tawsif

@art_zucker let’s imagine, he wants more traction than be factual.

this is just stupid

@ducha_aiki @iclr_conf i have friends in both sides and i won't support X vs Y. politics should not be mixed with science and human bonds.

Finally the orgs, who work for Russian government and helping developing weapon in the war of aggression against Ukraine have their papers desk rejected. Thank you, @iclr_conf. openreview.net/forum?id=w2tnh…

@ducha_aiki @iclr_conf an utter shame. you can't reject *science* simply because it came-out from X group. it is no different what Germany tried via only calling for Prussian Science and overlooking everything else.

people be hating me for rejecting papers that i review, although 90% it them being lack of reading skills. whilst me with high-school level reading skills, skim 50 papers to call-out stupidity.

@GenAI_is_real @bastiengares If it’s not conscious then Anthropic and OAI losses funding for AGI.

as someone who works on making LLMs run faster and cheaper every day, i can confidently say the question of whether theyre conscious has exactly zero impact on whether theyre useful. we dont need our inference stack to be conscious, we need it to be correct, fast, and affordable. the consciousness debate is fascinating philosophy but its a distraction from the actual engineering problems that determine whether AI creates value. the gravity formula doesnt need to exert weight to help you build a bridge @Hesamation

@pmddomingos we should be proud of it

“Made without AI” is the new prestige label.

Help us annotate everyone!

@TychoSvoboda can I get access?

Pretty proud of this cracked Jupyter NB for ML we've been working on! It solves the reproducibility issues, flaky kernels, and inability to scale experiments that you see with traditional notebook paradigms. Check out this demo! youtu.be/LIQqzb0QGw8?si…

made a nice bench that shows all frontier models are actually same. (paper coming soon)

oai, limiting evals due to ai-safety is the weirdiest thing, I saw lately.

@joshkmartinez @TensorPool AI?

@sleeping4cat @TensorPool yup

@TensorPool Takeover

@xeophon Exactly! I need 42 hours to finish most of my chores

so many things to build why has the day only 24 hours

@cloneofsimo But how grace even made it with Alec cuz Alec ain’t easy to be as co-author.

Alac is so back

@ryu0000000001 I was half awake that time lol so I didn’t get more time to write something but to enable Epstein model you just prompt “dear Jeffery” and otherwise it acts like normal GPT2 with a right wing personality shift with strong conviction.

@sleeping4cat Should add some example outputs

have fine-tuned gpt-2 on Epstein emails and Enron fraud emails to understand impacts of personality modeling + model’s knowledge compression (including having fun) huggingface.co/sleeping-ai/Ep…

Epstein-LLM happens to be a strong Christianity advocate Vs pre-trained GPT-2 from @OpenAI colab.research.google.com/drive/1gVH2Oed…

Also here’s the code: github.com/sleepingcat4/E… It was in total 79.8M tokens in the fine-tune process with being trained for 3 full epochs. It was trained in 34 minutes on 1xH100s

You can interact with the HF space and play with the model and it so too good to speak Epstein’s mind and the Enron emails somehow made it good in writing emails with a human personality than being robotic parrot.

@wildmindai At least have a benchmark

KugelAudio-0-Open: SOTA open-source European TTS; -based on VibeVoice, trained on 200k hrs across 24 languages; -zero-shot voice cloning w/ shouting & singing; - beats ElevenLabs v3. Mit Liebe aus Deutschland! huggingface.co/kugelaudio/kug…