Aller au contenu
debya
Français
Se connecter Inscription gratuite
Voir sur Bluesky Suivre ce compte

Fran Litterio

@fpl9000.bsky.social did:plc:j3d3ovtl7hs7sdjmvpezs6ay

Retired software engineer. Deadhead. AI enthusiast. Long ago, I implemented Bash's regex operator (=~). Signal ID: franl.99.

Sur Bluesky depuis le

Abonnés
2 904
Abonnements
344
Posts
789
Jours dans les classements
0

Résumé

Fran Litterio (@fpl9000.bsky.social) est sur Bluesky depuis le 15 février 2024. Le compte a 2 904 abonnés, suit 344 comptes et a publié 789 posts.

Présence dans les classements

Les meilleurs jours où les posts de ce compte sont entrés dans le top quotidien de Debya.

Ce compte n’est encore apparu dans aucun classement quotidien.

Posts aimés récemment

J’aime publics du réseau Bluesky.

  • Ethan Mollick@emollick.bsky.social

    Here is a mass attempted replication of old findings of LLMs with new models. Much conventional wisdom is outdated. Recent models no longer invent citations and both hallucination rates & stereotyping has dropped. But they are still sycophantic & lean left. zou-group.github.io/llm-claims-s...

    Voir le post
  • Singularity's Bounty e/CC@catblanketflower.yuwakisa.com

    If so, the models can write small universal function approximators Will be fascinating to map this jagged frontier

    Voir le post
  • Andrew Cullen@adrcullen.bsky.social

    Sure, but the required asterisk is huge 1) Qwen's loops were well known and understood already 2) It didn't improve math capacity per se, it removed the loops that cause Qwen to hit its token limit before returning a response. 3) For a model that doesn't loop it couldn't improve math performance.

    Voir le post
  • Ethan Mollick@emollick.bsky.social

    If this holds up, it is big: "We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training." rikiyat.github.io/blog/weight-...

    Voir le post
  • Sam Harsimony@harsimony.bsky.social

    After reading through this, I think experiments 1, 2, 4, 6, 8 might be too simple/easy. While experiment 3&5 are crazy. (and it failed expt. 7 and the Additional Experiment). I worry the "no tool use" is confounded by Astra maybe having a CPU in its weights. See another confounder below ...

    Voir le post

Connectez-vous pour en voir plus

Comptes bloqués

Sur Bluesky, les blocages sont publics.