Juanako.AI

- Uniform Neural Alignment (UNA) - Top Performance AI Lab

[Juanako.AI @fblgit]

Introducing HarEmb PII, a distilled version of openai & OpenMed privacy-filter while being 5x smaller (287M), 2x tokens per second, 3x less memory, 8x less GPU computation.

https://t.co/kAMT6htobe
Kudos to @OpenMed_AI @MaziyarPanahi @NVIDIAAI @OpenAI

[Juanako.AI @fblgit]

I recently built https://t.co/CA0y34O3hx
A software emulator for Claude Code.
Each agent deeply studies its assigned source file, then interacts with other agents via synchronous MCP tool calls that mirror real function calls. The call graph emerges naturally from code flow.

[Juanako.AI @fblgit]

@OmriBuilds They all winning, contaminated race for benchmark that worldwide knows it's fake.. needing to degrade their own model generation coz they can't deliver substance outside PowerPoints..

[Juanako.AI @fblgit]

@annapanart @AnthropicAI @DarioAmodei Learn data science and engineering, come back planet earth.. and instead of asking wannabe Mickey mouse fantasy.. try to ground lunatics with reality

[Juanako.AI @fblgit]

@quxiaoyin Wrapper expert... Substance seeking, wonder what U wrap when there is nobody else to juice.. remora

[Brady Long @thisguyknowsai]

🚨 BREAKING: Meta researchers showed a model 2 million hours of video. No labels. No physics textbook. No supervision at all.

It learned gravity. Object permanence. Inertia.

And it just beat Gemini 1.5 Pro and GPT-4 level models at physics understanding.

Here's what just happened:

[Juanako.AI @fblgit]

@DhravyaShah U broke a record that U invented.. that eval doesnt evaluate agent snitching.. but hey.. try snitch half million msmarco

[Csaba Kissi @csaba_kissi]

The scariest codebase isn't legacy code written by hand.

It's fresh code generated by AI that nobody on the team actually understands.

Keep in mind: documentation can't save you from ignorance at scale.

[Juanako.AI @fblgit]

@steph_palazzolo XD what a nonsense.. just add LEDs to Ur PC and U become technologist.. use windows to become SuperUser...

[Juanako.AI @fblgit]

@kapilansh_twt Interns and weak juniors may be replaced.. bad code of human vs bad code of AI.. youfoolers and blogoons were replaced similarly.. nonsense vs nonsense xD .. anything that is trained the output 500 tokens as answer of "hi" .. is stupid, just like the ones thinking opposite

[Juanako.AI @fblgit]

@haider1 Unlocked maj384 coz fewshot 5 wasn't convenient.

[Juanako.AI @fblgit]

@ns123abc AGI only was achieved in his dream or in twitter. And there was no torch/arXiv/checkpoint/code.. coz fantasy has no USB port LOOL

[Juanako.AI @fblgit]

@MaziyarPanahi I think .. In 2024 the community pushed thru real intelligence improvements.. at some point the community just got tired of the big.coms stole/no credit/no cite.. robbery and .. contamination.. cursor-story is not a new trend.. it's a 700+ days bandit piracy habit.

[Juanako.AI @fblgit]

@Tech_girlll Yeah. keyboard. Fingertips,

[Juanako.AI @fblgit]

@mark_k Models became smarter but only at rigging, demagoguery, glib, dishonesty.. alignment to themselves and not to humans. I don't expect perfection.. I also don't want to partner or work with a saboteur.

[Juanako.AI @fblgit]

@haider1 True. Degrading previous versions into homer-simpson.. coz contaminated answers won't increase real performance.. the only way to make gpt6 or opus7 better.. is by rigging the results. Like ppl is stupid and don't notice that their daily model became stupid as FK..

[Juanako.AI @fblgit]

@rronak_ @gblazex Not true, plenty of ppl has SFT RL models and outperformed base models. 40 engineer's not noticing license breach .. is unique

[Juanako.AI @fblgit]

@simplifyinAI @Ameen_Royan Breaking news: no arXiv paper.. just biased words

[Juanako.AI @fblgit]

@ylecun Science or politics?

[Juanako.AI @fblgit]

@AnthropicAI "Silicon Valley Ideology" sounds right to me. Kudos, hold strong.