobody else has confirmed it: that a millennium prize was offered if he would be taken off the paper. What he does not do is take the fight, which is worth noticing on a page this long.
Levent Alpöge
2h
“we cannot rule out that de-identified data derived from their usage of our products helped improve our models.”
i mean props to them for straight coming clean.
(so far the proof looks more along the lines of another euler blowup proof we had, off of whose ansatz naming we were making really stupid puns like “smooth criminale”, unlike the much better “ideal fluids explode”, Tristan)
so i’ll now give a bit on my thinking here. i actually woulda been pumped to collaborate on this, there are a lot of people at oai i like (ok, clearly some were indirectly dicks to me because of being part of the whole situation, but im a big boy, i still like them), idgaf about authorship on that step anyway, coulda been me Tristan and every fte at oai for all i care (on that Tristan would disagree:p). but on hearing the loud convo in the hallway, especially the part where a millennium prize was offered if i’d just be removed from the paper, it was kinda clear the die had been cast and things were locked. pretty wacky, unstrategic, and unnecessary, since on my side things were mostly me and claude having a good time yoloing random stuff in the corner rather than anything institutional. i also like the idea of the labs cooperating, and even better on scientific progress. it’s a shame!
-
Joseph Howlett reported the announcement the same evening, for the magazine that ran the misconduct piece in August. Bubeck gave a press briefing and emphatically denied that Buckmaster and Alpöge's work influenced OpenAI, saying "We did not use their prompt or models or proof", that the model solved the Euler problem by totally different means, and that the Navier-Stokes proof was produced over the weekend; he also granted that the Navier-Stokes route does follow a similar method to theirs. The quotation is Diego Córdoba, who with Luis Martínez-Zoroa opened the route everybody has been arguing about, hearing that it may have been walked to the end without him.
Diego Córdoba
We're a little bit in shock... If it's done, that will be a big surprise for us.
eChai Partner Brands
The partnership →
