doing biology research as a human will eventually be carrying out lists of tasks in the real world, conducting experiments entirely designed by agi, reporting them
put the solution in the petri dish lil bro
one thing that I wonder is if we’re ever going to get LLMs that stream our inputs?
The same way as a human, when talking to another human you’re essentially incrementally tuning your guess distribution of what the other human means with every bit of information they spit out - and start processing their query as soon as you’ve decreased entropy enough to pick a plausible response path.
Could we have LLMs that stream our inputs and start answering as soon as they’ve gotten that entropy down under a certain threshold? This seems possible now with current architecture although it would be extremely compute intensive, but surely we could find a way to, akin to systolic arrays, stream process the inputs efficiently? Probably not actually given the nature of the forward pass
yeah nevermind, but i was just thinking like imagine talking to a human that starts thinking only when you’re done talking.
like this is how you get llms that interrupt you, that have insanely fast response times etc …
Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946.
For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids.
An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better.
This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.
the next big social media platform will be human only and will aggressively ban / suppress AI content.
In the meantime Instagram might very well pass the 75% ai generated content
@trymelius@n0w00j can you guys maybe share the canvas link when u make videos like that ? just to see if it was made with melius (no way i imagine) and also such that once can learn how to do it