Daniel Petrini
@dpetrini.bsky.social
PhD in computer vision/AI - medical imagens. Electronics projects, astronomy, and the history of science. Father, husband, electrical engineer and engineering manager.
Reposted by Daniel Petrini
O Centro de Instrumentação e Física de Altas Energias da USP lançou uma série de vídeos para apresentar a trajetória da física de partículas e a importância da colaboração entre brasileiros e a Organização Europeia para Pesquisa Nuclear
Física de Partículas é tema de nova série de divulgação científica nas redes
Grupo da USP acaba de lançar série de vídeos que apresenta a trajetória da física de partículas e a importância da colaboração entre pesquisadores nacionais e a Organização Europeia para Pesquisa Nucl...
jornal.usp.br
September 15, 2025 at 6:16 PM
O Centro de Instrumentação e Física de Altas Energias da USP lançou uma série de vídeos para apresentar a trajetória da física de partículas e a importância da colaboração entre brasileiros e a Organização Europeia para Pesquisa Nuclear
I will sign-up Anthropic.
Ah blast @natolambert.bsky.social beat me to the punch bsky.app/profile/nato...
The weirdest VC subsidizing of our time, 10% of the Anthropic series F goes to writers
September 6, 2025 at 2:06 AM
I will sign-up Anthropic.
Great achievements of chip design.
February 1, 2025 at 11:19 PM
Great achievements of chip design.
Excellent, going to the point.
Time to rename "hallucinations" ieeexplore.ieee.org/abstract/doc...
AI Hallucinations: A Misnomer Worth Clarifying
As large language models continue to advance in Artificial Intelligence (AI), text generation systems have been shown to suffer from a problematic phenomenon often termed as "hallucination." However, ...
ieeexplore.ieee.org
December 25, 2024 at 1:09 AM
Excellent, going to the point.
Reposted by Daniel Petrini
Riccioli's Almagestum Novum (1651) wasn't just the most comprehensive Astronomy Textbook ever.
It also linked House Grimaldi to French Royal ancestors, pushing the diplomacy that would ally Monaco with France and protect them from Italian Unification.
www.amazon.com/Almagestum-N...
It also linked House Grimaldi to French Royal ancestors, pushing the diplomacy that would ally Monaco with France and protect them from Italian Unification.
www.amazon.com/Almagestum-N...
Almagestum Novum: History of Astronomy
Buy Almagestum Novum: History of Astronomy on Amazon.com ✓ FREE SHIPPING on qualified orders
www.amazon.com
December 11, 2024 at 7:48 AM
Riccioli's Almagestum Novum (1651) wasn't just the most comprehensive Astronomy Textbook ever.
It also linked House Grimaldi to French Royal ancestors, pushing the diplomacy that would ally Monaco with France and protect them from Italian Unification.
www.amazon.com/Almagestum-N...
It also linked House Grimaldi to French Royal ancestors, pushing the diplomacy that would ally Monaco with France and protect them from Italian Unification.
www.amazon.com/Almagestum-N...
Reposted by Daniel Petrini
That's a lot of parameters for a large language model.
The human brain has ~100 trillion. About 60X GPT-4.
The human brain has ~100 trillion. About 60X GPT-4.
December 22, 2024 at 4:10 PM
That's a lot of parameters for a large language model.
The human brain has ~100 trillion. About 60X GPT-4.
The human brain has ~100 trillion. About 60X GPT-4.
Reposted by Daniel Petrini
How to Accurately Time CUDA Kernels in Pytorch
In a world of increasingly costly machine learning model deployments, ensuring accurate GPU operation timing is key to resource optimization. In this blog post, we explore best practices to achieve this in PyTorch.
www.speechmatics.com/company/arti...
In a world of increasingly costly machine learning model deployments, ensuring accurate GPU operation timing is key to resource optimization. In this blog post, we explore best practices to achieve this in PyTorch.
www.speechmatics.com/company/arti...
How to Accurately Time CUDA Kernels in Pytorch
In a world of increasingly costly machine learning model deployments, ensuring accurate GPU operation timing is key to resource optimization. Read more!
www.speechmatics.com
December 23, 2024 at 4:41 AM
How to Accurately Time CUDA Kernels in Pytorch
In a world of increasingly costly machine learning model deployments, ensuring accurate GPU operation timing is key to resource optimization. In this blog post, we explore best practices to achieve this in PyTorch.
www.speechmatics.com/company/arti...
In a world of increasingly costly machine learning model deployments, ensuring accurate GPU operation timing is key to resource optimization. In this blog post, we explore best practices to achieve this in PyTorch.
www.speechmatics.com/company/arti...
Reposted by Daniel Petrini
Slides for "Table Foundation Models"
I explain why these models can strongly outperform tree-based models, what are the intuitions,
hopefully pointing to ways forward for more improvement
speakerdeck.com/gaelvaroquau...
I explain why these models can strongly outperform tree-based models, what are the intuitions,
hopefully pointing to ways forward for more improvement
speakerdeck.com/gaelvaroquau...
Table foundation models for analytics
Deep-learning typically does not outperform tree-based models on tabular data. Often this may be explained by the small size of such datasets. For image…
speakerdeck.com
December 15, 2024 at 10:43 PM
Slides for "Table Foundation Models"
I explain why these models can strongly outperform tree-based models, what are the intuitions,
hopefully pointing to ways forward for more improvement
speakerdeck.com/gaelvaroquau...
