khalil (ψ-bio/acc)

[khalil (ψ-bio/acc) @KhalilOuard]

On to the next data collection experiment

[alexei @alexeixbt]

the fastest way to shut down neuroplasticity is labeling yourself a failure. you have to be delusional enough to look at your worst attempt and call it data collection.

[khalil (ψ-bio/acc) @KhalilOuard]

Why are Claude 5 family models so excessively and unnecessarily verbose??

[khalil (ψ-bio/acc) @KhalilOuard]

The new marker for preclinical velocity?

[Polymarket @Polymarket]

JUST IN: Lab monkey prices in China have reportedly surged to roughly $26,000 each, as the biotech boom overwhelms supply.

[Aïna @ainafahy]

what if a letter could hide inside sound?

built this yesterday: you write a letter, it becomes a hidden, poetic message made of soundwaves.

used my own voice, cloned with @GradiumAI.
img2threejs by @NickDevFE.
inspired by Jen Cantwell's "Letter Home."

[François Chollet @fchollet]

Only Art and Science raise humanity to the sublime. But the magic isn't in the output. It's in the process of creation and discovery, and the experience of feeling and understanding.

[khalil (ψ-bio/acc) @KhalilOuard]

@ThinkingBone and where are all the urameshi pfps? befriending the hxh cast feels like i'm reaching out to my second cousins

[François Chollet @fchollet]

When the cost of execution drops, the value of taste, strategy, and architectural vision skyrockets. Previously, you were spending most of your cognitive budget on the micro. Now you are free to focus on the macro.

[khalil (ψ-bio/acc) @KhalilOuard]

@yacineMTB The real shock is how much AGI dogma runs on "borrowed faith" (vibes, Bostrom, EA), when, if anything, serious conviction should be downstream of NN awe

[François Chollet @fchollet]

The best way to understand a complex system is via edge cases and failure modes, because they define the contour of the system.

[khalil (ψ-bio/acc) @KhalilOuard]

@CardilloSamuel @MistralAI Mostly capital. Raising it and how much is even available to raise in Europe. Everything else is downstream. frontier AI is just too compute-intensive, so you're cooked if you can't match the GPUs or the infra. GDPR and regs are also frictions, but it's about money.

[khalil (ψ-bio/acc) @KhalilOuard]

@techartist_ open source ?

[khalil (ψ-bio/acc) @KhalilOuard]

@Nlacombe_ If the product experience can convey the same feeling of engagement as 2000s MMORPG games, it will resonate with millennials. And I'd disagree with the angel, gamyfing overstimulation is a noble cause

[Geoffrey Négiar @geoffnegiar]

We released our first public model: t0-alpha.

t0-alpha is a foundation model for time-series forecasting. It can handle predictive variables, which is important to optimize the context for what you are predicting.

It is Apache 2.0, so check it out!
If you'd rather play with it in your browser, no setup required: checkout https://t.co/GDZkAobxtm.

Download it here -> https://t.co/Gdi0sDQIrj

[Prompter @PromptLLM]

Throwback to this insane advice from Fable 5

[Brian Armstrong @brian_armstrong]

Agree - everything in hard tech is more difficult than software (especially in bio where we barely understand how it works)

Arguably the best way to do a biotech company is:
1. make a billion dollars in software
2. self-fund the biotech (at least in part, to de-risk it)

If you don't have a software billionaire patron, the uncanny valley (series C crunch) can be brutal in bio.

It's still worth attempting hard things that can move society forward even if difficult tho. Arguably you have a responsibility to do so once you've made money in software.

[khalil (ψ-bio/acc) @KhalilOuard]

if Mistral somehow has a competitive frontier model in the vault (they don't), this is a great case of using meme culture for viral marketing

[saperlipopette @__tommyFR__]

@arthurmensch 🇫🇷🇫🇷🇫🇷🇫🇷

[khalil (ψ-bio/acc) @KhalilOuard]

@bravo_abad Scaling laws were initially derived as a power law scaling of the loss function with model/dataset size. Improvement of the model's performance on downstream tasks is a byproduct

Do they observe a similar plateau on training objectives wrt size ?

[khalil (ψ-bio/acc) @KhalilOuard]

Clarity is found when looking past the self

[owl @owl_posting]

in order to best utilize ai, you need to skate on the edge of ai psychosis; the only thing saving you from falling into the soup being a mild distaste for the whole economic apparatus that forces you to talk to such an alien mind. the second you garner too much appreciation, you’ll be swallowed whole. of course, this ‘skating’ is the ideal, which is very difficult to maintain. in practice, you’ll spend some days dunked in the soup, headaches and indigestion as your nervous system is overloaded from reading several books-worth of semi-meaningless text. other days, you’ll climb out, dripping, a renewed commitment to love your humanness. then it repeats