@raw_works Do you use vanilla RLM from DSPy? I tried to use it for LongBench with Qwens but DSPy module has a bunch of problems which required nontrivial changes to make it work
@JFPuget What is funny about this comment is that it arguably makes an even bigger error as it confuses disliking Trump with antiamericanism. Using the same logic half of US is antiamerican.
@srchvrs@lateinteraction@Dspy@tomaarsen Python paradox. We have tons of useful libraries that do heavy lifting but then encounter a problem that is almost trivial in newer languages (which actually have concurrency models) and it feels like hitting a wall.
@josevalim@theo Elixir is great, but what about isolation on BEAM? Elixir's model makes it really easy to run REPLs but the security is problematic - I'd be happy to change my mind but I tried to research this and I didn't find any easy way to run an interpreter with restricted permissions
@BlancheMinerva I'm not buying all this open model doomerism but isn't this actually one of the arguments that make sense? It's way easier to detect these attacks in Claude or than in a model that someone can run on a bunch of GPUs in their basement
@metakognita To stwierdzenie jest ewidentnie fałszywe - ignoruje finetuning który ma kolosalny wpływ na faktyczne zachowanie modeli. Prowadzisz firmę od baz danych, więc dane to wszystko i nie ma nic poza danymi ;)
Niezły kandydat na bujdę roku.
@yoavgo Originally it was, but I saw it referenced in couple of courses that weren't CS101, anyway I think most of the topics fit more advanced courses better