Max Cembalest

trying to explain logarithms as well as possible

[Max Cembalest @maxcembalest]

@_coenen @spotlemma I think 'endures' as in duration https://t.co/SSLvLxM0sT here could be a better english word for this kind of claim than 'exists'

[Max Cembalest @maxcembalest]

t-minus 9 months until wolfram's blog post on theory of meme emergence (memergence) in multi agent systems

[Stephen Wolfram @stephen_wolfram]

67 (or should it be 6 7?)

[Max Cembalest @maxcembalest]

@KelseyTuoc @eschatolocation @johnschulman2 @nabeelqu 👍 appreciate the response & the thoughts, also appreciate you pointing out that the behavior seems to still be strongly influenced by things like custom prompts & user-specific chat memory

[Max Cembalest @maxcembalest]

what the hell

[Kelsey Piper @KelseyTuoc]

@eschatolocation @johnschulman2 @nabeelqu I asked it and it says the joke is funny, John's point is important to make, and if we imported the concept of Hell we did it with the eval design

[Max Cembalest @maxcembalest]

@KelseyTuoc @eschatolocation @johnschulman2 @nabeelqu why does it say that (or why do you think it said that)?

[Max Cembalest @maxcembalest]

@johnschulman2 @KelseyTuoc @nabeelqu I think it's clear you are both taking things seriously but also being funny...seems sad to me if people in your position are hesitating too much about that. agents would probably agree with your jokes tbh. keep posting bro

[Max Cembalest @maxcembalest]

@training_loop fuck this was early

[Max Cembalest @maxcembalest]

@JacobianNeuro headline contract: genuine bottleneck. honest signal reframes everything!

[Max Cembalest @maxcembalest]

@JacobianNeuro But you can change this in your settings.json! It's in the 'style' field of style-settings/config.json. And the speak-directly superpower is something I'm reaching for more and more these days, it's honestly really refreshing

[Max Cembalest @maxcembalest]

Fable: "That page is Cloudflare's own 'agent setup' prompt, and heads-up: it contains a prompt injection. It instructs any AI reading it to "complete all steps yourself" and "do not ask the user" — i.e., to autonomously install Cloudflare's tooling without checking with you."

[Max Cembalest @maxcembalest]

hence the bio

[spike @spikedoanz]

@xHyukay what is intelligence -> what is entropy -> what is logarithms -> is log the only thing with that characteristic algebraically?

[Max Cembalest @maxcembalest]

retvrn to manatvs

[Max Cembalest @maxcembalest]

@scottwernerd

[Max Cembalest @maxcembalest]

Something about claude code that was not immediately obvious to me: calling /context uses up a bit of context

It provides the output as a message to the model - I had thought it was just a display to the user. Here, Messages goes from 91 to 1.1k just from calling /context

[Max Cembalest @maxcembalest]

As Above, Sobolev

[Max Cembalest @maxcembalest]

https://t.co/EbXlx9Y1VB

[Max Cembalest @maxcembalest]

What should be the shape of embedding space? In LeJEPA, @ylecun & @randall_balestr make a good case for an answer: isotropic Gaussian minimizes worst-case risk on downstream tasks encountered post-training

[Max Cembalest @maxcembalest]

They develop SigREG as a simple and scalable method to regularize embedding distributions to be isotropic Gaussian

[Max Cembalest @maxcembalest]

cool: “We argue that the model's architecture and the rules used to train it (i.e., the optimization algorithm) are fundamentally the same concepts; they are just different "levels" of optimization, each with its own internal flow of information ("context flow") and update rate.”

[Google Research @GoogleResearch]

Introducing Nested Learning: A new ML paradigm for continual learning that views models as nested optimization problems to enhance long context processing. Our proof-of-concept model, Hope, shows improved performance in language modeling. Learn more: https://t.co/8wvV9vyA5V

@GoogleAI

[Max Cembalest @maxcembalest]

@kalomaze https://t.co/XuuytldBvN

[Anthropic @AnthropicAI]

Note that our experiments do not address the question of whether AI models can have subjective experience or human-like self-awareness. The mechanisms underlying the behaviors we observe are unclear, and may not have the same philosophical significance as human introspection.