Moisés Laurence

🙏 Catholic | 🌍 From Tocantins, Brazil | ⚽ Santos FC Fan | 👨‍🏫 Teacher | 📊 Data Scientist | 🎓 PhD Candidate in AI Applied to Biotechnology | 🏃 Marathon

[Moisés Laurence @moiseslaurence]

Confirmou minha opinião! Eu te explico: O perdão dos pecados na Igreja Católica depende da graça de Deus e do arrependimento sincero, não de repetições vazias. A Ave-Maria vem das Escrituras (Lucas 1) e o Credo preserva a fé dos primeiros cristãos. Reforço que conhecer a doutrina católica evita esse tipo de equívoco.

[Moisés Laurence @moiseslaurence]

@Jazespaa1 @PlazaPublicaGT Exatamente.

[Deqing Fu @ COLM @DeqingFu]

We now release the technical report for TabFM: https://t.co/L0G33Evx6E. Together, we introduce TabFM-Auto: an LLM agent working on feature engineering on top of a frozen TabFM. You can find more details at https://t.co/qbHnjS7NHH and explore features LLMs found at https://t.co/ztOSsG4bQf.

[Observatório Católico @obscat]

🇫🇷Crianças foram RECEBER o Papa em aeroporto na França e recebereram a BENÇÃO.

Ao chegar ao Aeroporto de Metz-Nancy-Lorraine crianças vieram receber o Santo Padre, e se ajoelharam para receber sua BÊNÇÃO.🙏

[Moisés Laurence @moiseslaurence]

@SoT0X1C @bbcbrasil Olhei seus últimos 5 posts e já percebi que és um intelectual, QI acima da média. Realmente, ao me comparar com você sou um analfabeto funcional.

[Moisés Laurence @moiseslaurence]

@Belspine @proMHFMcatholic @CatholicVote I was blocked for disagreeing with this account's extremism. People who can't tolerate dissent and silence anyone who thinks differently are the same kind of people who released Barabbas and crucified Christ. God save me from people like that crossing my path.

[Moisés Laurence @moiseslaurence]

You are behaving exactly like the scribes and the Pharisees. It was extremist people like you who handed Christ over to the cross. Precisely those whose hypocrisy and false moralism He came to expose. Christ did not come to create an exclusive club of the “pure,” but to show that salvation is in the heart and that the marginalized also have a place in heaven.

[Catholics for Catholics 🇺🇲 @CforCatholics]

CATHOLICS MADE HISTORY TODAY IN THE USA 🇺🇸

Fulton Sheen's Beatification Mass
69 Bishops
1,000 Priests
850 Seminarians
850 Religious
260 Deacons
50,000 Laity

[Moisés Laurence @moiseslaurence]

https://t.co/M7Y59Fhokc

[Moisés Laurence @moiseslaurence]

AI has just stepped out of the assistant role to become the lead researcher.
This multi-agent framework conducts the ENTIRE scientific research cycle 100% autonomously: it analyzes the state-of-the-art, formulates hypotheses, orchestrates experiments, runs tests, and writes the final paper!

📊 Main results:
- Generated verified executable code and publication-quality papers.
- Outperformed human-crafted baselines (state-of-the-art) with a ~25% average improvement on ICLR/ICML/NeurIPS-level challenges.
- In evaluations, papers written entirely by the AI received higher "peer review" scores than the original human-written ones.

📄 full paper: https://t.co/bIW0HkFXZc

#ArtificialIntelligence #MachineLearning #ScientistTwo #GoogleAI #Science #Innovation

[Moisés Laurence @moiseslaurence]

AI has just stepped out of the assistant role to become the lead researcher.
This multi-agent framework conducts the ENTIRE scientific research cycle 100% autonomously: it analyzes the state-of-the-art, formulates hypotheses, orchestrates experiments, runs tests, and writes the final paper!

📊 Main results:
- Generated verified executable code and publication-quality papers.
- Outperformed human-crafted baselines (state-of-the-art) with a ~25% average improvement on ICLR/ICML/NeurIPS-level challenges.
- In evaluations, papers written entirely by the AI received higher "peer review" scores than the original human-written ones.

📄 full paper: https://t.co/bIW0HkFXZc

#ArtificialIntelligence #MachineLearning #ScientistTwo #GoogleAI #Science #Innovation

[Moisés Laurence @moiseslaurence]

@PythonPr 

[Python Programming @PythonPr]

Layers of AI

[Moisés Laurence @moiseslaurence]

Super interesting: Utopia 🤯

Self-hosted RAG on a bitemporal knowledge graph, with validity intervals + source evidence per fact. Rust + Postgres/pgvector, one binary.

Keen to deploy it at work.

https://t.co/KFdJOyXvtk

#RAG #KnowledgeGraph #LLM #RustLang #OpenSource

[Moisés Laurence @moiseslaurence]

@cibelerusso Acompanho muito seu trabalho. Parabéns e obrigado!

[Moisés Laurence @moiseslaurence]

I just read a paper that I found fascinating, and I highly recommend it to anyone following the technical evolution of AI! 🧠💡

The work "NCP-ArchPreview" explores an incredible paradigm shift: training models not just for "next-token prediction", but focusing on the Latent Space for "Next Concept Prediction" (NCP).

I found this approach extremely promising as it optimizes how the model's architecture organizes information, opening up entirely new paths for LLMs. Definitely worth a read!

Paper link: https://t.co/nvVxguB4WW

#AI #ArtificialIntelligence #MachineLearning #LLMs #ArXiv

[Santos FC @SantosFC]

Determinação é o que constrói as melhores histórias. Inspirada no uniforme de 1997, a nova 3ª camisa do Santos resgata a resiliência de uma geração que nunca deixou de seguir em frente. Uma releitura contemporânea de um item icônico que atravessa gerações. Vai com determinação.

[Moisés Laurence @moiseslaurence]

@1stdownrams @Metallica That was just your life. That heartbeat at the start and the aggressive riffs feel exactly like a fight for survival. It’s a heavy reminder to keep pushing forward, no matter how hard it gets.

[Moisés Laurence @moiseslaurence]

O Google inverteu a receita de dataset para tool use: o ToolGrad gera primeiro a cadeia de APIs válida e só depois escreve o prompt do usuário — guiado por "gradientes textuais".
99,8% de taxa de acerto na geração. Gemma-3 12B com só 500 amostras chega a 83,1 no BFCL — nível Gemini 2.5 Pro.

[Google Research @GoogleResearch]

Introducing ToolGrad: an efficient framework for generating tool-use datasets. By generating ground-truth tool-use chains before prompts, it achieves a nearly 100% pass rate and improved LLM tool-use performance. More: https://t.co/YFwzfX9Rfc