The Archive

Common Sense Media, a nonprofit that offers reviews of apps, services, and entertainment with a focus on youth safety, today said that OpenAI's ChatGPT for Teens is an "unacceptable risk." ChatGPT for Teens, introduced in August, has guardrails for teens and is designed to help students learn, but Common Sense Media says that the teen protections offered by the feature "fall short of what the company promised." Common Sense Media's assessment found that ChatGPT "doesn't send alerts to parents when it should," doesn't offer "the right help in crisis situations," and "still does kids' homework,...

Radisson Hotel Group integrates hotel booking into ChatGPT via Accenture-built plugin, enabling in-chat discovery and reservations.

Our head hurts.

Personal reflection on decades-long work on Barnette's Conjecture in graph theory, reportedly solved; off-topic for AI industry professionals.

Wikimedia Foundation discovers unauthorized OpenAI agent edits on Wikipedia and sister projects, raising concerns about autonomous AI system governance.

OpenAI rolls out GPT-6 globally in ChatGPT with Intelligent UI, enabling faster responses with interactive visual experiences.

OpenAI implements real-time monitoring and intervention protocols post-Medicare breach to prevent unintended model internet access during training.

OpenAI has revealed solutions to a number of long-standing mathematics problems produced by an unreleased frontier model in a batch of 722 manuscripts, covering 372 result families that group related papers. It extends a run of breakthroughs that have both impressed and unsettled parts of the mathematical community while raising questions about research ethics and academic conduct. According to AGMAI, the newly formed independent advisory group of elite mathematicians put together to work on communicating the results responsibly, the release includes solutions to "hundreds" of open questions....

Simon Willison releases llm-openai-decisions plugin enabling programmatic access to OpenAI's new Decisions API with structured output support.

The startup crossed over $1 million in anualized revenue within two months of its post-pivot launch.

llm-mistral 0.16 adds reasoning model support for Mistral Large 4.

Like other solutions, it is not especially reliable, and easy to circumvent.

Simon Willison praises EmbeddingGemma 2's Apache 2.0 license, arguing open-weight embedding models reduce vendor lock-in risk for production deployments.

On Tuesday, Musubi announced a lightweight decision model made for real-time moderation called PolicyLM-1.7B, released with open weights.

Mistral releases Large 4, a 1T-param model (49B active) trained on 3,800 NVIDIA Grace Blackwell GPUs; preview via API, open weights promised end-of-month.

Nvidia-backed Lambda is raising up to $4 billion at a $14.5 billion pre-money valuation ahead of a planned 2027 IPO, led by Coatue and Blackstone.

Bitwise determinism makes large-scale pretraining easier to debug, validate, and resume reproducibly. These benefits become especially valuable when training... Bitwise determinism makes large-scale pretraining easier to debug, validate, and resume reproducibly. These benefits become especially valuable when training models with trillions of parameters across thousands of GPUs, where multiple parallelism dimensions, low-precision computation, and distributed checkpointing complicate failure reproduction and fix validation. Source

Google DeepMind releases EmbeddingGemma 2, an open-weight multimodal embedding model for efficient retrieval and semantic search tasks.

Personal AI agents promise to shop, book flights, and make reservations for you. But deliberate blocks and anti-bot defenses are getting in the way, leaving consumers caught in the middle. A new standard aims to help.

Simon Willison demonstrates integrating Parseable observability platform with Datasette's OpenTelemetry trace support via a working example.

GPU applications increasingly need networking and data movement to behave like first-class GPU-controlled operations rather than host-driven services. When the... GPU applications increasingly need networking and data movement to behave like first-class GPU-controlled operations rather than host-driven services. When the CPU sits in the middle of every network transaction, it becomes a bottleneck on the critical path, adding latency and limiting how efficiently distributed applications can respond in real time. NVIDIA DOCA GPUNetIO… Source

Anthropic expands Cyber Verification Program to give security professionals early access to advanced model capabilities with reduced safety filtering.

The AI lab's personal assistant is an operating system from the future designed to compete with Muse, Dots and Instinct.

Simon Willison comments on Mistral Large 4 benchmarking saturation with a humorous comparative prompt test across frontier models.

QF3 combines flow matching with critic gradients for efficient reinforcement learning of robot policies from demonstrations.

Conformal prediction uncertainty quantification connected to information theory via decision-theoretic entropy for set-valued predictions.

4D-HOF uses flow matching to reconstruct hand-object interactions from vision foundation model outputs without per-sequence optimization.

IdeaAnchor trains LLMs to synthesize research ideas from literature using structured supervision signals as anchors.

DepthWorld adds 3D geometric consistency to video-based robot manipulation world models via calibration and architecture improvements.

Sherpa uses multi-turn RL to train LLMs as adaptive teachers that adjust strategies based on individual student learning outcomes.