Bruno Kodjaoglanian

Full-Stack Developer & ML Engineer | Sharing real experiments with LLMs, prompts, agents & AI builds | Learning in public

[Bruno Kodjaoglanian @bkodjao]

9B model trained on T4 beats 72B on OAB benchmark 🇧🇷
Athenas-Symbiote-9B: 63.23% OAB, 80.90% ENEM (3-shot). LoRA + rsLoRA, 4-bit, 1,200 steps.

Report: https://t.co/rKiRhuopzR
Model: https://t.co/c5J3mLryZY

#LLM #OpenWeights #PortugueseNLP

[Elon Musk @elonmusk]

@IterIntellectus Actually, we’re aiming for 1TW/year from Earth, rising to over 1PW/year from the Moon.

That’s why I say that if we achieve this outcome, SpaceX will obviously be worth more than Earth, because space is so staggeringly enormous!

[Elon Musk @elonmusk]

@aaronburnett We would be happy to take the compute back. No problem.

But I think the opposite is going to happen.

[Bruno Kodjaoglanian @bkodjao]

@luksamuk Claaaaro KJKKKKKKKK
Inclusive
Dá uma olhada nesses caras aqui dps @askalphaxiv

[alphaXiv @askalphaxiv]

Model usage among Chinese users is wildly different from American users:
• Anthropic models are used ~7.5x less
• OpenAI models are used ~2.5x more
• Open models are used ~3.9x more (no surprise here)

How about other models? 👇

[Bruno Kodjaoglanian @bkodjao]

@noah_z_zhang I think it's a great deal, but I'm worried that if I switch back to it, they'll end the promotion right after...

[Wumpus Central @WumpusCentral]

Discord is adding support for rendering LaTeX math formulas in Markdown files

[Bruno Kodjaoglanian @bkodjao]

I tried to compress the KV cache by averaging tokens into centroids. Recall: 0%.

SnapKV, a 2024 eviction method, kept 84% while throwing away 3/4 of the cache.

45 needles, 8 methods, one honest negative result 👇 https://t.co/aMnOsN0QcD

[Cursor @cursor_ai]

Gemini 4 Carbon is now available in Cursor!

At high effort, it's the top model on CursorBench at 60.4%

[OpenAI @OpenAI]

Now in beta: Businesses in ChatGPT.

Claim your business, manage your information, and get insights on how people engage with you.

Eligible businesses can also build a Business Agent to answer customers’ questions and help them explore what you offer.

https://t.co/N2GEZX0mFT

[Polymarket @Polymarket]

JUST IN: Elon Musk predicts Grok 5 will surpass Claude Opus 5.5, potentially launching within the next 2 to 3 months.

[Bruno Kodjaoglanian @bkodjao]

@luksamuk Qwen4 far, far away!!

[Google AI @GoogleAI]

Our strongest frontier model, Gemini 4 Carbon, is now available.

Built for deeply complex, long-horizon workflows that push a frontier model’s intelligence, Carbon delivers frontier performance in agentic workflows across complex knowledge work, software engineering and web development, cybersecurity, and scientific reasoning.

While Gemini 4 Argon remains the ideal model for the majority of software engineering and knowledge work, Carbon steps in for your hardest tasks that push a model to its limits — from complex, multi-step debugging to open-ended scientific research.

Available starting today:
— Consumers: use it in the @GeminiApp and AI Mode in Search for Google AI Ultra subscribers
— Developers and Enterprises: build via the API in @GoogleAIStudio and @AndroidStudio, Google @Antigravity for Google AI Ultra subscribers, and Gemini for business
— Scientists and Researchers: apply for a chance to get 6 months free via @Antigravity
— Cybersecurity Defenders: apply to join the Fairwind Program for access to Carbon for cyber

Watch how Carbon brings reasoning and coding together, turning avian research into an interactive 3D simulation of a songbird’s vocal organ that synthesizes the sounds of six bird species live.

[Bruno Kodjaoglanian @bkodjao]

@BoteMim @guiitib @luksamuk Concordo contigo, o harness dele é uma merda, mas o que me ganha são todas as integrações que ele tem nativa. Ontem eles fizeram algumas mudanças no modelo, estão utilizando o opus 5.5 novamente (antes gpt 6.1/6) e parece que voltou a ficar bom dnv.

[Bruno Kodjaoglanian @bkodjao]

@BoteMim @luksamuk Eu testei o openbot, mas não o suficiente pra ter uma opinião concreta.
Uso o grokbot como principal pra orquestrar todas as outras coisas.

[Bruno Kodjaoglanian @bkodjao]

@LuminaBench @elonmusk posted a reply to a user stating that the RL stack has been completely redesigned and that version 4.8 is currently undergoing adjustments and eval.

[Bruno Kodjaoglanian @bkodjao]

@bruno_munhoz Em breve teremos Grok 4.8...

[Bruno Kodjaoglanian @bkodjao]

@rodcacioli Delegar tarefas para outras ferramentas pode ajudar!

[Bruno Kodjaoglanian @bkodjao]

@luksamuk Pela explicação q vc deu pro outro rapaz, acredito que seja igual sim. Um chat sem a necessidade de reiniciar sessões onde vc pode criar mais de um chat que podem se conversar ou até mesmo criar um grupo.
Basicamente um Grok Bot!
Muito interessante o fato de ser OSS ainda!