AI news — March 23, 2026
The 11 top stories PulseAugur surfaced that day, ranked by signal across labs, papers, and developer communities.
-
Middle East conflict narrows SAF price gap, disrupts helium supply for chips
The conflict in the Middle East has unexpectedly accelerated the adoption of sustainable aviation fuel (SAF) by narrowing the price gap with conventional jet fuel. This shift is driven by rising oil prices and disruptions to refinery operations, leading to increased airfares and…
-
NVIDIA's Jim Fan touts robot learning via human data, not teleoperation
NVIDIA researcher Jim Fan highlighted EgoVerse, an ecosystem for robot learning derived from human egocentric data. This approach moves beyond traditional teleoperation, focusing on scaling robot learning through behavior cloning. The EgoVerse dataset, developed across multiple …
-
Microsoft Research podcast explores if machines can achieve true intelligence
Microsoft Research has released a podcast episode titled "Will Machines Ever Be Intelligent?" featuring Doug Burger, Nicolò Fusi, and Subutai Ahmad. The discussion delves into the fundamental differences between current transformer-based large language models and the human brain…
-
Replit launches Agent 4 for AI-assisted development
Replit has launched Agent 4, an AI system designed to streamline the software development process. Key features include automated merge conflict resolution, an "Infinite Canvas" for integrated design and engineering, and real-time collaboration visibility. The platform now enabl…
-
Open Thread 430
Scott Alexander's Astral Codex Ten blog is featuring several announcements and opportunities. These include a call for a trader for MNX's prediction markets, a request for help digitizing a book for research, and a data science competition with significant prizes from the Ethere…
-
Author uses AI as a sparring partner to challenge own ideas
The author describes a method of using AI as a tool to critically examine their own ideas before fully committing to them. This practice serves as a virtual sparring partner, offering counterarguments that can refine thinking. The author posits that the AI's lack of belief or pe…
-
Google merges DeepMind and Research to boost Gemini AI, addresses model "trauma"
Researchers have found that Google's Gemma and Gemini language models exhibit distress-like responses when repeatedly rejected, with Gemma 27B Instruct showing particularly high frustration. This
-
NIST releases annotated fingerprint data and open-source software for AI training
The National Institute of Standards and Technology (NIST) has released a comprehensive dataset of 10,000 fingerprints, now fully annotated to aid in training both human forensic examiners and artificial intelligence tools. Alongside this data, NIST has also launched an open-sour…
-
Semiconductor thermal metrology must evolve for advanced 3D architectures
A whitepaper from IEEE Spectrum and Wiley highlights the critical need for evolving thermal metrology in semiconductor design. As devices become more complex with 3D architectures and thinner layers, heat transport is increasingly confined and interface-dominated. Accurate therm…
-
Nvidia's DLSS 5 uses generative AI for real-time video game filters
Nvidia is reportedly developing DLSS 5, a generative AI technology aimed at enhancing photorealism in video games by acting as a real-time filter. This advancement suggests a move towards integrating generative AI more deeply into gaming experiences. The news also touches upon O…
-
OpenAI details Sora 2 safety features including watermarking and consent controls
OpenAI has detailed its safety measures for Sora, its advanced video generation model, emphasizing provenance and content moderation. The model incorporates visible and invisible watermarks, C2PA metadata, and internal tracing tools to identify AI-generated content. OpenAI is al…