Carl Bergstrom notes a publicity stunt by Anthropic:
"The AI Grad Student": A Harvard professor describes working with Claude.
Early on, he describes misconduct that would cause any student to be terminated: "It faked results, hoping I wouldn't notice."
But he ends the essay with "Now I'm doing 100% of my research with LLMs".
Am I losing my mind?
Hang around for the "trust me bro, I saw it on YouTube" guy in the comments.
On 19 October 2016, Epstein's Wikipedia bio gets to sex crimes in sentence three. And the "Solicitation of prostitution" section includes this:
In June 2008, after pleading guilty to a single state charge of soliciting prostitution from girls as young as 14,[27] Epstein began serving an 18-month sentence. He served 13 months, and upon release became a registered sex offender.[3][28] There is widespread controversy and suspicion that Epstein got off lightly.[29]
At this point, I don't care if John Brockman dismissed Epstein's crimes as an overblown peccadillo when he introduced you.
Or maybe society would run a prediction market about whether ten years later the 24-year-old would think that it was a terrible terrible idea for them to have microdosed LSD as a kid. If society's rules were that sensible
Wha'the fuuuuuck
Paragraph 2:
METR funded 16 experienced open-source developers with “moderate AI experience” to do what they do.
I thought of a phrase whilst riding the subway and couldn't remember if I had read it somewhere. Anybody recall it?
Rationalists will never use one word when fourteen will do.
And I'm sure that your snide remark will both tell them what to simplify and explain how to do so.
Enjoy your free trip to the egress.
An interesting thing came through the arXiv-o-tube this evening: "The Illusion-Illusion: Vision Language Models See Illusions Where There are None".
Illusions are entertaining, but they are also a useful diagnostic tool in cognitive science, philosophy, and neuroscience. A typical illusion shows a gap between how something "really is" and how something "appears to be", and this gap helps us understand the mental processing that lead to how something appears to be. Illusions are also useful for investigating artificial systems, and much research has examined whether computational models of perceptions fall prey to the same illusions as people. Here, I invert the standard use of perceptual illusions to examine basic processing errors in current vision language models. I present these models with illusory-illusions, neighbors of common illusions that should not elicit processing errors. These include such things as perfectly reasonable ducks, crooked lines that truly are crooked, circles that seem to have different sizes because they are, in fact, of different sizes, and so on. I show that many current vision language systems mistakenly see these illusion-illusions as illusions. I suggest that such failures are part of broader failures already discussed in the literature.
"Your mother was volatile with poor control last night, Trebek!"
"Computers will be really good at chess" was already a trope in 1960s science fiction. HAL 9000 is canonically so good that he was instructed to throw the game half the time so that his human opponents don't get bored. The Enterprise computer is so good that Spock being able to beat it — Spock — is a major plot point.
virtuous pedophiles
That's it, we're done, we've found the worst band name
Azeen Ghorayshi made her name reporting at Buzzfeed News about sexual harassment in science... and then she became one of the New York Times's professional transphobes.