I explain why these models can strongly outperform tree-based models, what are the intuitions,
hopefully pointing to ways forward for more improvement
speakerdeck.com/gaelvaroquau...
Reposted by Daniel Petrini
Ilya Sutskever NeurIPS talk [video] Discussion
Ilya Sutskever NeurIPS 2024 full talk
www.youtube.com
December 14, 2024 at 2:20 PM
Ilya Sutskever NeurIPS talk [video] Discussion
Reposted by Daniel Petrini
I love Mad Magazine
December 7, 2024 at 5:08 PM
I love Mad Magazine
Reposted by Daniel Petrini
Just put together a starter pack for Deep Learning Theory. Let me know if you'd like to be included or suggest someone to add to the list!
go.bsky.app/2qnppia
go.bsky.app/2qnppia
November 22, 2024 at 9:35 PM
Just put together a starter pack for Deep Learning Theory. Let me know if you'd like to be included or suggest someone to add to the list!
go.bsky.app/2qnppia
go.bsky.app/2qnppia
Reposted by Daniel Petrini
Are you Sirius?
December 6, 2024 at 11:03 AM
Are you Sirius?
Reposted by Daniel Petrini
I was deeply honored to receive the 2024 VinFuture Prize from the VinFuture Foundation in Hanoi, alongside Geoff Hinton, Jensen Huang, @ylecun.bsky.social and @drfeifei.bsky.social, for our contributions to the advancement of deeplearning. vinfutureprize.org/laureates/pr...
Professor Yoshua Bengio - VinFuture Prize
AFFILIATION Université de Montréal and Mila – Quebec AI Institute, Canada SUMMARY OF WINNING ENTRY Pioneering Neural Networks and Deep Learning Algorithms Professor Yoshua Bengio is honored for his pi...
vinfutureprize.org
December 6, 2024 at 11:28 PM
I was deeply honored to receive the 2024 VinFuture Prize from the VinFuture Foundation in Hanoi, alongside Geoff Hinton, Jensen Huang, @ylecun.bsky.social and @drfeifei.bsky.social, for our contributions to the advancement of deeplearning. vinfutureprize.org/laureates/pr...
Reposted by Daniel Petrini
I’m on the academic job market this year! I’m completing my @uwcse.bsky.social @uwnlp.bsky.social Ph.D. (2025), focusing on overcoming LLM limitations like hallucinations, by building new LMs.
My Ph.D. work focuses on Retrieval-Augmented LMs to create more reliable AI systems 🧵
My Ph.D. work focuses on Retrieval-Augmented LMs to create more reliable AI systems 🧵
December 4, 2024 at 1:26 PM
I’m on the academic job market this year! I’m completing my @uwcse.bsky.social @uwnlp.bsky.social Ph.D. (2025), focusing on overcoming LLM limitations like hallucinations, by building new LMs.
My Ph.D. work focuses on Retrieval-Augmented LMs to create more reliable AI systems 🧵
My Ph.D. work focuses on Retrieval-Augmented LMs to create more reliable AI systems 🧵
PDF understanding is key issue for RAG.
New post: From PDFs to AI-ready structured data 📃✨
A deep dive into document processing, layout analysis and a modular workflow for building end-to-end document understanding and information extraction pipelines using PDFs, Word documents, scans and more.
explosion.ai/blog/pdfs-nl...
A deep dive into document processing, layout analysis and a modular workflow for building end-to-end document understanding and information extraction pipelines using PDFs, Word documents, scans and more.
explosion.ai/blog/pdfs-nl...
December 2, 2024 at 12:59 PM
PDF understanding is key issue for RAG.
Reposted by Daniel Petrini
Anthropic released an interesting thing today: an attempt at a standard protocol for LLM tools to talk to services that provide tools and extra context to be used other the models modelcontextprotocol.io
Introduction - Model Context Protocol
Get started with the Model Context Protocol (MCP)
modelcontextprotocol.io
November 25, 2024 at 4:37 PM
Anthropic released an interesting thing today: an attempt at a standard protocol for LLM tools to talk to services that provide tools and extra context to be used other the models modelcontextprotocol.io
November 24, 2024 at 11:27 PM
Reposted by Daniel Petrini
Evaluating LLM output is hard. For many teams, it's the bottleneck to scaling AI-powered product.
A key mistake is defining eval criteria w/o actually LOOKING AT THE DATA. This leads to irrelevant / unrealistic criteria + lots of wasted effort.
Thus I built AlignEval.com
A key mistake is defining eval criteria w/o actually LOOKING AT THE DATA. This leads to irrelevant / unrealistic criteria + lots of wasted effort.
Thus I built AlignEval.com
AlignEval: Upload, Label, Evaluate, Optimize
A prototype tool to help you label data, evaluate output, and optimize prompts.
AlignEval.com
October 31, 2024 at 2:11 AM
Evaluating LLM output is hard. For many teams, it's the bottleneck to scaling AI-powered product.
A key mistake is defining eval criteria w/o actually LOOKING AT THE DATA. This leads to irrelevant / unrealistic criteria + lots of wasted effort.
Thus I built AlignEval.com
A key mistake is defining eval criteria w/o actually LOOKING AT THE DATA. This leads to irrelevant / unrealistic criteria + lots of wasted effort.
Thus I built AlignEval.com
Reposted by Daniel Petrini
