clems

il m’arrive de tweeter avec le hashtag #vendredilecture (c’est faux vive travis scott)

[clems @ephemerde]

🙏

[clems @ephemerde]

@_poolday_ mega cool! can’t you suggest pools created automatically from matching screenshots content? i tend to take a lot of X posts screenshots as ideas to explore / things to keep an eye on. would be nice to have them grouped together so they become actionable

[clems @ephemerde]

crazy how things shift. a month ago i was about to cancel my claude code sub because the token limits were outrageous compared to codex. since they augmented capacity, not only do i get to do more, but it now feels codex limits burn faster (no tested though, just a feeling)

[clems @ephemerde]

spotify should add comments. with Fable 5 out, some cracked spotify engineer could vibecode it by friday. tell your PM it’s “community engagement” and “competitive retention” and enjoy the promo

[clems @ephemerde]

a-t-on déjà vu pire qu’un contrôleur de train ouigo? bonne question

[clems @ephemerde]

@jpschroeder obvious: they are not allowed to use Mythos because it’s too powerful

[clems @ephemerde]

@modal a solution to the exact problem i ran into, will give it a try!!

[clems @ephemerde]

andrej just made overfitting the validation set attractive

[Andrej Karpathy @karpathy]

I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then:

- the human iterates on the prompt (.md)
- the AI agent iterates on the training code (.py)

The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc.

https://t.co/YCvOwwjOzF
Part code, part sci-fi, and a pinch of psychosis :)

[clems @ephemerde]

i’m burning my budget quickly though, gotta be cautious!

[clems @ephemerde]

only four days into February, and I can already tell my claude code subscription is the best €20 i’ll spend this month

[clems @ephemerde]

it’s not magic, it makes mistakes, it needs nurturing. but with care, the potential is enormous. for the first time, I’m giving life to ideas i never had the time to prototype, and the psychological relief of that alone is priceles

[clems @ephemerde]

it’s not magic, it makes mistakes, it needs nurturing. but with care, the potential is enormous. for the first time, I’m giving life to ideas i never had the time to prototype, and the psychological relief of that alone is priceless

[clems @ephemerde]

rechute c t dingue

[clems @ephemerde]

@AstroMagazine_ c’est chaud c’est chaud

[clems @ephemerde]

oui je viens de replay chienne de trappe life y a quoi

[clems @ephemerde]

24 de fève quel projet les mecs ptn

[clems @ephemerde]

24 de fève quel projet les mecs ptn

[clems @ephemerde]

@alexisharold14 @AstroMagazine_ j’attends depuisss 🤝

[clems @ephemerde]

oh le but de fou mdrr #OLTFC

[clems @ephemerde]

@mryya17 jamais goûtée mais je refuse de croire qu’elle est meilleure que l’incroyable mousse de michel & augustin

[clems @ephemerde]

très fort

[clems @ephemerde]

manu koné c’est fort

[clems @ephemerde]

pythagore sous hypotésnus