Fable got me try out Claude Code again and I'm shocked at how much worse Claude Code is than Codex. Anthropic might be cooked now that GPT 5.6 seems to be taking the performance flag and people are gonna see what it's like to use Codex.
@ErikVoorhees@sethmobr@bourscheid Who paved the roads they drive on, who, who guards their blocks, who defends their ability to do free trade and speak their minds? I'm not against the rich but obviously they are under taxed. Who cares about the absolute amount? It's about stabilizing society.
@mitchellh Is the denounce part really necessary? I would imagine the main goal is to block low effort "contributors", who will circumvent by creating new accounts anyway. If you only do vouching, then you could bootstrap by auto-vouching contributors to well run open source projects.
@lexfridman The SOTA in AI the past year moved from being scientific to being mostly engineering. If you're interviewing an AI researcher, I'm curious how they view that development, and if you're interviewing an AI engineer, I'm curious what scientific breakthrough might disrupt them
@headinthebox Same reason scripting languages did in the early 2000's! But imagine how much better these agents can still get just by engineering not even through LLM improvements. If we never improved past GPT3.5 we'd still have amazing coding agents today.
@simonw This is Europe so excuse the small amounts, but in a previous company we spent ~$60k on two projects to finetune segmentation models, and then Meta released SAM and it was better at both. It'll be the 2nd AI winter before I'll waste money on finetuning again..
@wintergatan Not to double down on the sycophancy but to me your videos capture an artist struggling to create beauty more than a van Gogh painting does.
@wintergatan Not sure if you're reminded often of this, but the video where you once again take the angle grinder to the marble machine a couple years back is the most beautiful YouTube video I have ever seen. You are making art, in your music, in your machines, in your videos, in your life.
@pli_cachete Rocket engines. If you don't do it well the first time your rocket is gonna blow up and it'll cost you hundreds of millions. Better put a lot of sensors on those puppies.
.@windsurf_ai just to let you know, both yesterday and today on separate projects I had to ask cascade to teach itself how to properly use the `view_file` tool
@ylecun Right, except it's not about error but about ability. If there is a 10% chance the AI is unable to do a thing, its chance of success is low after ten hours. But for errors it could self correct and make the chance of error per step arbitrarily low (limited by its ability)
@signulll They don't risk the base by backing Elon. Their base isn't Trump supporters, it had to be seduced into tolerating Trump by Chamath arguing he isn't so bad once you know him. I bet they got into a heated and unresolved argument and it made the pod unreleasable.
It's not true though. Peaceful inhabitants of Jerusalem are unlawfully evicted and replaced by Israelis. Peaceful farmers on the west bank are displaced by violent settlers. This video is blaming the boiling frog for jumping. If Hamas stopped fighting maybe the frog would boil and less people would die, but it wouldn't be peace.
@BillAckman@realDonaldTrump If he doesn't strike at least one good deal in the next couple days the whole world economy will grind to a halt. You forget the other countries have markets that are collapsing as well, I don't think the negotiation position is as bad as you say.