Funny thing: most #LLM components have been modified during the years except residual connections. We thought it was not necessary, but new research are showing the contrary. #AI #transformer #residuals medium.com/gitconnected...

#CodingAgents are the new frontier. Context files are used to help agents understand the codebase, and now is a must to have an AGENTS.md in your repository. What if this is hurting your #agent performance while increasing cost? I discuss that here levelup.gitconnected.com/context-file... #AI #LLM

Learning from the Crowd: Can One Agent Think Like Many? How Collective Intelligence Can Be Compressed into a Single Model medium.com/gitconnected... #AI #LLM #artificialintelligence #agents #multiagents

Funny thing, being rude with AI improves performance "Contrary to expectations, impolite prompts consistently outperformed polite ones, with accuracy ranging from 80.8% for Very Polite prompts to 84.8% for Very Rude prompts" arxiv.org/pdf/2510.04950

An interesting perspective by Nature www.nature.com/articles/d41...

#tabular data are everywhere (from medicine to finance). However, they have been neglected by the #AI revolution. In this review, we discuss why it is difficult to work with this data and which models to use (from #neuralnetwork to #LLM). www.techrxiv.org/users/961472...

Researchers evaluated 19 language models on over 76,000 participants and discovered that specialized post-training methods increased persuasiveness by up to 51%. But, the more persuasive the LLM, the less accurate is arxiv.org/pdf/2507.13919

@salvatoreraieli.bsky.social explores how meta-learning helps CNNs grasp abstract relationships like "same-different," bridging the gap between human and machine vision.

Can AI learn to think like us? @salvatoreraieli.bsky.social explores how meta-learning helps CNNs grasp abstract relationships like "same-different," bridging the gap between human and machine vision.

Can AI learn to think like us? @salvatoreraieli.bsky.social explores how meta-learning helps CNNs grasp abstract relationships like "same-different," bridging the gap between human and machine vision.

This survey classifies LLM reasoning by when it occurs (inference or training) and system type (standalone or agentic), highlighting trends like learning-to-reason and agent-based workflows, with methods like prompt tuning, output refinement, and PPO. arxiv.org/abs/2504.09037

Can AI learn to think like us? 🤔 This new article by @salvatoreraieli.bsky.social explores how meta-learning helps CNNs grasp abstract relationships like "same-different," bridging the gap between human and machine vision.

The Basis of Cognitive Complexity: Teaching CNNs to See Connections Transforming CNNs: From task-specific learning to abstract generalization towardsdatascience.com/the-basis-of... thanks @towardsdatascience.com to publish my latest article

Only the Beginning Matters: How the LLM Decides Where to Focus Attention Understanding How the First Token Shapes LLM’s Focus and Responses levelup.gitconnected.com/only-the-beg...

If you (like me) are brushing up on your #python plotting using #matplotlib or #seaborn, check out the python graph gallery by Yan Holtz: python-graph-gallery.com Being more versed in R, I've found it hard to find others using python (but the #PydyTuesday hash helps). 🙋🏻‍♀️ Say hi if that's you! #databs

Sadly realizing that the Italian expression "conosco i miei polli" ("I know who I'm dealing with", literal translation: "I know my chickens") cannot be used in English. is a pity because this expression conveys well how someone is expected to do something stupid

Thinking Machines: A Survey of LLM based Reasoning Strategies This survey offers an overview and comparison of current reasoning techniques and presents a systematic review of reasoning-imbued language models. arxiv.org/abs/2503.10814

"Now that all mouths are burned, no one should be satiated. Let all lips be withdrawn and one last knowing grin remain for all after all vanities have evaporated." Today on ergot.: 'Rye Dance' by Salvatore Raieli @salvatoreraieli.bsky.social www.ergot.press/authors/Salv...

Thank you @ergot.bsky.social for publishing my work

The Socratic AI: Knows Others, But Not Itself How Language Models Excel at Prediction but Fail at Introspection levelup.gitconnected.com/the-socratic...

5 Haikus of Artificial Intelligence The inner poetry of artificial intelligence medium.com/data-science...

Tomorrow on ergot.: Seven new Rye Dances from Salvatore Raieli @salvatoreraieli.bsky.social, a debut fiction publication.

TxAgent is an AI-driven system that assesses drug interactions, contraindications, and patient-specific data to create adaptive treatment plans. zitniklab.hms.harvard.edu/TxAgent/

Are Open Models Really Open? Many AI models labeled as "open" come with restrictive licensing terms. Google's new models and similar releases from Meta raise concerns about commercial limitations, which could affect smaller companies that depend on these technologies. techcrunch.com/2025/03/14/o...

As every week, the #ML and #AI news (latest articles for LLM, resources, news, and perspectives) for the last week (10-16 March). github.com/SalvatoreRa/...

Planet Definitions xkcd.com/3063

An interesting survey on how to make LLM more sustainable arxiv.org/abs/2412.04782

Notice to flat-Earthers all around the globe: Please alert the authorities if tonight’s Total Lunar Eclipse looks like this:

The TechCrunch AI glossary Cool resources to learn about some common generative AI terms techcrunch.com/2025/03/02/t...

An intriguing review about LLM and medicine arxiv.org/abs/2502.17132

Is Wikipedia an Endangered Species? Is ChatGPT Its Predator? Exploring the Impact of Large Language Models on Wikipedia’s Future levelup.gitconnected.com/is-wikipedia...

As every week, a collection of the best news on #AI and #ML. Here (github.com/SalvatoreRa/...), you can find the news for the week of 3-9 March.

A Malicious Seed: Fine-Tuning #LLMs to Grow Bad Behaviors How Narrow Training on Insecure Code Leads to Broad #AI Misalignment and Deception medium.com/gitconnected...

Well, why am I not surprised? More generally, benchmarks for AI should be rethought. A simple number does not show how well a model really performs and what its capabilities are techcrunch.com/2025/02/22/d...

A cool application of AI www.eurekalert.org/news-release...

A cool survey about Protein-specific large language models (Protein LLMs), models that can predict the structure of the proteins, their function or even design them arxiv.org/abs/2502.17504

As every week, the most important news about #artificialintelligence, plus the most important articles, repositories, perspectives, and much more Yoy can find here for the last week: github.com/SalvatoreRa/...

As every week collection of the the best #ML and #AI news every week (research, news, resources, perspectives). For the week 17-23 February github.com/SalvatoreRa/...

And so we set out to understand _feedforward_ graphs (i.e. graphs w/o back edges) ⏩ Turns out these graphs are rather understudied for how often they are used in practice! 😮 We work towards a framework to analyse them, which I hope both my GNN and my LLM friends will enjoy 🚀

As many areas, science itself will be transformed by AI. We have now released a state-of-the-art survey on this emerging field. We discuss methods and tools for AI-enhanced search, ideation, experimentation, content generation and peer review. arxiv.org/pdf/2502.05151

www.unite.ai/transformers... #AI #artificialintelligence

When Elon was a kid, survival for ALL, the most common childhood cancer, was <10% Now it’s >90% due to funded collaboratives (high indirect costs) whose work ultimately SAVES THOUSANDS OF LIVES EVERY YEAR. For other cancers there’s still so much progress to be made. I think it’s worth it.

As every Sunday, a collection of the most important AI and ML news of the week 3-9 Feb. They are grouped into the latest research, news, resources, and perspectives #AI #ML #artificialintelligence #machinelearning #LLM you can find here, together with previous weeks: github.com/SalvatoreRa/...

You: "We must build robots to automated human labor" Apple: "We must build the pixar lamp" machinelearning.apple.com/research/ele...