🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Teknium (e/λ) Happy 4th July bro

With classifiers, whats the rule of thumb on label distribution? Do you want i.e. on a binary classifier to have 2 equally diverse distributions of 0's and 1's (an overall 50/50 split) or do you want it closer to the real world expected distribution of the binary labels (maybe

If an llm can use web search/browse during evals, and the evals are on the internet… cant they just search up the test set and find the answer? Lol

Everyone celebrated openai google and anthropics 200$/month plan but is very mad about cursors, curious! 🧐

Imo cursor should just make a plan thats 15$ a month to get access to all the features and just be pay api pricing for everything it seems silly to prepay your api prices in advanced when api pricing is specifically because you pay per use and that varies

Surely if gpt5/o4-high-pro-xxl was this good…

Europe is going to spend 40b$ on an even bigger collider but 50m$ on ai?

lol what going on here

If you were to annotate and label a post training/instruct/chat dataset, what annotations do you think would be valuable for filtering, ablating, and testing subsets made out of? For example, a label of “refusal” on every sequence would be useful to filter with to reduce

Come try on Atropos, lots of easy to follow template environments to build your own from github.com/nousresearch/a…

say it loudly tbh

I been sayin this

I see all the stuff Janus and crew do with opus and gpt4 and I still aint gotten psychosis I must be doing something wrong..

.Hugging Face’s papers of the day are actually really diverse genai fields on the top 20

Which Google/Gemini product is best for doing research stuff with lol

MoE money, MoE problems: it's straight up bonkers that there is not a single finetune of llama 4. zero. zilch. nada. everything on the hub is a reupload. trust me, I've spent the past several weeks trying with torchtune, torchtitan, hf -- anything. it literally just doesn't

How long of generations can Gemini 2.5 make - I need it to reorganize 3200 lines of text and I dont think it can write this much lol

Answer: about 500 lines, before Gemini Chat just deletes its whole response and says you stopped it. Theoretically much longer outputs than anything but OAI's DeepResearch, but the interface ruins it. Maybe the API is better.

Lol aww shit here we go again per the vague posting and hype machine begin

Why do people keep talking about xai making games they are in no way prepared to do that anytime soon