←
valy545 posts
valy avatar

valy

@faychuk

Threads about tech | (҂◡̀_◡́)ᕤ

◷ Joined July 2022▧ 76 Photos and videos
valy avatar
valy@faychuk·Apr 28

I have lost $250k over the past 2 years, follow for more financial advice

Post attachment
valy avatar
valy@faychuk·Oct 8

if your inbox looks like you have money, your AI agent will suggest you more expensive options everywhere

valy avatar
valy@faychuk·Oct 8

a dolly parton fan asked chatgpt for a new yorker style cartoon and got one signed "BLOPER" - the faked the real signature of cartoonist brendan loper, and people actually believed it was legit

Post attachment
Andrej Karpathy avatar
⟳ valy retweeted
Andrej Karpathy@karpathy·Oct 2

We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks: Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better: Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better: Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better: Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work! In summary: - As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding. - Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.

Post attachment
valy avatar
valy@faychuk·Sep 20

three researchers landed a PR into the internal OpenAI's github repo: they have leveraged a libheif heap overflow on openai's discourse forum due to the outdated docker image, then "sign in with openai" SSO from there into an employee's codex, and finally submitted a PR as a proof total spend was under $3,000 in tokens, 72 hours, and the won is $6,500 bounty

Post attachment
valy avatar
valy@faychuk·Sep 17

hugging face's bill for the openai agent that broke into its infra - $100M in compute for the community plus the full execution traces, invoice denominated in GPUs this time

Post attachment
valy avatar
valy@faychuk·Sep 17

an insane raspberry pi 5 e-ink frame that listens for birds and draws them as 1800s illustrations - birdnet, 800 hand-cut engravings covering 400 species, sized by body mass, fully local

Post attachment
valy avatar
valy@faychuk·Sep 17

guy ported his 1993 amiga game to godot over a july 4th weekend - claude read 72,758 lines of 68000 assembly, rebuilt lost source, reassembled the shipped binary byte-identical, decoded all five level maps

Post attachment
valy avatar
valy@faychuk·Sep 16

iOS 27 release code has a full inference provider protocol - claude or gpt can receive siri's own planner prompt and tool definitions and run the phone, apple built the swap before announcing it

Post attachment
valy avatar
valy@faychuk·Sep 5

apple ran forensics on an ex-engineer's macbook and found a confidential schematic, an LTspice run in march, icloud syncing the file to a mac mini, instructions to destroy evidence - your laptop remembers everything

Post attachment
valy avatar
valy@faychuk·Sep 5

a from-scratch 8-layer transformer, 1.5 hours on one 5090, 67 cents total - 44% on ARC-1, matching TRM at $158, sample efficiency is the whole game now

Post attachment
Solana Sensei avatar
⟳ valy retweeted
Solana Sensei@SolanaSensei·Sep 2

me going to work so i can afford to buy more crypto

0:10