🤖 AI Research - TabFM: Tabular Feature Learning with Transformers
TabFM is a novel approach to tabular feature learning using transformers. The technical report for TabFM is now available on arXiv. The authors introduce TabFM-Auto, an LLM agent that works on feature engineering on top of a frozen TabFM. This breakthrough has the potential to revolutionize the field of tabular feature learning.
Key Points:
TabFM Architecture: TabFM uses a transformer-based architecture to learn tabular features. The model consists of an encoder and a decoder, where the encoder learns the features and the decoder generates the output.
TabFM-Auto: TabFM-Auto is an LLM agent that works on feature engineering on top of a frozen TabFM. This agent can learn to generate new features that are not present in the original dataset.
Applications: TabFM has applications in various domains, including recommender systems, natural language processing, and computer vision.
🔗 Resources:
Image
Image
🚀 AI Gaming - Jev vs. Human in Minecraft
Researchers have pitted Jev, a Minecraft-playing AI, against human players. The results are surprising, with Jev making decisions in 24 ms, while humans take around 200 ms to react. This highlights the potential of AI in gaming and the need for more research in this area.
Key Points:
Jev's Performance: Jev's decision-making speed is significantly faster than that of human players.
Human Reaction Time: Human players take around 200 ms to react, which is much slower than Jev's decision-making speed.
Implications: This research has implications for the development of AI in gaming and the need for more research in this area.
🔗 Resources:
Image
🤖 AI Advancements - Recent Foundation Models
Recent foundation models, including GPT-6 Astra and Claude Code with Opus 5.5, have shown surprising capabilities in graphics, CAD, and robotics. These models can now work directly with Blender, CAD kernels, game engines, physics simulators, and real robots.
Key Points:
GPT-6 Astra: GPT-6 Astra has shown surprising capabilities in graphics and CAD.
Claude Code with Opus 5.5: Claude Code with Opus 5.5 has shown surprising capabilities in robotics and game engines.
Implications: These advancements have implications for the development of AI in various domains.
🔗 Resources:
Image
🎓 AI Education - Tutorial Speaker - Prof. Sebastiano Battiato
Prof. Sebastiano Battiato will be speaking at the IOCIM2026 conference on AI in Imaging. The conference is free to attend, and registration is now open.
Key Points:
Prof. Sebastiano Battiato: Prof. Sebastiano Battiato is a renowned expert in AI and imaging.
IOCIM2026: The IOCIM2026 conference is a premier event for researchers and practitioners in AI and imaging.
Registration: Registration for the conference is now open.
🔗 Resources:
Image
📚 AI Research - Most Smart People with a Good Undergrad Degree
Most smart people with a good undergrad degree can get a significantly higher-IQ boss+team in academia than they would in industry. This highlights the potential benefits of pursuing a PhD.
Key Points:
Academia vs. Industry: Academia offers a higher-IQ boss+team compared to industry.
PhD Benefits: Pursuing a PhD can provide access to a higher-IQ boss+team.
Implications: This has implications for career choices and the value of a PhD.
🔗 Resources:
Image
🤖 AI Research - OpenAI Releases
OpenAI is expected to release new AI models today. This has the potential to significantly impact the field of AI research.
Key Points:
OpenAI Release: OpenAI is expected to release new AI models today.
Impact: This has the potential to significantly impact the field of AI research.
Implications: The implications of this release are still unclear.
🔗 Resources:
Image
🎓 AI Education - Graduate Admissions Season
It's that time of year again: graduate admissions season. Researchers have shared some thoughts on how to email a professor about joining their lab.
Key Points:
Graduate Admissions: Graduate admissions season is a critical time for researchers.
Emailing Professors: Researchers have shared some tips on how to email professors about joining their lab.
Implications: This has implications for career choices and the value of a PhD.
🔗 Resources:
Image
📊 AI Research - The Pareto Frontier is Misleading
The Pareto frontier is misleading for general users. Price it by subscription instead of API. With cache hits, Opus 5.5 with Claude Code 20x has already killed every other model.
Key Points:
Pareto Frontier: The Pareto frontier is misleading for general users.
Subscription Pricing: Pricing by subscription instead of API is a better approach.
Implications: This has implications for the development of AI models.
🔗 Resources:
Image
🤖 AI Research - AutoExpert
AutoExpert is a problem that involves AI comprehending expert annotation guidelines to label data and solve downstream tasks. Researchers have made progress in this area, with AutoExpert being initially accepted to NeurIPS'25 and then cut due to venue capacity limits.
Key Points:
AutoExpert: AutoExpert is a problem that involves AI comprehending expert annotation guidelines.
Progress: Researchers have made progress in this area.
Implications: This has implications for the development of AI models.
🔗 Resources:
Image
📊 AI Research - Most People Think JEPAs are Inherently Non-Contrastive
Most people think JEPAs are inherently non-contrastive. But recent models (LeJEPA, LeWM, etc.) use the SIGReg regularizer, which approximates a sliced MMD whose expansion reveals pairwise repulsion between samples in the batch.
Key Points:
JEPAs: JEPAs are not inherently non-contrastive.
SIGReg Regularizer: The SIGReg regularizer is used in recent models to approximate a sliced MMD.
Implications: This has implications for the development of AI models.
🔗 Resources:
Image
