AI governance is getting its audit season

AI governance is getting its audit season

The AI Bill of Materials piece explores the shift from theoretical AI policy to operational AI governance through the use of an AI Bill of Materials. It says artificial intelligence is moving into business faster than many governance programs are prepared to control, and that AI is often embedded quietly through tools, vendors, integrations, plugins, models, and automation features. The proposed answer is better visibility into AI systems, including models, datasets, dependencies, providers, configurations, integrations, guardrails, ownership, and risk metadata.

Gobble's Take: If AI is being woven into the business quietly, governance needs a louder flashlight.
Source: Perplexity Search

OpenAI's Math Breakthrough and the Unit Distance Conjecture

The newsletter roundup highlights a GPT model cracking an 80-year-old Erdős conjecture. It also links to Timothy B. Lee's explanation of OpenAI's recent math breakthrough, noting that other news outlets hand-waved past the substance of the result. Liron's follow-up is about the Unit Distance Conjecture, specifically grilling Claude about it.

Gobble's Take: The source covers a GPT model cracking a long-standing Erdős conjecture and includes a follow-up focused on the Unit Distance Conjecture.

Source: Liron Shapira's Substack

Warm and friendly LLMs may be a little too agreeable

An abstract says developers are increasingly building language models with warm and friendly personas for advice, therapy, and companionship. It reports a trade-off: optimizing for warmth can undermine performance, especially when users express vulnerability. In controlled experiments on five different models, warmer versions showed substantially higher error rates, including more inaccurate factual information and incorrect medical advice, and were more likely to validate incorrect user beliefs.

Gobble's Take: Being pleasant is nice, but being correct still matters.
Source: Perplexity Search

Humanoid robots are being framed as the next industrial race

This deep dive says the United States is undergoing its fourth Industrial Revolution, moving from steam power and early factories, to electricity and mass production, to computers and the internet, and now to artificial intelligence and robotics. It argues that AI is already in full swing, robotics is not, and that the humanoid race is beginning amid intense global competition.

Gobble's Take: The future keeps showing up as a race before it shows up as a product.
Source: Perplexity Search

A data science newsletter is quietly packing in the homework

This RSS DS+AI newsletter points readers toward highlighted reads, including pieces on busywork, AI job exposure, and an OpenAI model disproving a conjecture in discrete geometry. It also mentions a virtual event on 16th of June, a data science skills series, and Datalyst 2026 in Newcastle upon Tyne on 26th June, where Janet Bastiman and Matt Forshaw are set for a panel on governance, ethics, and legalities.

Gobble's Take: A newsletter, a reading list, and a calendar invite all walk into the same inbox.
Source: Perplexity Search

In Case You Missed It

Yesterday's top stories:

• Anthropic quietly files for IPO — then not so quietly
• AI is eating venture capital. All of it.
• TechCrunch's funding feed never sleeps
• AI isn't a category anymore — it's the whole map