offensive bias

why tell a dead man the future

[offensive bias @ffensivebias]

none of the introspection matters if you win

[offensive bias @ffensivebias]

dags/workflow graphs are the tool shaped object final boss… why would you look at a graph when you can ask a model, there is (nearly) by definition no more complete representation of your graph outside latent space, even formal approaches like gql failed miserably

[offensive bias @ffensivebias]

@awilkinson @sama Would be interesting if it’s a UX performance issue downstream of benefitting from substantially more reasoning tokens

[offensive bias @ffensivebias]

@dwarkesh_sp nearly as retarded as saying 1 car is worth 300 horses

[offensive bias @ffensivebias]

@varunram Same experience, these guys are metal

[offensive bias @ffensivebias]

yeah I’m watching the odyssey the way I goddamn want to

[offensive bias @ffensivebias]

https://t.co/5deTOJhmZS

[Zhengyao Jiang @zhengyaojiang]

Our RSI system AIDE² has two autoresearch loops.

An inner loop, just like a normal autoresearch agent, optimizing code against an eval.

An outer loop, optimizing the inner-loop agent's harness code against the inner loop's average score across different benchmarks. (2/7)

[offensive bias @ffensivebias]

given that agi is more or less already here I wonder if our benchmarks objectively have to move a derivative up toward some idea of how effectively models can improve themselves (on current benchmarks? unclear)

[offensive bias @ffensivebias]

@IterIntellectus embed a glock in it and strap this on an amputee immediately

[offensive bias @ffensivebias]

they tried this on me it was all the shrek movies except 3

[Amir Zamir @zamir_ar]

Turns out it's possible to generate videos that maximally excite an arbitrary brain region using a simple search-based algorithm. It's a fully computational approach, so it's another way to speculate what a brain region represents, alongside other neuroscientific methods.

Select an arbitrary brain region->algorithmically generate a video that jacks it up. See the visuals on the webpage https://t.co/UzKyN8MHda

In silico (for now)

[offensive bias @ffensivebias]

Sol post training Luna is notable here

[offensive bias @ffensivebias]

it’s a fairly bearish indicator for asi that mythos or whatever is still being evaluated on terminalbench and that there is no clear goalpost on self improvement at all outside a vague notion that eng is moving faster inside labs

[offensive bias @ffensivebias]

must be a fun day to hold an open ai bag

[tobi lutke @tobi]

I shipped more code in the last 3 weeks than the decade before. The top AI models / agentic systems right now are an entirely different thing to what people used until the beginning of December.

[Bryan Cheong @bryancsk]

Aging is such a cruel curse, you notice it in everything except yourself until it is too late.

[offensive bias @ffensivebias]

I have never seen such ladder kicking of the spirit

[offensive bias @ffensivebias]

re post ai disillusionment in software

it’s curious how so many have come out of the woodwork now with the sole, certain message that every normative wealth generation opportunity in your time is clearly gone and away, RE, crypto, swe, etc

[offensive bias @ffensivebias]

a weirdly subtle and specific character trait I will continue to fail to codify well is folks’ interest and willingness to shape the world in their image. I think for a few people I know hard drugs actually developed this

[ftlsid @ftlsid]

there comes a point when only total commitment, with all of the risk that it entails, will allow you to proceed

[offensive bias @ffensivebias]

Brb using my tariff check to make my monthly 50 yr

[offensive bias @ffensivebias]

surely this is resolved by giving alexandr wang another 20 billion dollars

[near @nearcyan]

Meta AI's live demo failed for the entire minute 😢