Prompt Notes #13 - weekly AI digest
AI goes fast, helping you (and myself) catching up.
Welcome to the 13th edition of Prompt Notes, your new favourite weekly AI digest.
This edition will cover the week from 15 June to 21 June.
Note: the newsletter will be taking a break next week.
What caught my attention
I decided I’m not talking about Anthropic or OpenAI this week. You have no idea how hard that is. Every other headline has one of their names in it. The US government cracked down on Anthropic’s newest models, OpenAI tripled its revenue while burning through $34 billion, and Noam Shazeer jumped from Google to OpenAI. That’s three lead stories I’m leaving on the table.
But if I cover them again, this newsletter becomes a two-company recap with a fancy header. So here we are. The biggest AI news week of the month, and I’m exercising restraint.
Consider it character development.
TLDR
ChatGPT’s market share fell below 50% for the first time while still leading with 1.1 billion monthly users — Gemini and Claude are closing the gap. (TechCrunch AI)
Samsung deployed ChatGPT Enterprise and Codex to employees worldwide — one of OpenAI’s largest enterprise rollouts to date. (OpenAI Blog)
Adobe is rolling out “creative agent” functionality across Photoshop, Premiere, and other Creative Cloud apps — users describe the task, the software executes it. (The Decoder)
Yann LeCun warned that AI labs like OpenAI and Anthropic face a “big bubble explosion” because operations are investor-subsidized and costs aren’t declining fast enough. (The Decoder)
Kuaishou’s Kling AI video unit is in talks to raise $2 billion at an $18 billion valuation led by General Atlantic. (Bloomberg Technology)
Full roundup
Models & Research
At a Stanford talk, Sam Altman defended LLM scaling and cited OpenAI’s recent disproof of a mathematical conjecture as evidence that earlier researchers underestimated what scaling could do. (The Decoder)
OpenAI researchers showed that reinforcement learning on small doses of desired behavioral traits such as truthfulness and corrigibility improved performance on 44 of 53 benchmarks. (The Decoder)
Z.ai’s GLM-5.2 became the top open frontend coding model according to recent evaluations. (Latent Space)
Cohere released North Mini Code, a model offering developers more transparency and control than frontier models from Anthropic and OpenAI. (AI Business)
Products & Tools
Samsung Electronics deployed ChatGPT Enterprise and Codex to employees worldwide, marking one of OpenAI’s largest enterprise AI rollouts. (OpenAI Blog)
Google added a Gemini-powered AI assistant to Ad Manager that provides personalized answers to campaign-related questions. (Social Media Today)
OpenAI released “Record & Replay” for Codex on macOS, converting a demonstrated workflow into a reusable skill. (The Decoder)
AWS launched Continuum, which detects and fixes code vulnerabilities, and Context, which builds a knowledge graph from corporate data to give AI agents business context. (The Decoder)
Adobe is rolling out “creative agent” functionality across Photoshop, Premiere, and other Creative Cloud apps that lets users describe tasks for the software to execute. (The Decoder)
Google released Android 17 and Wear OS 7 with expanded Gemini features on Pixel devices. (TechCrunch AI)
ChatGPT opened self-serve advertising access to all advertisers after previously requiring invitations. (Neil Patel)
Databricks introduced Lake Transactional/Analytical Processing architecture to support AI agents accessing operational and analytics workloads. (SiliconANGLE AI)
Databricks unveiled CustomerLake, an agentic customer data platform, at its Data + AI Summit. (MarTech)
Google Cloud placed agentic AI at the center of its enterprise strategy. (AI Business)
TechCrunch published instructions for disabling Gemini AI features and “write with Gemini” pop-ups in Google Docs. (TechCrunch AI)
Infrastructure & Compute
Schneider Electric and Foxconn formed a partnership to deliver integrated AI data centers combining compute, power, and cooling. (Capacity Global)
The DOJ argued that xAI’s unpermitted gas turbines are required for national, economic, and energy security. (TechCrunch AI)
Regulation & Policy
Anthropic received a June 12 export-control order to block foreign access to Fable 5 and Mythos 5 and responded by disabling the models globally. (The Verge AI)
The White House told Anthropic that Fable 5 cannot be rereleased unless its guardrails are made impossible to circumvent, a requirement security experts say cannot be met. (Wired AI)
SK Telecom’s access to Anthropic’s Claude Mythos via Project Glasswing was halted after US officials cited the South Korean firm’s alleged ties to China. (The Decoder)
The European Commission is evaluating the global impact of the US order that forced Anthropic to shut down Fable 5 and Mythos 5. (The Decoder)
Dozens of cybersecurity experts urged the White House to lift export-control restrictions on Anthropic’s Fable and Mythos models to avoid limiting defenders’ capabilities. (TechCrunch AI)
Anthropic’s emphasis on safety commitments has allowed it to prioritize business interests while challenging US government directives. (Stratechery)
Anthropic stated that its Mythos model is so capable at finding software vulnerabilities that it was released only to 200 partner organizations. (Bloomberg Technology)
Big Tech lobbyists continued pushing for federal AI legislation that would preempt state-level rules. (The Verge AI)
Business & Funding
OpenAI reported $5.7 billion in Q1 2026 revenue and $3.7 billion in operating costs, both triple the prior-year figures, with stock-based compensation exceeding $2.3 billion. (The Decoder)
OpenAI spent $34 billion over the past year, significantly more than the prior period. (The Decoder)
Kuaishou Technology is in talks with General Atlantic to raise $2 billion for its Kling AI video unit at an $18 billion valuation. (Bloomberg Technology)
ChatGPT’s market share fell below 50% for the first time while still leading with 1.1 billion monthly users, followed by Gemini and Claude. (TechCrunch AI)
Cohere reported a surge in inbound interest from governments and investors seeking non-US models after the Anthropic restrictions. (Bloomberg Technology)
Investors anticipate a wide-scale liquidity event from SpaceX’s IPO that will benefit private AI companies preparing to go public. (Bloomberg Technology)
Yann LeCun warned that AI labs such as OpenAI and Anthropic face a “big bubble explosion” because operations are subsidized by investors and costs are not declining fast enough. (The Decoder)
Workers at South Korean tech firms received bonuses worth millions of won, prompting the Bank of Korea to warn of upward pressure on inflation. (CNBC Tech)
Microsoft Azure Core CTO Marcus Fontoura is joining Anthropic to help build next-generation AI systems. (Data Center Dynamics)
Google researcher Noam Shazeer is joining OpenAI. (Bloomberg Technology)
Amazon MGM Studios dropped the nearly finished OpenAI drama film “Artificial” after signing a $50 billion partnership with OpenAI. (The Decoder)
Society & Culture
A UC Berkeley study of more than 500,000 grades found that courses heavy on writing and coding saw grades jump after ChatGPT launched, with the effect appearing mainly in homework. (The Decoder)
Signal’s Meredith Whittaker stated that AI chatbots “are not your friends” and are not conscious or sentient beings. (TechCrunch AI)
Wired published 28 tips for engineering more effective ChatGPT prompts. (Wired AI)
Ahrefs analysis of 137,000 domains found that 97% of llms.txt files received zero requests from AI retrieval bots. (SEJ Generative AI)
On the Odd Lots podcast, Anthropic co-founder Jack Clark and the company’s top economist discussed AI’s labor-market impacts and societal effects. (Bloomberg Technology)