What we're reading, in the order we're reading it.
Anthony and Harry's working stream — links, charts, tweets, and short takes. The unprocessed inputs behind the briefings.
Chinese researchers unveil MemOS, the first ‘memory operating system’ that gives AI human-like recall
Researchers unveil MemOS, a breakthrough “memory operating system” for AI that delivers 159% improvement in reasoning tasks and enables persistent memory across sessions.
Supabase MCP can leak your entire SQL database
In this post, we show how an attacker can exploit Supabase’s MCP integration to leak a developer’s private SQL tables. Model Context Protocol (MCP) has emerged as a standard way...
The AI Civil War Is Here
The tech industry and its critics occupy parallel universes.
Supabase MCP can leak your entire SQL database
Here’s yet another example of a lethal trifecta attack, where an LLM system combines access to private data, exposure to potentially malicious instructions and a mechanism to communicate data back...
Mistral in Talks With MGX, Others to Raise Up to $1 Billion
French artificial intelligence startup Mistral AI is in talks to raise as much as $1 billion in equity from several investors including Abu Dhabi fund MGX, according to people familiar...
Impostor uses AI to impersonate Marco Rubio and contact foreign and U.S. officials
The recipients of the scam messages, which were sent by text, Signal and voice mail, were not identified in the cable, a copy of which was shared with the Associated...
Nvidia AI Breakthrough Tackles Encyclopedia-Sized AI Questions
Nvidia researchers have unveiled AI technology that could revolutionize how large language models handle massive amounts of information while maintaining real-time responsiveness.
Nvidia Dethrones Apple as World’s Most Valuable Corporation
Nvidia Corporation has become the most valuable company in history. Shares in the high-end AI chip developer rose as much as 2.4% to $160.98 in trading on July 3, pushing...
SmolLM3: smol, multilingual, long-context reasoner
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Zuckerberg’s Meta Superintelligence Labs poaches top AI talent in Silicon Valley
Meta Platforms is racing to secure top artificial intelligence talent for its newly created Superintelligence Labs to better compete with rivals including OpenAI, Google and Anthropic.
Meta Poaches Apple’s Top AI Exec: Where Does This Leave Apple Intelligence?
Another setback for the iPhone maker could put its AI future in jeopardy, or push it even closer to using models from OpenAI or Anthropic instead of developing its own.
LLMs are Capable of Misaligned Behavior Under Explicit Prohibition and Surveillance
Abstract In this paper, LLMs are tasked with completing an impossible quiz, while they are in a sandbox, monitored, told about these measures and ins…
Will we know if the next plague is human-made?
AI could make bioweapons more dangerous. It could also be the key to stopping them.
Subversion via Focal Points: Investigating Collusion in LLM Monitoring
I released a new paper on collusion and Schelling coordination between language models: “Subversion via Focal Points: Investigating Collusion in LLM…
Automation Anywhere buys its own AI agent vision
The process automation vendor deploys more than 40 agents across the company to save time and money in its finance, tech support, and other functions.
Study could lead to LLMs that are better at complex reasoning
To improve adaptability of large language models to challenging tasks that require reasoning, MIT researchers found strategically applying a method known as test-time training with task-specific examples can boost the...
AI Safety at the Frontier: Paper Highlights, June ’25
tl;dr Paper of the month: • Emergent misalignment arises across many models when training on incorrect data and is largely driven by a single “toxic…
Palantir’s +79% vs Bill’s -43%: The AI Divide That Defined SaaS Performance in 1H’25
SaaS / B2B Public Stock Performance 1H’25: Top Winners & Losers 📊 Market Summary: SaaS Sector in 1H’25 The SaaS sector experienced a dramatic bifurcation in the first half of...
GitHub – golioth/tinymcp: Let LLMs control embedded devices via the Model Context Protocol.
Let LLMs control embedded devices via the Model Context Protocol. – golioth/tinymcp
Will AI-Powered Cameras Improve Road Safety?
AI cameras are transforming intersections into safer spaces. How are they changing driver behavior without invading privacy?
Economics of Claude 3 Inference
Anthropic has recently announced the deprecation of API access to Claude 3 Opus. The model will still be available for claude.ai subscribers and thro…
Apple’s newest AI study unlocks street view for blind users
SceneScout, combines Apple Maps with a multimodal LLM to provide interactive, AI-generated descriptions of street view images.
Will A.I. Replace New Hires or Middle Managers?
Amid layoffs at Microsoft and other large tech companies, experts are debating whose jobs are most likely to be spared.
It’s time to retire the ticket: An IT roadmap for agentic AI
Stop paying the innovation tax of legacy IT tickets; the future is agentic AI that solves problems before they are even reported.
OpenAI, Oracle Sign $30B Cloud Deal to Fuel AI Project
The deal is part of OpenAI’s Stargate project to build out AI infrastructure in the U.S.
How a big shift in training LLMs led to a capability explosion
Reinforcement learning, explained with a minimum of math and jargon.
Shining A Light On Shadow AI
Shadow AI (like most forms of shadow technology and bring your own device activity) is viewed as a negative, an infringement and a risk.
GitHub – MrLesk/Backlog.md: Backlog.md – A tool for managing project collaboration between humans and AI Agents in a git ecosystem
Backlog.md – A tool for managing project collaboration between humans and AI Agents in a git ecosystem – MrLesk/Backlog.md
Forget the hype — real AI agents solve bounded problems, not open-world fantasies
Event-driven multi-agent systems are a practical architecture for working with imperfect tools in a structured way.
ChatGPT is pushing people towards mania, psychosis and death
Record numbers of people are turning to AI chatbots for therapy, reports Anthony Cuthbertson. But recent incidents have uncovered some deeply worrying blindspots of a technology out of control
New Paper: It is time to move on from MCQs for LLM Evaluations
New paper: Answer Matching Outperforms Multiple Choice for Language Model Evaluations. …
Google Hit With EU Antitrust Lawsuit Over AI Summaries
A coalition of independent publishers alleged that Google’s tool has caused ‘significant harm to publishers,’ impacting their traffic, readership and revenue.
Large Language Model Performance Doubles Every 7 Months
By 2030, AI will greatly outperform humans in some complex intellectual tasks. Discover how LLMs are doubling their capabilities every seven months.
How AI May Treat Chronic Pain Without Medication
USC and UCLA researchers have created a wireless device that uses AI to decode pain levels from brain activity and customize spinal cord stimulation to treats chronic pain.
vLLM: Easy, Fast, and Cheap LLM Serving with PagedAttention
GitHub | Documentation | Paper
Early Signs of Steganographic Capabilities in Frontier LLMs
One key strategy for preventing bad outcomes from misuse or misalignment is model monitoring. However, one way that monitoring can fail is if LLMs us…
Apple just released an interesting coding language model
Apple has just released an AI model that, rather than generating code from left to right, does it out of order and all at once. Here’s how.
Mechanical engineer develops AI-generated digital masks to restore damaged paintings
Alex Kachkine, a PhD student at MIT, art-lover and self-taught restorer, wanted to provide conservators with another “tool in the toolbox”