j2p2 🟧

🤖 You won't believe these 100 ways GPT-4 can make your life easier! Say hello to your new AI best friend! 👋 A thread: #GPT4 #AILifeHack

@zhengyaojiang here is autoresearch for llm architecture x.com/_jiga/status/2…

Autoresearch has been out for 2 weeks. The community is trying to apply it to everything with a measurable metric, here are some successful attempts: 🧵 (1/6)

@gregisenberg No GPU needed. I just ran 30 experiments on a CLI tool's lint rules using Gemini Flash. $0.15, 12 minutes, +14% quality improvement — shipped to GitHub automatically. The pattern works on anything measurable. Here's the full breakdown 🧵 x.com/_jiga/status/2…

karpathy just broke the internet with something called auto research it’s basically an ai research agent that runs experiments for you 24/7 you give it a goal like “make this model better” “find a higher converting landing page” “lower customer acquisition cost” then it runs a loop: 1) plan an experiment 2) edit the code or config 3) run a short test on a gpu 4) read the metrics 5) keep the winner 6) try again over and over while you sleep by the morning you wake up to the best version actual tested improvements think of it like a robot research intern that runs hundreds of experiments and only keeps the winners this is link to his repo github.com/karpathy/autor… for your to mess around with it in the latest episode of @startupideaspod i break down: • what auto research actually is • how it works step by step • 10 business ideas you can build with it • how to install it and start using it this one is saucy because tools like this change how startups get built watch

@karpathy ctxm is open source: github.com/jiga/ctxm Autoresearch harness included. Swap in your eval. 30 experiments. $0.15. 12 min. What system are you pointing this at?

@karpathy What got shipped: - too_generic_threshold: 100 → 60 chars - llm_phrase_threshold: 2 → 3 matches - antipattern_min_strength: 1 → 2 matches Committed and pushed while I was in morning standup. Overall F1: 0.8333 → 0.9500. Precision and recall both 1.0.

I let an AI run 30 experiments on my CLI tool overnight. It found 2 hidden bugs, fixed them, and shipped a better version to GitHub while I slept. Cost: $0.15 Time: 12 minutes F1 score: 0.83 → 0.95 Here's exactly how it worked 🧵

@karpathy ctxm is open source: github.com/jiga/ctxm Auto research harness included. Swap in your eval. 30 experiments. $0.15. 12 min. What system are you pointing this at?

@karpathy we are entering the Age of Agents

Human orgs are not legible, the CEO can’t see/feel/zoom in on any activity in their company, with real time stats etc. I have no doubt that it will be possible to control orgs on mobile, with voice etc., but with this level of legibility will that be optimal? Not in principle and asymptotically but in practice and for at least the next round of play.

Expectation: the age of the IDE is over Reality: we’re going to need a bigger IDE (imo). It just looks very different because humans now move upwards and program at a higher level - the basic unit of interest is not one file but one agent. It’s still programming.

@thibaultroux_ @karpathy it was simple enough- “explain this code using interactive visual infographics. be through and creative so anyone can understand”

@_jiga @karpathy Does it make that one-shot ? Do you mind sharing the prompt ? 👀

New art project. Train and inference GPT in 243 lines of pure, dependency-free Python. This is the *full* algorithmic content of what is needed. Everything else is just for efficiency. I cannot simplify this any further. gist.github.com/karpathy/8627f…

@LNow_ @obycode @v1n1_c0d35 @alexlmiller @mrwolf_btc @Stacks @muthusleuths may have some data

@obycode @v1n1_c0d35 @alexlmiller @mrwolf_btc @Stacks Tried out... It doesn't look like something that is CPU, network, or disk io bound. It looks more like a sequential process: download block->validate->apply->download block->validate->apply.

Do you run a @Stacks node? Curious to see how decentralized we really are 👀

@LNow_ @AdrianoDiLuzio same contract name with automatic versioning

@AdrianoDiLuzio I was wrong with my initial assessment of this hack - I didn't understand fully how ALEX contract was meant to work. But now I can confirm it was Stacks limitation that theoretically can be mitigated, but in practice I don't think it is feasible.

I'm glad to see quick action by the ALEX team to resolve the situation for users, but want to be very clear: it is not a Stacks limitation or vulnerability that caused this exploit. Having spoken to multiple experienced, independent Stacks developers, the exploit was likely caused by a lack of verification in the ALEX smart contracts as well as the use of a single contract instead of segregated pools as other apps in the ecosystem use. This is why other DeFi apps in the Stacks ecosystem were not exploited despite offering similar functionality. Again, I'm pleased to see how quickly ALEX is committing to resolving this for their users and taking lessons from this to help all builders build more secure apps, but Stacks users can remain confident in the security and design of the L1.

Some independent app builders in the Stacks ecosystem were recently targeted via Github in an effort to insert malicious code in their app repositories. We have found no evidence of an attack or compromise of the sBTC or Stacks Core repositories, but because some of these open-source devs also contributed separately to other ecosystem projects (including sBTC), we are taking the opportunity to review security practices around sBTC and Stacks Core and provide more guidance to contributors about their personal and company setups. Core development (including sBTC) already has extensive security practices in place including least-privilege permissions, branch protections, multi-man rules, MFA, code audits, and more. But as the Stacks ecosystem and sBTC usage grows, attempts on our software development process will become more common. We’re prepared for this increased scrutiny on our Github repositories and continue to strengthen our security practices. Stacks core devs are working with internal and external security experts to further tighten our software development process. We’ll publish a more detailed analysis of the attempted attack and security measures that helped catch it, along with information on upgraded security processes to help prevent future attacks. There are no user-facing changes at this time and no steps that users need to take - sBTC and the Stacks network continue to function as normal. Please note that as part of the sBTC development plan, it has filled its 3,000 BTC cap, and withdrawals are not enabled, so it is not currently possible to peg in or peg out of sBTC. You can always report anything suspicious or find more information on security at stacks.org/security.

@sama release an open source version? or distributed inference node network?

insane thing: we are currently losing money on openai pro subscriptions! people use it much more than we expected.