San Francisco
- Tweets 1,581
- Following 563
- Followers 16,988
- Likes 3,009
Molly Cantillon retweeted
🧵Today I am hosting my birthday party on the USS Hornet, the Aircraft Carrier that did the Apollo 11 recovery and was pivotal in WWII.
When inviting people, most peoples reaction was "wait, you can do that????!!!!" Well it almost didn't happen!
Much of a life is shaped by what happens to reach you: the job you heard about secondhand, the friend who introduced you to your partner, the book that set a career in motion. These encounters are mostly accidental. The world holds far more people, ideas, and opportunities than anyone could survey, so each of us sees only the thin slice that drifts our way.
2
6
47
6,742
we forget why agents are addicting...
attention is the oldest luxury. for centuries, the privilege of handing your affairs to another mind was reserved for only those who could afford staff.
intelligence can now concentrate on a single life, and now on a billion of them at once.
3
32
4,693
several notable observations on gpt-live:
- zero dead air on pickup (handshake in 1 vs 6)
- genuinely concurrent mid-call handoffs and tool calls
- live has a chronic lisp problem lol
Just posted our technical deep dive on the GPT-Live system, by @zahanm and yours truly.
10
5,870
Molly Cantillon retweeted
An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science.
We believe it will be a major step for scientific reasoning. openai.com/index/ten-advance…
10 proofs from our next major model Astra on long-standing open problems in mathematics and theoretical computer science (also including new circuit lower bounds for computing the permanent!)
GPT-5.6 has already enabled so much exciting work in math and science. Can’t wait to see what comes next!
839
2,528
17,520
13,350,943
we all agree this should exist.
labs wont do it because their economics require serving the mean.
no one cares about you, in particular.
truly out of distribution performance demands obsession over ONE person’s context. getting as close as fanatically possible.
so we start bespoke, at the highest-leverage seats, where a six-figure token bill is a rounding error
we owe the world this one
heynox.com/caspian
dear claude code & codex teams, please, for the love of god, where is my executive super assistant that has:
(1) a deep understanding of me via great memory, just pack 200k context with every chat. you can build this personal store from past chats, but also i'll just give you all my data, respond to 100 different personal questions, give you all my Apple Notes and iMessage
(2) a no-chat interface. i don't want something that forgets me everytime, that i have to skip to the right chat. just ditch chats! its all one chat history, same as a real assistant. compaction and retrieval cannot be that hard
(3) i can trust to reasonable certainty with my credit cards, info, email access, etc. and i can delegate to go do things for me. i just want something that can book a flight, hotel, refill my runpod credits, update all my old passwords, cancel all unused subscriptions
(4) it is good at pinging me by sending me a text when i need to answer something for it or do something it can't. good back and forth communication. it also can't die all the time.
i do think the idea of 'responses' is wrong, it should just always be 'on' in idle mode, ready to act on scheduled tasks. it should have lots of webhooks/awaits, for example, maybe it should text me everytime i receive an important email that needs my attention. a small model can scan everything and then call a tool to wake up
BIGGER ASKS:
(5) please let it call people for me. go schedule drs. appointments, sit on hold for me, etc. same with access to imessage/email etc on just handling things for me that i don't want to care about.
(6) computer use that doesn't interrupt my work (works on another desktop on my mac/its own windows?) and is faster than me at filling forms and doing things. like how are models slower than me at typing and filling out forms than me still!
i will pay you $5k a month minimum for this. if you need me to buy some physical hardware or rent some cloud thing that's fine. i'm fine with taking on some security risk, it's that valuable, you take risk with a human EA too
chatgpt almost feels deprecated as a product compared to codex now. like models in codex are significantly less helpless and more capable than chatgpt, such that i don't want to use chat anymore (other than as a google replacement). plz fix or just build an integrated experience
- will
12
1
78
27,361
Codex cooked so we thought, so shall we.
Check out our permissions dialog for accessibility settings :) @heynox
Asked computer use to reverse engineer itself and vibecode this lil guy 🙈 github.com/zats/permiso
11
7
272
67,440
It is open sourced by @zats github.com/zats/permiso
1
1
15
3,591
Molly Cantillon retweeted
two friends used AI to clone Epstein's Gmail
it's been used by over 150 million people
and it was built in just 5 hours – here's how they did it
62
133
1,199
164,113
i read california's entire $498 billion budget so you don't have to
the wealth tax is supposed to fix our deficit. instead its accelerating it
39
111
978
412,170
CA faces an extreme concentration problem: a tiny number of taxpayers matter enormously. what happens if they all leave?
i tracked every billionaire leaving california, where they went, how much tax revenue walked out the door
scraped 10-K filings, IRS migration data, census finance records, and FOIAs
of the 12 publicly leaving, they take $1.07T in wealth - and with it, $12.7B in tax revenue. that's 19% of the entire deficit
the budget is broken at every level. SF spends 5.7x more per homeless person than K12 student.
4
26
270
21,483
the wealth tax is meant to fix this. instead it's accelerating the exodus
see it at where.tax
plug in your income to see where your $$ go
5
8
133
8,146