vessenes

I'd like to note that we should remember a formalized Lean proof does have value in that it enters the pantheon of true things other Lean proofs can rely on. Agreed that for the humans, descriptions and being able to 'grok' the proof / assess it for new tools and concepts is extremely helpful.

Pretty efficiently, apparently, since it saturated ARC-AGI-3 in half of the predicted time, and according to the Chollet blog post on the fly created dense DSLs to describe and analyze individual games.

Weave robotics (no affiliation) claims they are folding over 1,000 lbs of laundry a week in homes right now: https://www.weaverobotics.com. I have no idea how well these work.

Ooh interesting. Got a link? I guess you’d want to find out f that’s a weird motion tic falling out of rl training or the much simpler explanation. I thought figure was publicly clear the robots were operating autonomously.

I guess the VLA proposal is that frontier models will keep extending the planning and action horizon and that the action models will keep improving simultaneously leading to a moment when they suddenly get useful.

This seems likely to me, but I’m not a robot guy. We have certainly seen this 10x in value with 2-5x increase in context and horizon dynamic play out at least three times since 2019 with llms, and with self driving cars; having watched it I’m inclined to believe it can occur again.

If you tour SpaceX’s hawthorn factory they will tell you: “raw materials come in, rockets go out.” Spacex manufacturing had a mix of robots and humans manufacturing together on the floor the last time I was there, like six years ago. But it was definitely deeply vertically integrated manufacturing.

Similarly BYD in china has plants over an acre in size that run lights out with one human worker. My understanding is these are also vertically integrated, taking inspiration from Tesla and SpaceX.

I’m happy to learn more about the material world from people at any walk of life; right now we have humans building robots out of raw materials and robots building robots out of raw materials already in the world. My prediction is that humanoid robots building robots will scale rapidly in the next two years; please give me the bear case though, I’m genuinely interested.

This is a recent essay on robotic workers that does not mention Figure’s 24+ hour livestream of a robot doing package management on a live assembly line.

The real reason it’s hard to assess where we’re at with industrial robots is that some of the people who’ve gone hardest at it are secretive, and also believe in power law scaling. So, the first 1,000 robots will only be used to build the next robot factory, which also will be used to build the next really big robot factory. When we see humanoid robots actually out in the wild for sale at retail, they are going to come very very quickly, at scale.

Figure and Boston Dynamics sold out next year’s production runs almost instantly sometime last year; I think it’s just incredibly hard to overestimate the impact humanoid robots are going to have on the global economy.

thank you!

Thank you. Yes I would like to hear more about that - I’d be interested to read details about limiting some of its native tool use.

The title is the worst part of the essay. Which is super interesting, to wit: fable's a very capable model when it comes to transforming concepts in and out of structural descriptions, and you can use it (along with a test suite I presume) to transpile a codebase. I wouldn't have thought to do this, but I think it makes sense, and I like it - it's using model intelligence at a few different steps for sensible things. Thanks for the writeup. De clickbait your title though or you'll keep getting clickbait rage responses :)

It's been a busy month at OpenAI.

I'm looking forward to seeing the increased coordination and engineering skills from Astra - one of the charts shows it roughly 2-3x better in 50% of the tokens from 5.6 sol, which I find to be very capable, if still a bit 'linearly minded' when given instructions. Even in fast mode, I wish sol were quicker, so token efficiency is greatly appreciated.

Adding these cyber capabilities has let me do a bunch of low grade IT tasks around my house I've been putting off, like updating an old home assistant raspberry pi, and one way to use the cyber capacity for good is liberating (and keeping free) weird cloud hardware we have floating around the house, so I'm hoping for some nice dividends in terms of true ownership of hardware we've got.

Felix, just poking at this, and it is MUCH more pleasant to talk to, thanks to your teammates for the work.

Combo of that, laziness and load-bearing language + the penchant for making up weird dense conceptual names pushed me to sol 5.6. They seem to indicate it is a less annoying writer in the announcement so I’m curious to try it out, though.

