ftxbro

53 ms·

631.
by ftxbro 3y ago
When I got to slide 5 I LMAO when they wrote "LLVM doesn’t remove the dead loop" in bright red it reminded me so much of the geohot drama on his batshit twitch archive on youtube where some AMD GPU driver author says something abo
632.
by ftxbro 3y ago
Anyone who doesn't know, this is by Chris Lattner the famous hacker who made projects like LLVM, Clang, Swift. https://en.wikipedia.org/wiki/Chris_Lattner He is writing this as part of a marketing blitz for his ne
633.
by ftxbro 3y ago
No. This tokenization is huffman encoding and word2vec is embedding in cartesian space. They are very different.
634.
by ftxbro 3y ago
I find it difficult to believe such a strong position. For example consider some hypothetical alternate timeline like if there were some physicist connectionist hackers who made powerful supercomputers with LLM capabilities back when we onl
635.
by ftxbro 3y ago
Artificial intelligence with superhuman cognitive capabilities has never been made before, so of course there will be a whole constellation of new words and terminology and connotations to go along with it. There is also a corporate rush to
636.
by ftxbro 3y ago
Yes the title is sensationalist. The real question in the body of the article is "What's the ideal programming language for LLMs to output?" I thought that question was interesting and I'd already posted a 'ask hac
637.
by ftxbro 3y ago
Interesting. Good to know that I'm thinking in similar ways to Chris Lattner. Or I suppose it's possible that his blog post is only pandering to dumb ones like me who he knows will ask this dumb question so he's pre-emptively
638.
by ftxbro 3y ago
this has 'microsoft word is racist bc it let me type the n word' energy
639.
by ftxbro 3y ago
> "Aksoy's businesses collected at least $100 million, much of which he pocketed." > "Under the terms of the plea deal, he is required to forfeit $15 million in illicit gains"
640.
by ftxbro 3y ago
interesting take
641.
by ftxbro 3y ago
i don't know but it won't be available for public access. it's like asking what 2030 nuclear weapon control and simulation codes will look like.
642.
by ftxbro 3y ago
this some cyberpunk shit
643.
by ftxbro 3y ago
good point. also, if there was only one of each then there might have been a difference between them other than how they looked like.
644.
by ftxbro 3y ago
> "Nerds are usually ugly" maybe nerds from 80s movies
645.
by ftxbro 3y ago
> "“For example, whether the game is biased toward one player, whether it’s too short or too long, too simple or too complex. These indicators can tell us which rulesets were more likely to have been the games that people enjoyed pl
646.
by ftxbro 3y ago
> "Iraq adventure was a pointless slurry" I wonder where 'pointless slurry' is meant on the scale of justifications or rationalizations of war? Maybe more favorably than 'quagmire' but less than 'good s
647.
by ftxbro 3y ago
I feel like the design of programming languages has involved some tradeoffs that have been informed by biological human psychology. I'm wondering if different tradeoffs would be more beneficial to AIs that are based on LLMs. Obviously
648.
by ftxbro 3y ago
yes it's obvious that the LLM will do better with existing languages that have large corpuses than with hypothetical languages that don't exist
649.
by ftxbro 3y ago
> "the halo error (a cognitive bias whereby our perception of a person is shaped by our opinions of other traits) as applied to risk tolerance, which it proven to be vulnerable to stress and concupiscence." so i know a lot of w
650.
by ftxbro 3y ago
> "We’ve not seen the paper yet so will have to go by the press office’s description, which outlines an experiment where 111 male early-stage investors are shown video of the same start-up business idea presented by a conventionally
651.
by ftxbro 3y ago
now do one for single drop at home blood testing
652.
by ftxbro 3y ago
> "Something in the Apple omertà makes me uncomfortable naming my collaborators as I normally would, even as I discuss the project itself." I don't understand this. It's things they made them sign? I saw another artic
653.
by ftxbro 3y ago
> "I've noticed something else in the last 24 hours, as initial hands-on impressions have been published by various outlets (ours is here). There are, as far as I can find, no actual photos or videos of any writer or YouTuber w
654.
by ftxbro 3y ago
i think they are the most visible ones. it's not like you are going to see what a database programmer working at oracle does at their job.
655.
Design a programming language for LLMs to learn?
3 points by ftxbro 3y ago|8 comments
656.
by ftxbro 3y ago
I've been watching these and so far I think midjourney 5 is still best for plain text to image. If you want highly specific models (like one that only makes trains or whatever) then I think stablediffusion is better. For image-to-image
657.
by ftxbro 3y ago
just a heads up this post is likely going to the creepy backrooms of algolia https://news.ycombinator.com/highlights
658.
by ftxbro 3y ago
it sound like stackexchange getting disrupted by llm
659.
by ftxbro 3y ago
> You're eliminating all the nuance. OK I'll put some nuance. My nuanced view is that many of the real AI boffins like Ilya Sutskever know that the 'stack more layers' LLM approach to AI is going to unlock superhuman
660.
by ftxbro 3y ago
Just imagine any time that a dangerous thing is on the horizon. There will literally always be people pointing out that it's possible that the ones saying it's scary are saying it for their own gain. That's all this article i

More ›