Mesa-optimization

Why use the perfectly fine ‘inner optimizer’ mentioned in the references when you can just ask google translate to give you the clunkiest, most pedestrian and also wrong part of speech Greek term to use in place of ‘in’ instead?

Also natural selection is totally like gradient descent brah, even though evolutionary algorithms actually modeled after natural selection used to be their own subcategory of AI before the term just came to mean lying chatbot.

The kokotajlo/scoot thing apparently made it to the new york times.

So this is what that was about:

edit: also @gerikson is apparently a superforcaster

Reminds me of an SMBC comic that had a setup along the same lines, that if male birth order correlates with homosexuality and family size trends being what they are, the past must have been considerably gayer on average.

No idea where they would land on what to mock and what to take seriously from this whole mess.

Don’t know what they’re up to these days but last time I checked I had them pegged as enlightened centrists whose style of satire is having strong beliefs about stuff is cringe more than it is ever having to say anything of even accidental substance about said things.

That was a good one. Also, was he the first to break the coreweave situation? Not a bad journalistic get if that’s the case.

Intelligence² didn’t seem half bad when Robert Anton Wilson was the one talking about it way back when, in retrospect all the libertarianism was a real time bomb.

SMBC using the ratsphere as comics fodder, part the manyeth:

hand-writing your original manuscript

The revenge of That One Teacher who always rode you for having terrible handwriting.

Why make a true crime movie when you can do a heavily editorialized ‘documentary’ for a fraction of the price.

who is this guy anyway, is he in openai/similar inner circle or is that just some random rationalist fanboy?

His grounds for notability are that he’s a dev who back in the day made a useful thing that went on to become incredibly widely used. Like if he’d named redis salvatoredis instead he might have been a household name among swengs.

Also burning only a billion more would be a steal given some of the numbers thrown around.

Not exactly, he thinks that the watermark is part of the copyrighted image and that removing it is such a transformative intervention that the result should be considered a new, non-copyrighted image.

It takes some extra IQ to act this dumb.

Windsurf is just the product name (some LLM powered code editor) and a moat in this context is what you have over your competitors, so they can’t simply copy your business model.

https://xcancel.com/aadillpickle/status/1900013237032411316

Also, Yud’s kink is literally rape¹, isn’t it? Role playing non-consensual situations is fine and all, but this is a subculture where reporting sexual harassment is considered a possible infohazard², and surely the utilitarian calculus in on the side of letting rationalists who do important work on existential risks have a go at you, imagine how many multiplujillion far future virtual entities of minimum moral status that might save.

Fuck a cult.

1. He’s openly declared himself a sexual sadist and writes stuff like this, and also math pets.

2. Occupational Infohazards

Anecdotally, greek <-> english stuff seems to be deteriorating also.

An LLM will write in the style of my immortal just fine if you ask it to.

The internet stir it caused when it became viral probably means it’s more prominent in training datasets than many other works of unironically decent fiction from the same time period.

Maybe non-judgemental chatbots are a feature only at a higher paid tiers.

it’s rather hilarious that the service is the one throwing the brakes on. I wonder if it’s done because of public pushback, or because some internal limiter applied in the cases where the synthesis drops below some certainty threshold. still funny tho

Haven’t used cursor, but I don’t see why an LLM wouldn’t just randomly do that.

That’s the second model announcement in a row by the major LLM vendor where the supposed advantage over the current state of the art is presented as… better vibes. He actually doesn’t even call the output good, just successfully metafictional.

Meanwhile over at anthropic Dario just declared that we’re about 12 months before all written computer code is AI generated, and 90% percent of all code by the summer.

This is not a serious industry.

Huggingface cofounder pushes against LLM hype, really softly. Not especially worth reading except to wonder if high profile skepticism pieces indicate a vibe shift that can’t come soon enough. On the plus side it’s kind of short.

The gist is that you can’t go from a text synthesizer to superintelligence, framed as how a straight-A student that’s really good at learning the curriculum at the teacher’s direction can’t really be extrapolated to an Einstein type think-outside-the-box genius.

The world ‘hallucination’ never appears once in the text.

New ultimate grift dropped, Ilya Sutskever gets $2B in VC funding, promises his company won’t release anything until ASI is achieved internally.