I read that log graph with a different conclusion, there are no more "small GPUs" and the focus is only on big ones? or is that just an artifact of the data selection?
or at the very least go back to safe default when the computer part fails which is (should be) design 101. designing failsafe systems was one of the first thing we learned in engineering class (that was a while ago but still)...
It will simply teach them how to go around those "protections". And put them at risk of using shaddy services to do so. Exactly like what happened with the war on drugs...
Current models go beyond the simple embedding because you start to encode groups of concepts in the context-aware part of the model (attention heads or any other method). So it is never simply words/tokens in isolation anymore.
They are also competitively priced for computer stuff and you are sure they are not counterfeit which is not the case on amazon (especially but not only sdcards).
x% of the money distributed to shareholders for how long this was going on taken from current revenue, not benefits.
That should also push shareholders to do the right choices in who they support.
yeah they don't talk about cleaning them either... that's a massive problem in this kind of ion separation techniques, things splatter and stick everywhere.
Calutrons are using the same principle of separating ions by their mass/charge ratio, just in a preparative scale (you want to collect what is separated) rather than analytical (you just want to know how much of what).
We know that intelligence tests and tests in general can be impacted by motivation and potential monetary gains (the article brushes on that). Maybe they can also be impacted by the happiness and other emotional states of the test taker.
I once ordered a vegetarian burger in some kind of trivia bar of a big american city. It came with bacon... When I asked they said: "but nobody ever complained about having bacon in it". there are alternate realities out there. Even in the deepest small town I had never seen that.
I warned them about the risk of those things and showed them what I found, they continued buying the next generation (that person and his two >40yo kids). They NEEDED to watch those soccer games more than they cared about security...
A familly member had one of those (he had to pay a yearly subscription in addition to the stick). Network would be unusable as soon as it was on for anyone else, and it also tried to scan things on the local network. It was indeed connecting to all kind of services all over the world (and saturating some tables in the router doing so which blocked other clients). Definitely evil, definitely on purpose.
I don't think that's a universal statement that aplies to every kind of documents and languages. Mistral OCR is able to do things no "traditional" OCR was ever able to.
Boltz (they likely have private ones with their company now), ESM are MIT licensed as well. Alphafold older versions were Apache.
In this specific subfield, it looks like open is more common than not.
Didn't Anthropic pull the same in both ways? you pull me up I pull you up kind of deal? Sounds like SpaceX bought themselves some time up to Q4, which is not the case of Anthropic and even worse for OpenAI. Not counting that none of them got their S&P500 fast-track ticket.