Self

builder. exploring the void.

On a quest for the sublime and eternal

[Self @SelfInfinity]

@rh__147 @AndrewYNg @syncthing For that I use the obsidian app and the official sync service on iOS

[Self @SelfInfinity]

@EMostaque @Kimi_Moonshot this is why China will indeed win the ai race.

EU ngmi

[Self @SelfInfinity]

@rh__147 @AndrewYNg I pay for the service they offer, but I also use @syncthing (opensource) to sync the obsidian vault between devices (eg between macOS and Linux)

[Self @SelfInfinity]

@kandros5591 @AndrewYNg good point, it’s not open source

[Self @SelfInfinity]

@soumithchintala now join @karpathy the greats do their own thing

[Self @SelfInfinity]

@FU_joehudson Fully feeling and welcoming that fear is a loving and embodied act.

those charts look like if someone wanted to Fourier transform fear into Nirvana. I’d call this bypassing by intellectualization.

[Self @SelfInfinity]

@VikParuchuri nice is this a follow up to Surya or something else completely?

[Self @SelfInfinity]

@hardmaru @yule_gan Reminds me of your old blog posts on cma es , cool stuff https://t.co/IyYxRxL6AN

[Self @SelfInfinity]

@coffeebreak_YT humanity will be drowned in endless slop, creativity censored by big companies, running on rails

[Self @SelfInfinity]

I wonder if LLMs actually are indeed a local optimum, that instead of leading to progress, stalls human progress, maybe for a long time, because we did bet on the wrong horse.

So we essentially start with a local optimum to initialize the LLM with pre-training, with a massive human bias, which itself could also be a wrong local minimum, of course. Reinforcement learning itself, obviously, is way too brute force

[Self @SelfInfinity]

@karpathy @simonw indeed they switch coding clis more often than their underwear. if tomorrow’s one is better than today’s, they’ll switch in an heartbeat and scream "the old one is so terrible" despite being scifi good just weeks ago

[François Chollet @fchollet]

The 3rd edition of my book Deep Learning with Python is being printed right now, and will be in bookstores within 2 weeks. You can order it now from Amazon or from Manning.

This time, we're also releasing the whole thing as a 100% free website.

I don't care if it reduces book sales, I think it's the best deep learning intro around, and more people should be able to read it.

[Self @SelfInfinity]

@markmanson hm that does disqualify most of twitter 🤔

[Self @SelfInfinity]

@NielsRogge have to agree, also the llama 2 releases back then were nice as well. while models may have become better since then, those first smallish opensource LLMs were the pioneers

[Self @SelfInfinity]

I have found that writing unit tests, integration tests, for instance with PyTest or vitest , are among the highest ROI activities to counteract slop and the increase in entropy in the codebase, a.k.a. more garbage created by the AI agents. Also, regular code reviews by bigger reasoning models, independent of the CLI agents (cc, codex cli, pick your cli of the week), is also worthwhile.

[Self @SelfInfinity]

@samuelcolvin @colinhacks Zod and Pydantic are among my fav libraries in TS/Python

[Self @SelfInfinity]

@karpathy This space is moving so fast, it's incredible. Anyone still remembers Cursor? Or Windsurf? Do they still exist? last week was Claude Code. Now it's Codex. Next week, what will it be?

[Self @SelfInfinity]

@FU_joehudson From you I’ve learned the term "welcome".

Not as a subtle strategy to get rid of it but to truly love it, unconditionally.

Reminds me of Rumi's guesthouse ❤️

[Self @SelfInfinity]

@OpenAI codex cli leads currently, claude code is close behind.

The world has forgotten about the cursors and windsurfs.

the space is moving fast