That would be my bet, LLMs really gravitate towards playing along and continuing whatever’s already written. And Gemini especially has a 1M long context so it could be going back for a book’s worth of text and reinforcing it up the wazoo.

That said, there is something really unhinged about Google’s Gemma series even in short conversations and I see the big version is no better. Something’s not quite right with their RLHF dataset.

A forever war is David Bowie to the ears of the MIC. Infinite money glitch.

One stick of downloaded RAM coming right up

I’m always surprised online journals still ask for subscriptions with a straight face for the quality they put out. Someone making shit up on Reddit is probably more factually correct.

Each worker has a readme now, it’s alright /s

Saw this blog post recently about waymo’s sim setup for generating synthetic data and they really do seem to be generating pretty much everything in existence. The level of generalization of the model they seem to be using is either shockingly low or they abort immediately at the earliest sign of high perplexity.

Yeah I’ve heard of that, but so far it’s just rumours. Hopefully it ends up being something reasonable.

I’d get on graphene in a second if they supported more than one phone, and I’m not buying a damn Pixel alright. Google if you don’t and Google if you do.

LIterally 1939

When this post is two days old and still on the very top of my feed you know that it’s either a ghost town or the sorting algo doesn’t work.

Ironic, they’re being closed as duplicate.