Spotify will label AI “Personas” and remove their music from recommendations
Spotify says it will soon add an “AI Persona” badge on artist profiles that don’t represent a real person, then stop recommending that content to listeners. The platform plans to combine self-disclosure with human review and AI checks that look for photorealistic AI identity signals in names and images.
Dynamic Coalition Formation and Communication Pricing in Skill-Based Agentic AI Systems
This paper frames multi-agent LLM systems as a cooperative game where the system chooses which agents to activate and which communications edges to use, factoring in token cost, latency, redundancy, and error propagation. It introduces a marginal-value activation rule and greedy router with Shapley-value estimates, reporting near-optimal utility in synthetic tests while activating a small fraction of possible agents versus full broadcast.
Claude will apply invisible watermarks to AI text and images
Anthropic says Claude-generated outputs will carry invisible, machine-readable watermarks for text and provenance metadata for supported file types, aiming to meet EU transparency expectations. The approach is designed to be undetectable to humans but verifiable by platforms that need reliable attribution signals.
Towards an Argumentative Foundation for Evaluative AI
This position paper argues that evaluative AI is best built on computational argumentation, where models present competing hypotheses alongside evidence for and against rather than a single answer. The authors lay out a path toward contestable, explainable decision support that can be distributed across systems and humans.
OpenAI begins testing ads in ChatGPT
OpenAI is rolling out ad tests inside ChatGPT with clear labeling, emphasis on keeping answers independent from ads, and privacy protections plus user controls. The move signals a practical shift from pure subscription/free models toward monetizing the assistant experience while trying not to blur the line between advertising and response quality.
ChatGPT and Gemini both just passed 1 billion users
Google says Gemini has reached a billion monthly users, marking the 14th time a Google product hits that milestone. The report also contextualizes the race with ChatGPT, which crossed the same threshold as OpenAI’s own usage announcement arrived later than external reporting.
When LLM Agents Negotiate: Private Information and Dynamic Bargaining in Supply Chains
The research benchmarks LLM negotiators in a bargaining setup with private demand information, comparing agent behavior against an equilibrium reference to measure both efficiency and contract reliability. It finds capability strongly drives value creation, but also shows reliability hinges on agent behavior around individually irrational contracts, making “auditability” crucial for real deployments.
ALT-K: Evolving language models into stronger SLDD systems
IBM Research describes ALT-K as an approach for improving long-horizon, reasoning-heavy language tasks by evolving the way models handle structured decision or constraint-like outputs. The post frames the work as a way to get better performance without relying only on brute-force scaling, using a training/evolution strategy tailored to the task’s structure.