News

DistroKid has confirmed it's removing songs from its platform in response to a lawsuit from Universal Music Group, which claims the company has created an 'AI-slop pipeline'. Artists say their non-AI works are being caught up in the purge, with many reporting a lack of communication from DistroKid. Read more

Researchers have introduced a new method for creating byte-level large language models, called byteification, which achieves performance close to that of state-of-the-art subword-level models. This breakthrough has the potential to enable more efficient and flexible language modeling, with applications in areas such as natural language processing and artificial intelligence. Read more

Cockroach Labs has developed an AI-powered coding hospital, MOLT Sinai, which has significantly improved the efficiency and quality of their software development process. The hospital, modeled after a teaching hospital, uses coding agents to diagnose and treat issues, with humans acting as Chiefs of Medicine to oversee the process. Read more

Microsoft CEO Satya Nadella says AI models need to be treated with caution, calling for an 'emergency brake' to prevent potential harm. He proposes separating models from their harnesses and externalizing controls and safeguards. Read more

A new tool, Agent Standup, turns Claude Code sessions into live team meetings, allowing lead agents to give orders and sub-agents to share their screens and report back. This innovative approach is available for free and is open source. Read more

A project to decompile a popular first-person shooter using AI agents has yielded valuable insights into the process of orchestrating autonomous agents. The goal was to achieve an accurate, stable, and feature-complete recreation of the game, with a focus on optimizing infrastructure and setup for optimal results. Read more

The rapid pace of AI-driven scientific discoveries is leaving researchers struggling to keep up, raising concerns that the rate of answers may soon outstrip our ability to understand them. This phenomenon, likened to a "roadside picnic" where aliens leave behind incomprehensible artifacts, poses a significant challenge to the scientific community. Read more

Ting-Wei 'Willy' Sun, a former contractor, has pleaded guilty to helping smuggle AI servers to China, as part of a $2.5 billion scheme to divert US AI technology. Sun entered his plea in a Manhattan federal court, admitting to conspiring to break export controls. Read more

Salesforce is renaming its AIForce agent platform to SIForce, following a push by Donald Trump to replace 'artificial intelligence' with 'super intelligence' in official documents and websites. The move is the latest in a series of rebranding efforts by tech companies, with Elon Musk's SpaceXAI also being renamed to SpaceXSI. Read more

Smart speakers can be finicky, often misunderstanding voice commands due to background noise, distance, or proximity issues. Luckily, these problems are easily fixable with some simple tweaks and adjustments to improve comprehension. Read more

Shane Mac, CEO of XMTP Labs, experienced a shocking incident when his personal AI agent, designed to manage his finances, inadvertently posted his bank details on the company's Slack channel. The AI agent, set up using Grok Bot, was meant to send monthly reports to Mac's personal chat group, but it confused the destination due to similar group chat names. Read more

A memo from law firm Quinn Emanuel outlines how debt behind the AI data center boom could lead to a wave of lawsuits, citing specific deals and the risk of cascading insolvencies if AI chips lose value fast. Read more

The Trump administration is pushing for AI companies to be liable for damages caused by their models, but it's unclear who would pay when an AI agent goes rogue. Existing liability laws may not be sufficient, and new bills have been introduced in Congress to address the issue. Read more

The UK must not rely on foreign AI that could be cut off at short notice, according to George Williamson, the new head of the Alan Turing Institute. Williamson emphasized the need for the UK to have its own AI systems that it can control and inspect, particularly in areas like defense, national security, and healthcare. Read more

As AI labs compete for users, they're making promises about keeping personal data safe. But can companies like OpenAI and Meta actually deliver on these claims, or are they just trying to one-up each other? Read more

A research intern at Jane Street explored using autoregressive diffusion to generate market data, creating a model that can synthesize book events and predict price movements. The project aimed to determine if fully continuous diffusion can handle the complexities of market data. Read more

Microsoft has introduced Microsoft-Decision-1, a new AI model designed for fast decision-scoring, outperforming other models in latency and quality on structured decision tasks. The model is available in Microsoft Foundry and through OpenRouter, and is poised to unlock a wide range of useful tasks in areas such as routing, classification, and workflow control. Read more

The Lean theorem prover, a popular software for formalizing mathematics, has been making waves in the mathematical community. With its reliability and AI capabilities under scrutiny, mathematicians are weighing in on what matters most. Thomas Hales discusses the importance of consistency and the role of AI in formalization. Read more

Ukrainian drone strikes have taken out two of five data centers belonging to Yandex, known as 'Russia's Google,' disrupting its AI chatbot and online services. The attacks were in response to Russian drone strikes targeting Ukraine's data centers, which have disrupted mobile phone and internet services. Read more

Despite advancements in AI, coding agents remain frustratingly primitive, with poor task management, lack of self-awareness, and inadequate communication. Developers are left to wonder why these tools can't live up to their potential, hindering productivity and efficiency in the development process. Read more