TLDR AI 2025-06-04

Delve ran 4 placements across 2 TLDR newsletters and brought in 66 leads, with a meaningful chunk from enterprise. The result: $1M in attributed pipeline and a 52x ROI.

The case study breaks down their strategy, results, and example ads.

Read the Delve case study.

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

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.

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.

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.

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

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.

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

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

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.

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.

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.

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.

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.

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