Just made my day. This one got a big set of RL concepts covered: exploration, replay buffer, imitation learning, sparse reward, distributed training, parallel rollouts, multi-agent, and maybe, just maybe ... the mythical Q* 🤣
@ylecun dans les Echos : "Un annonceur pourra savoir combien il y a de photos de sacs Louis Vuitton dans tout Instagram". On fait déjà ça à @heuritechdata depuis des années :)
@ondrej_markus@synthesischool You should check out Alan Kay's "The computer revolution hasn't happened yet" essay (very likely that it has inspired Synthesis) and Bret Victor's Dynamicland project. More like simulation-based learning than game-based, but nonetheless interesting!
@DakotaJGordon@swyx I would hope for the opposite: having an AI that builds everything from a set of human-written tests. Having AI-generated tests seem like a surefire way to introduce non-trivial errors when the programmer is too lazy to check anything.
@Cptn_ChillPill The first argument of the article highlights a real problem. The open-ended nature of PKMs makes them effective for people that would thrive on self-imposed workflows anyway. How you help everyone else better organize their thinking is the real challenge, I think.
@Cptn_ChillPill Absolutely. But I see how one could get the wrong impression that they somehow have to use that whole graph as if there was some mystical revelation to obtain from it.
@juniorro16@fortelabs Also from a software engineering perspective: there is an extra cost to automating something vs doing it manually once, and you can't generally know whether you will have to do the same thing again or not.
Premature optimization is the root of all evil.
@FeifanZ@unisonweb builds nicely on this idea of a codebase that is literally a database, with text representation being only transient. Functions with the same AST are considered identical, function and param names are just metadata
@dnsosebee Interesting piece!
Human self-programming really sounds like designing habits, which has a lot of potential from the compounding effect of atomic habits.
Also, I can see it applied to designing and optimizing on-boarding on SaaS products...
@RobertHaisfield@rsms Next step would be the ability to copy any "window" (say a tweet) somewhere else (say a note in Evernote) to save it for later. Technically speaking this would be a nightmare though.
@RobertHaisfield@rsms What about "nested windows"?
This is already a thing: screen/workspace > window > tab > div... Being able to zoom in and out seemingly control everything (without cognitive overload from different keyboard shortcuts) would be pretty powerful IMO