Alex

it’s just tweaking weights to minimize loss

[Rohan Pandey @khoomeik]

been in ML research for 7 years, wrote a paper on compression & scaling laws, and passed openai's information theory interview

yet the latest 3b1b *still* gave me fresh intuition on entropy

either i'm an impostor or 3b1b is the greatest teacher of all time

[Alex @Fax3l]

@Ced_haurus @GabeLogan974 https://t.co/rrZxnph0fI

[☠ Bluetouff @bluetouff]

l0g est maintenant MCP Native
👉 https://t.co/kmltsqT2gI
et AI agents ready
👉 https://t.co/724yYf567e

[Alex @Fax3l]

@bluetouff ““Frequently the messages have meaning; that is, they refer to or are correlated according to some system with certain physical or conceptual entities. These semantic aspects of communication are irrelevant to the engineering problem.”

[Alex @Fax3l]

@bluetouff C est quand meme tordu de name drop C shannon et de proposer l opposé de ses fondamentaux…

[Alex @Fax3l]

@R_Nikaido @Kessentchu 👀😎

[Alex @Fax3l]

@anshulkundaje I like to think about it with how robots operate:
One sees a dishwasher door and picks an action based on a probability distribution learned from training. The other sees the same door and understands it needs a torque vector because it models the physical dynamics of the world

[Alex @Fax3l]

@kerighan2 @Fabien_Mikol Par contre c est un peu borderline de dire « prédire la bonne valeur d’un calcul mathématique »? On veut le calculer, pas le predire / estimer par une distribution proba ?

[Delip Rao e/σ @deliprao]

In-context learning in LLMs

[Alex @Fax3l]

@docteur_TJ C est au tel ou par une app?

[Alex @Fax3l]

@asmah2107 Return HTTP 402 to every user except 10 of them.
By the time users figure out what the status code means and wonder why the site is asking for payment, the cache has already been rebuilt.

[Alex @Fax3l]

@ThePrimeagen Use pbpaste / pbcopy.

[Alex @Fax3l]

@nickaturley Allow some chats (or some prompts within a chat) to intentionally use only a very small recent context window, instead of continuously sending the entire conversation history to the model.

[Alex @Fax3l]

@Tz_AesticV @jam3scampbell Yes and it was somehow working and fun.
the overall experience needs some tweaks, but i enjoyed it

[Alex @Fax3l]

@chrisgpt It is a bit misleading to say that “most token pairs don’t matter.”. A pair may seem meaningless at layer 2, yet is important by layer 12…

[Alex @Fax3l]

@Tracteur947 @teille2selecto @Fabien_Mikol 👀

[Alex @Fax3l]

@tunahorse21 @torumaike Ahahah good old time!!!