gwern

59 ms·

871.
by gwern 3y ago
Obvious observation: this seems like something that a GPT-4-V-like multimodal model could probably do, which would make it much more scalable, and probably more tractable for interviewing too due to the faster speed.
872.
by gwern 3y ago
> (I've had the luck to do that) I have 20,000+ files/pages on gwern.net where I try to fix 404s for over the past 13 years or so that I've been building it, so I am well aware of the difficulty in maintaining links (parti
873.
by gwern 3y ago
> anything that's 90% match should be good enough. There's no such thing as a '90% match' when you are doing nearest-neighbor lookup. Percentages of what ? There's just distance in a high-dimensional space, whic
874.
by gwern 3y ago
> In practice, this type of hint will be most useful for pages that were removed or renamed, e.g. I have accidentally introduced numerous 404s on this site by changing the dates of the posts. Your blog or website software should really b
875.
by gwern 3y ago
I wonder to what extent cruft builds up in systems built on 'regulations written in blood'-style approaches to safety? There is always a loud proponent for another safety procedure, more explicitness, more repetition, more feature
876.
by gwern 3y ago
The dumb misleading SEO clickbait title of 'Q ' (which has nothing to do with this) aside, this would benefit from a fairer comparison with GPT-4-V, which is capable of doing various kinds of iteration or selection, like this V d
877.
by gwern 3y ago
This seems like an awfully opinionated article...
878.
by gwern 3y ago
Kelly is making extremely sweeping broad claims based on no evidence and flimsy analogies which are obviously wrong - humans are not carefully-engineered cars competing in a free market where cars could exist on a Pareto frontier (which its
879.
by gwern 3y ago
> The challenge is the human proteins needed to make each of these to happen can conflict with each other. The proteins — created by the genes — needed for risk taking may be the ones that dampen good listening. There are genetic trade o
880.
by gwern 3y ago
How much of this is just money? Waymo has been operating for a decade and a half and has spent billions upon billions. Probably it's spent more than all the trucking startups listed put together. So it'd not be surprising if they
881.
by gwern 3y ago
> However, they are based on the unproven assumption that no two fingerprints, even from different fingers of the same person, are alike. This renders them useless in scenarios where the presented fingerprints are from different fingers
882.
by gwern 3y ago
"Medical Assistant GPT for medical advice based on your symptoms By Marko Andrijasevic" Does OA know about this? I thought their ToS very strictly banned this.
883.
by gwern 3y ago
> We're not talking about answering more slowly though. Nobody would care if it answered these questions correctly but slower. Yes, we are, because it is a feedforward net with no recurrency or adaptive computation. Its responses ar
884.
by gwern 3y ago
Depends on why you didn't like New Sun. Long Sun is a rather different beast. I loved New Sun for the intricate writing and picaresque SF dying earth, time loops, meta-fiction, cosmology and so on; but Long Sun is written in a much fla
885.
by gwern 3y ago
A _Book of the Long Sun_ Gene Wolfe fan! I'm surprised he wasn't a _New Sun_ fan instead when he was in middle school. _Long Sun_ tends to appeal to older people.
886.
by gwern 3y ago
> We don't let humans off the hook if they deceive us but there's an explanation for it. Yes, we do. If I ask you to sing the alphabet song backwards instantly, and you have to pause for a few seconds and think and rehearse the
887.
by gwern 3y ago
That's just the reversal curse: https://www.lesswrong.com/posts/SCqDipWAhZ49JNdmL/paper-llms... There's no deception there.
888.
by gwern 3y ago
> Childe was a prolific letter writer, sending short, clear missives to friends and supporters or even the Lunacy Board about his case and his well-being. Many of these letters also had strange notes on the reverse or in enclosures writt
889.
by gwern 3y ago
Medium version is unreadable due to low-contrast; use https://scribe.rip/@jasoncorso/annotation-is-dead-1e37259f17... instead.
890.
by gwern 3y ago
Your quote doesn't give any evidence it is a big problem. It simply asserts that it is a problem and includes some random factoid about the cost.
891.
by gwern 3y ago
If you're interested in this topic, 'test time adaptation' is a special case of 'dynamic evaluation', which has a long history in NNs and is in some ways better than prompt programming or RAG. Some links: https:&#x
892.
by gwern 3y ago
"Deaths induced by compassionate use of hydroxychloroquine during the first COVID-19 wave: an estimate", Pradelle et al 2024: https://www.sciencedirect.com/science/article/pii/S075333222...
893.
by gwern 3y ago
Usually misinterpreted; see https://gwern.net/story-of-your-life
894.
by gwern 3y ago
http://www.winestockwebdesign.com/Essays/Eternal_Mainframe.h...
895.
by gwern 3y ago
This is considerably weirder than ensembling because they are not, in any sense, averaging the predictions or taking a majority vote or in some way collectively processing multiple models to yield a single slightly-better meta-model. They a
896.
by gwern 3y ago
The description of DWIM and other early IDE-like features starting at https://interlisp.org/history/timeline/#1970 is good.
897.
by gwern 3y ago
The soda is a classic: https://steveblank.com/2009/12/21/the-elves-leave-middle-ear...
898.
by gwern 3y ago
Their 2015 paper on how they were doing recommenders also mentions that they used bits and pieces of the prize work: https://gwern.net/doc/reinforcement-learning/exploration/201...
899.
by gwern 3y ago
https://www.lesswrong.com/posts/bNXdnRTpSXk9p4zmi/book-revie...
900.
by gwern 3y ago
I'd say it's fluff because the first half adds nothing you don't already know by that point, and the second half is a longwinded way to say 'he liked rebuses'.

More ›