TLDR AI 2025-06-04

Read the Delve case study.

NOTEBOOKLM NOW SUPPORTS PUBLIC SHARING

Google's NotebookLM now lets users share notebooks publicly via links. Viewers can interact with AI-generated summaries and questions while source content remains read-only.

YOSHUA BENGIO'S AI SAFETY LAB

Yoshua Bengio launched LawZero, a nonprofit AI safety lab backed by $30M in funding from major philanthropic donors.

GITHUB LAUNCHES COPILOT SPACES

Spaces allows developers to curate code, documentation, and custom instructions for Copilot, turning it into a subject matter expert that can be shared across organizations. Files and repositories added to Spaces automatically stay updated as code changes.

EFFICIENT ONLINE LEARNING WITH TRL AND VLLM

Hugging Face integrated vLLM directly into TRL to reduce inefficiencies in training with GRPO, an online learning algorithm.

MY AI SKEPTIC FRIENDS ARE ALL NUTS

A veteran developer rails against the many highly-skilled programmers still rejecting LLMs based on experiences with early chatbots, missing how modern coding agents autonomously navigate codebases, run tests, and iterate on failures. He dismisses common objections: developers already review all code before merging regardless of source, and hallucinations become irrelevant when agents can automatically compile, catch errors, and retry until tests pass. LLMs might displace developers but software engineers similarly automated away administrative roles like travel agents, record store clerks, and countless other professions.

WHY I HAVE SLIGHTLY LONGER TIMELINES THAN SOME OF MY GUESTS

AI progress over the last decade has been driven by scaling training compute of frontier systems. This can't continue beyond this decade. After 2030, AI progress has to mostly come from algorithmic progress. However, since the low-hanging fruit will have already been plucked, the yearly probability of AGI craters.

DNA REPRESENTATION LEARNING

SPACE is a supervised learning method for genomic profile prediction that uses a Mixture of Experts model.

DIFFUSION-BASED MEDICAL TREATMENTS

DIME is a diffusion-based model designed to estimate the joint distribution of interdependent medical treatment outcomes.

JIGSAWSTACK LAUNCHES OPEN-SOURCE DEEP RESEARCH TOOL

The framework orchestrates LLMs, recursive web searches, and structured reasoning to generate reports that would take a human hours or days to complete. JigsawStack offers control over research depth, breadth, model selection, and output formatting while maintaining strong citation transparency.

VIBE-CODING IDEAS TO GIVE STARTUP GTM TEAMS AN EDGE

A startup advisor demonstrates building a polished ROI calculator for a manufacturing SaaS company in under two hours using Bolt.new, transforming a spreadsheet into an interactive tool that frames value for executives. Other examples include conference scraping tools, meeting prep dashboards, and feature prototyping that previously required engineering teams or expensive agencies but now cost around $70 in monthly subscriptions, arguing this enables non-technical teams to prove value and move faster than competitors.

WHEN WILL WE PAY A PREMIUM FOR AI LABOR?

AI agents often outperform humans at a fraction of the cost, not yet commanding premium prices due to ongoing technical evolution and perceived risk. Waymo, achieving significant safety improvements, remains cheaper than other options, reflecting a common pricing strategy across startups. However, in scenarios where AI's constant vigilance and processing power are crucial, premium pricing might emerge.

PREDICTING AND EXPLAINING AI MODEL PERFORMANCE: A NEW APPROACH TO EVALUATION

Microsoft researchers developed ADeLe, a framework that predicts and explains AI model performance on new tasks by assessing them across 18 cognitive and knowledge-based scales. ADeLe uncovered limitations in current benchmarks and created detailed ability profiles for various LLMs, highlighting discrepancies in strengths, weaknesses, and specific abilities. The framework showed an 88% accuracy in predicting AI success, promising improvements in AI evaluation, policymaking, and deployments.

OPENAI'S VULNERABILITY REPORTING

OpenAI introduced a policy for coordinated disclosure of third-party software vulnerabilities found by its AI systems.

AMAZON MGM FAST-TRACKS FILM ABOUT SAM ALTMAN'S DRAMATIC OPENAI FIRING AND REHIRING

The studio is eyeing Andrew Garfield to play Altman, with Monica Barbaro (“A Complete Unknown”) as CTO Mira Murati and Yura Borisov (“Anora”) as co-founder Ilya Sutskever.