ChatGPT beyond novelty: how people are putting it to work
OpenAI’s global usage breakdown shows how ChatGPT adoption is spreading in real-world patterns, not just curiosity sessions. The report highlights country-level shifts in how people prompt, iterate, and rely on the tool as everyday workflows evolve.
OpenAI is giving ChatGPT free users unlimited text chats
The Verge reports that OpenAI is removing rate limits for free and “Go” tier users for text-only conversations. A new “Think” button is also coming to boost reasoning on harder questions, while limits remain for uploads and images.
FinProBench: Evaluating Financial AI Agents with Role-Grounded Rubrics
FinProBench argues that judging financial agents requires more than task prompts – it needs tacit standards pulled from what real professionals actually deliver. The benchmark’s Role-Grounded Rubric Construction pipeline builds rubrics from practitioner deliverables and shows a major win for role-specialized work where “prompt-only” evaluation falls short.
Baseten on Hugging Face Inference Providers 🔥
Baseten expands how teams deploy and scale models by tying into Hugging Face’s inference provider ecosystem. The move is aimed at reducing friction from experimentation to production, especially when you need consistent performance across model backends.
MatrAIx: Simulating the World with 8.3 Billion Persona Agents
MatrAIx introduces a population-scale simulated-user testing setup built around Persona 8B – a dataset of 8.3 billion persona records. It’s designed to stress LLM-driven products across realistic decision behavior (hesitation, persistence after failure, latency tolerance) using thousands of trials across chat and app environments.
The left and right agree on one thing: no data centers
In a Verge Decoder episode, the reporting zeroes in on how “data center backlash” cuts across political identities – with localized environmental concerns turning into surprisingly bipartisan or populist coalitions. The discussion frames protests as community control fights over concrete infrastructure, not abstract AI policy.
FinPerMA: Event-Grounded Personalized-Memory Benchmark for LLM Agents
FinPerMA tests whether personalized-memory agents can actually update a user model when material events happen, using theory-informed event shocks and longitudinal checkpoints. The benchmark finds that even frontier LLMs and multiple memory setups lag far from saturation, with retrieval sometimes beating “purpose-built” memory when preference signals get diluted.
Improving GPT‑5.6 Sol in ChatGPT – expanding access to GPT-5.6 Luna for free users
OpenAI says it’s improving GPT-5.6 Sol with better accuracy and consistency, while also broadening access so free users can use more of the GPT-5.6 family. The announcement pairs quality upgrades with product-level expansion of “everyday” usage paths.
Suno shares plans to combat spammy AI music
Suno lays out new watermarking and fingerprinting measures aimed at making AI-generated tracks easier to identify and harder to mass-spam. The company also signals partnerships and “transparency tools” as it tries to align with emerging platform and regulatory expectations.
A Long-Run Persistence Theory for AI Systems under Redundancy-Adjusted AAS
This arXiv paper tackles a foundational question: can an AI system keep running through repeated cycles without accumulating unbounded structural “aging”? Using a redundancy-adjusted Artificial Age Score, it formalizes when aging stays bounded (even indefinitely) versus when persistence carries increasing burden – a framework built for analyzing long-run agent reliability.