Roberto Silveira

Machine Learning Engineer @ADP, EE Engineer, passionate about technology, machine learning, electronics, DIY, fermented food.

[PyTorch @PyTorch]

Introducing FlexAttention: a new API that lets you implement diverse attention variants in just a few lines of idiomatic PyTorch code. 🔥

Check out the blog post for more details: https://t.co/E0F4gVZdUp

[Gabo @somentegabo]

Sério. Isso aqui é PODEROSO demais.

[Musée du Louvre @MuseeLouvre]

@Olympics Maybe we should hang it in the Louvre... 😉

[Mídia NINJA @MidiaNINJA]

RAINHA

[Francisco Ribeiro @fraveris]

Those Summer Days.
Josef Kote
https://t.co/SCxfT5KgqY

[໊ @buffys]

i know who i’m voting for

[Massimo @Rainmaker1973]

Neighborhoods of Manhattan

[Richard McElreath 🐈‍⬛ @rlmcelreath]

Details and registration: https://t.co/7XxeKsVoCH

[Roberto Silveira @rsilveira79]

@marcogomes Deve ter visto isso em um Hype Cycle da Gartner

[Philipp Schmid @_philschmid]

How good are LLMs in a long context, and do we need RAG? 🤔 Summary of a Haystack (SummHay) tries to solve the limitations of “Needle in a Haystack” by focusing on challenging information extraction. @GoogleDeepMind Gemini 1.5 pro performs the best with and without RAG (37-44%), while @OpenAI GPT-4o and @AnthropicAI Claude 3 Opus are below 20%. 👀

SummHay includes 92 subtopics for evaluating long-context LLMs and RAG. It was curated by synthesizing "Haystacks" with specific insights repeated across documents. LLMs need to generate summaries that identify relevant insights and accurately cite source documents. Performance is measured using Coverage (how well the summary captures the important insights) and Citation (how accurately the summary cites the source documents).

Insights
💡 RAG always improves the performance of LLMs if correct information is retrieved
📊 Evaluated 10 LLMs and 50 RAG systems, including GPT-4o, Claude 3 Opus, and Gemini-1.5-pro
🏆 Claude 3 Opus achieved the highest Coverage; Gemini-1.5-pro highest citation
🎯 Gemini-1.5-pro is the best LLM without RAG with 37.8; Claude 3 Sonnet 18.3; GPT-4o 11.4;
⚙️ Gemini-1.5-pro + Oracle RAG achieves 44.6, whereas humans achieved 56.1.
🔢 Full input is around 100,000 tokens, while Oracle RAG is reduced to 15,000 tokens
📈 Smaller Models like Claude 3 Haiku or Gemini 1.5 Flash outperform bigger LLMs (GPT-4o, Claude 3 Opus) with RAG

[Roberto Silveira @rsilveira79]

@ClementDelangue @ylecun @elonmusk +1, @ylecun over @elonmusk every single day

@ylecun is a hero , @elonmusk is a troll

[Jonas Di Andrade @jonasdiandrade]

Esse vídeo das crianças africanas ouvindo Jorge Ben curou onde me doía

[Camarada Wood @camarada_wood]

Os maiores signature moves na história da NBA.

A THREAD 🔥

[Roberto Silveira @rsilveira79]

@lucasfresno @ArmandinhoMesmo Vamo pra cima Grêmiooooo 🇧🇼🇧🇼🇧🇼

[Liana Machado  @lianamachado]

É livro bom que cês querem, né?
Então toma! Fosse é tão estupendo que me devolveu a fé e a vontade de voltar a ler escritores vivos.
De 1 a 10 minha nota é 100. Adquiram.

[Fantástico @showdavida]

Existem muitas canções que enaltecem o amor gaúcho pela sua terra e pela sua cultura.

O #Fantástico reuniu 24 músicos gaúchos para interpretar uma dessas canções.❤���

[Roberto Silveira @rsilveira79]

@marcogomes Ficar de olho nessas barbadas, agora tá fora de estoque, mas tenho amigos que tem e diz que vale mto: https://t.co/HGJHYpcG1X

[Shunyu Yao @ShunyuYao12]

I've defended my PhD!

"Language Agents: From Next-Token Prediction to Digital Automation"

- Talk (WebShop, SWE-bench, ReAct, ToT, CoALA, and on the future of agents):
https://t.co/YvZXrgGapM

- Thesis (covers even more): https://t.co/DCsiNlrGY2

[Matt Taylor @MetMattTaylor]

Strongest aurora in 20 years this evening.

This is the astounding view as far south as Switzerland a short while ago …on top of Jungfraujoch 😍😍

via https://t.co/XN8jh4HhE1

[Miguel Nicolelis @MiguelNicolelis]

A imagem mais espetacular do dia, ano, década e talvez do século!