I ask a lot of dumb questions. I work on infra, systems and stuff @meta. Still here for fintwit, but posting incessantly on https://xitter.cf/t.co/4w3jVQl0Dy
Third circle of hell
Joined April 2014
- Tweets1.3K
- Following2.2K
- Followers307
- Likes5K
Anhad Jai Singh retweeted
I led engineering at Google DeepMind.
Today, I'm proud to introduce Fo to give personal AI something no lab ever has... Humans.
Other personal AI's pretend AI can do everything. Fo employs humans to do tasks that AI cannot.
- 2x better at real-world task completion (beats other agents by 69%)
- 94% trust rate (4x less likely to leak private info vs Muse, Instinct)
Sign up for free: wajo.ai/join-wajo
As people have pointed out, this is basically Karpathy's auto research idea, but you probably want something like github.com/uditgoenka/autore… which is ready to set up and use. If you've tried that and it didn't work - tell us what didn't work
What's the puzzle? I'm wondering if it does much better given the move sequence chart from the start vs. the board image itself.
The simplest answer here is that it's probably not in the training dataset of the frontier labs yet - when it is, they'll probably get better. This is the whole jagged intelligence idea.
Good list to see everything that shipped, but putting dispatch in D tier (while missing channels entirely?) and putting Computer use in C tier is criminal.
They are both actual game changers. Auto mode should be nowhere close to S tier.
Anthropic shipped 120+ features in 90 days across Claude Code, Cowork, and Claude.
I ranked every single one.
S tier, A tier, B tier, C tier, D tier. What to adopt now, what to skip, and 4 workflows that chain them together:
🔗 news.aakashg.com/p/anthropic…
Was listening to the pod and one thing that stood out to me was the confusion between Tracy's question about memory notes (something chatGPT web does) and file writing for Claude code and compaction. I feel like you guys should actually talk to @bcherny for way more deets!
Anhad Jai Singh retweeted
Claude Code with Opus 4.5 is a watershed moment, moving software creation from an artisanal, craftsman activity to a true industrial process.
It’s the Gutenberg press. The sewing machine. The photo camera.
FWIW - I think these are active decisions made inside the big frontier model labs all the time - datasets for things like Coding, Math, Music, Writing, Quant skills are high ROI to add. biz/co will pay for them. Poker, Chess, Pokemon etc aren't, but you could be pretty easily.
I have a different take on this - LLMs are smart over what they've seen before. Poker/chess just aren't dedicated datasets for training frontier models yet. If someone took the corpus of all hist games + annot. for a good/bad plays, fed it into pre-training, perf would 🚀
Can one of the VCs in SF pls fund an internet company that provides good Internet and half decent service?
AT&T and Xfinity cannot be the best of what Silicon Valley has to offer. It's not venture scale; it won't make $$$, but pls do it anyway, just for for the love of the game.
What do you folks use to organize knowledge in 2025?
I consume a bunch of media - blogs, podcasts, news articles, tweets/threads etc etc.
Tell me what systems/tools work for you - anything from handwritten notes, to a personal obsidian, NotebookLLM - anything you actually use
Terminal rendering magic is very archaic and probably one of the less understood/used/loved for UI systems today - I find it a bit funny, but if not for claude code it would likely have stayed that way.
