techno yoda

sf exploration 📸 (forever expanding)

the ending of the book was the best of all of them. Like woooow

chapter 92 of parade of horribles was absolutely fabulous. DCC book series is so damn good. Thank you @mattdinniman for such a beautiful gift to this world.

snow crash is a much better term for AI psychosis or someone getting one shotted by a clanker

@__paleologo the pope got snow crashed !

Dario Amodei gave a free Mythos sub to Leo and within five minutes Leo lost his faith due to a vulnerability.

@GergelyOrosz if you don’t eat the code you write then you can’t build meaningfully long lasting things. software longevity requires each line to be drastically intentional. A code base without diligently managed history is no different then AI slop.

That feeling of: "I'm in the middle of the code... oh, this is such a nasty hack. OK, let me clean it up as I go. [2 hours pass] OK, it's done, now let me get back to where I was." It just never happens as organically as I use AI agents. I no longer spot stuff as I don't "live in" the code...

@FangYi11101 @fleetingbits any metric derived purely from revenue is a perverse metric.

@fleetingbits Tbf both of these are fake

Sobering to learn that a prominent AI commentator can’t tell operating margin from gross margin. Be careful where you get your info.

@mon__lim most people I have met across continents in my life don’t possess the acumen to ask different questions when they meet someone new. They always ask the same thing and the convo becomes regurgitative. for me this question is absolute taboo. it’s where creativity dies.

mfw ppl at SFO use “so what do you do?” as icebreakers lol. going straight in for the kill

@DavidSHolz the right question is not cluster size. It’s training duration. Can run a training job for 10 mins on 1024 nodes but exponentially harder when doing it for 3 weeks straight

how many gpus do you think a single researcher can handle at once for a single big training job without any help? (assume it's set up with slurm or something)

@AvitalBalwit @TheFP real gods move atoms and alter local physics.

New essay for @TheFP on searching for God in Silicon Valley. "They are building something that has brought them, unexpectedly, to the edge of where He would be."

@mahlenr search is everywhere to see. Even in modern day AI technoyoda.github.io/agent-search.h…

One time I asked a friend who was working on an AI search platform why he found search so interesting that he wanted to focus specifically on it. He said it’s not that he found it especially interesting, but that everywhere he looked the problems eventually reduced to search problems. He had a belief that everything in Life, at every layer of complexity, was ultimately a search problem. If there is something you want to achieve, there exists a series of actions that will get you there. Search is the process of finding it. Google became arguably the most important company in the world by being the first to make search work really well at the layer of human knowledge. There are so many layers of reality still waiting for their search solution.

@leothecurious hardwarelottery.github.io

now this is cool research

@PhiloGroves define the terms please

I think we will have AGI within 5-10 years, but also confident ASI will take far longer. We are missing a fundamental piece of the puzzle. A distilled model is never better than the original; with frontier models, humanity is the original model.

@gleech given how the principal agent problem gets exacerbated by agents, general autonomy is very hard to build. narrow autonomy is possible but if you really care about autonomy then any thing that has a clanker in the middle eventually transforms to deterministic code

are you using your own AI agents for anything besides search/lit review, coding, chat

@ayushjaiswal ARR is nothing but a vanity metric. DB gives liquidity to employees. DB is a healthy business with strong enterprise sales and profitable margins. Cursor is none of that. growth != healthy business

Cursor is at $3B ARR, Databricks at 5B. Databricks is 13 years old, cursor is ~4 years in. Wild.

@ChombaBupe wen u keep using anthropocentric language, you are cursed to not see what few can. change you language and you will be able to see a different future. the nature of a thing is more important than its form

Brah, I am a telecommunications engineer by profession, antenna design is a way more sophisticated open-ended problem than even the mathematical problem OpenAI model solved.

@IceSolst the only balanced take ever is “reduce your dependencies as often as you can”. be more vigilant with deletions and conservative about additions. written code should be about longetivity and “building it once thoughtfully”. you cant build towers to the moon on fluid foundations

“Rewrite your dependencies” is such a stupid take Help me secure my supply chain “yeha bro jus t uhh fork all ur dependencies”

@alpercanbe it’s very simple : robots vs wrestlers

i wonder why they need this much space

click farms and monetization spirals. if spewing hate gives you monies why would ignorant people work hard and spend brain cells thinking. only thing 🇺🇸 listens to is $$$. throw enough, you can buy an president or policy. there is tons of prior art on this. reddit.com/r/todayilearne…

@raunakdoesdev i didnt realize what a nice bubble we lived in on X and what a cesspool it was outside

the replies to this are beyond unhinged

@theo @GaryMarcus be better. x.com/GaryMarcus/sta…

@GaryMarcus You know you can mention this without tagging him right? Purely performative to do such. > "and LEARN something." This type of non-reply isn't on the pyramid, but it should be. It's more pathetic than anything else listed.

Fuck this OpenAI employee, seriously fuck him. Also read this quantitative study, which I had nothing to do with, that says my technical predictions have bee largely correct. github.com/davegoldblatt/… I should not have to put up with this level of unhinged hostility, simply for defending views that have frequently been challenging to his company. Nor should I have to put up with him neither giving evidence nor allowing me to respond. This man appears to be both a coward and a liar, with only insults, literally a bottom feeder in @Paulg’s pyramid of argumentation. That this empty slander is all he has got is perhaps a sign of how desperate OpenAI has become.

@KevinNaughtonJr now multiply each side by N where N > 1 million

this engineering shit really isn't that complicated