Ironically one of their demos is speeding up inference - do us normies get to do that with Anthropic tech??

I love this. I wanted the disk noise, and I got it. I will say that’s a lot faster than my drives were in the late 1990s though.

Yes. Most of us are, still. The frontier is currently both at expanding ‘common sense’ / non-cheating outcomes for imprecisely specified software (that’s all software), and at expanding autonomy - ability to work longer unsupervised with success, oh, and also at expanding outside contextual reasoning about what’s being built, as in “hmm, that doesn’t look right or make sense, let me explore that.”

Simon’s karma exceeds yours by about 1,300x — so most of us disagree with you. Like me, I disagree with you. I like the pelican benchmarking, and I like it when Simon stays on top of models for both public good reasons and because it saves me time and effort.

That would be awesome. Probably worth getting a v1 and not upgrading the firmware just to be sure

Their current messaging is "we're busy, no promises" on dev kit or API access -- where did you see this open promise? I'm super curious.

UPDATE: That's their kickstarter customer service. Their subreddit has founders liberally promising API/devkit access, targeted at Q1 '27

This is cool. May I suggest that you make the standard deformation much less spherical? I do not get nauseous that easily and it made me a little nauseous, feeling like I was falling. I would not want to watch that while driving in a car. But 1/3 of the way from ‘flat’ to that would be super useful.

Long time go player here - this is really cool!

I only played a little, but it is clear that this game strongly prefers tight extending connections. In go terms “territory.” I am not clear if ‘influence’ is as useful a concept here; it seems very strong to make tight connections.

One of the many things that make go great is that a range of moves in relation to other stones provide a set of trade offs - more certain connection but less total influence / future territory ‘claimed’ - I’m not clear your variant has this. If so, it would be really useful to explore these move sets and give the player some basic tutorials about them. If it doesn’t, then I humbly suggest you change the rules so that it does. Perhaps tight groups can extend their ‘non placement’ range a little more, or extend their distance at which they connect up to a lone stone?

Last note, any beginner go tutorial explains that it costs more stones to live in the middle than it does the side, and that costs more than the corner. I’d be super interested to see some minimal living group definitions in the tutorial.

Again, thanks! this is really cool.

She was a founder, influential and one of the most liked people in the world. Seems fine to me.

That…is the opposite of the facts reported in that story you linked.

Counterpoint - I like Dan's site. I think it fits his vibe and priorities, and I don't find it unreadable. I can always pick my browser width and font size with a couple keyboard touches, and I never have to fight his idea of how I should want to read it.

s/Unix/The CLI/

But yes. That’s true.

This article sent me to Joyce's letters to his wife (https://www.theparisreview.org/blog/2018/02/02/james-joyces-...) which are ... dirty. Amazingly so. Inspiring use of English.

Agreed that you almost certainly can just embed the original with most modern embedding models.

well don't stop there, tell us more

Oh I love this a surprising amount. Also a good reminder of some of the dividends of vibe coding -- this would have been an extremely useful tool for at least two decades of web designers if published in the early 2000s. Even now for a type nerd who is not a designer, it was fun to play with -- really cool.

This is a tough read. I'm not sure who it's aimed at, since the article mentions only 14% of Americans (surely one of the most climate denying countries) don't believe climate change is real. It's alarmist in a way that is whistlecalling but not in a useful way.

It also lies with graphs in a way that I just hate -- we have surface temperature charts that show growth trends going all the way back to ... 2024. Please. This page from wikipedia shows a much longer range with a lot more context: https://commons.wikimedia.org/wiki/File:Holocene_Temperature..., and reasonable qualifiers on the methods used to go back.

Useful proposals always welcomed, ideally ones that in a game theoretic sense do not reward defection and are realistic about the economics of releasing more CO2 as part of power production, but this essay isn't it.

Agreed that confirming “This proposition has been faithfully translated to Lean” matters. As a side note, though, this is an area where LLMs used interactively can be helpful.

Also in the case of these recent counterexamples turning up, we have good qualitative evidence that this pathway is effective.