Hearing reports of startups paying for what they thought was human data but turns out it’s mostly AI-generated. Usually it’s not the vendors intentionally tricking customers, but smart students getting paid per task will find ways to automate annotations, pairwise comparisons,

I wonder if Michael Jordan and Tom Cruise know that their names are default examples in NLP papers.

Soon we’ll get Claude Code for gene editing!!

It's revealing that when we first made a chatbot, we wanted it to be our therapist: ELIZA (1966) Its inventor Weizenbaum became disillusioned and left the field. 60 years later, AI-induced psychosis is trending again. Human vulnerabilities are never patched.

SILICON VALLEY IS IN A CRISIS Our best minds are being wasted on marginal AI wrappers & safe bets. To combat that, we're announcing Choose Good Quests by savant a conference to gather 500 exceptional talents to take on hard & important problems. November 6-7 in SF 👇👇

Pro athlete: "Just wanna say God has a plan for everything. All praise to the man upstairs for giving us this dub tonight!" ML researcher: "We offer no explanation as to why these architectures seem to work; we attribute their success, as all else, to divine benevolence."

It's 2025. Developers now do all their work in the 2 weeks following the launch of a new coding agent, when tokens are heavily subsidised. Agent gets nerfed. Wait patiently for the next launch.

the real agi benchmark should be "based on this graph, list the days I missed" why are vlms still so bad at this 🤷‍♂️

one of the best years for sports in recent memory

early on ppl thought / still think vibe coding platforms are thin wrappers over models... it's becoming clear that they're actually the eval engines that lead to better models meaning their value is roughly what labs are willing to pay for human creativity & taste

The internet makes the world flat. My parents learned English with textbooks & flashcards, so while they can read & write at a graduate level, they speak English with an accent. Language learning has flipped. Gen Z around the world now talk in an American Twitch-streamer

products that have investors as early adopters: - ride-hailing - email apps - meeting prep / notes - dictation - virtual assistants - people search engine

emergent strategies are pretty neat

We just need a systems view of everything.

Is it weird that I trust the AI's answer more if it thought for longer? I must be in the <1% of users still manually selecting o3 vs. default router model?

Nano Banana (brilliant researchers, could use better PMM) adding colour to B&W photographs 😮

The best eng didn't go to Stanford/MIT. They probably went somewhere with "Polytechnic" in the name.

The world trends toward complexity. Startups constrain that complexity, until they become successful & complex themselves. AWS simplified servers. Now there are startups that simplify AWS. Stripe simplified payments. Now startups seek to simplify Stripe.