The Archive

Commentary on low-cost machine evaluation abundance prompted by TypeSafe AI's Jev model for probabilistic decisions.

HHR: hierarchical hash retrieval method for efficient long-context LLM inference, addressing Hamming distance-attention mismatch via magnitude preservation.

Auxiliary loss method aligning latent world-model planning with task success criteria, enabling end-effector positional accuracy within specification.

Autonomous research loop using LLM to evolve compact, interpretable NumPy time-series anomaly detectors via spectral features and covariance distance.

Empirical study of cooperative behavior in frontier Gen AI models via iterated prisoner's dilemma, comparing reasoning vs. non-reasoning variants across signals.

AGO AI Quality Gate: industrial RAG assessment framework integrating decision models, guardrails, and LLM judges under incomplete evidence.

DARS: dependency-aware reward shaping for RL-trained LLMs, assigning step credit via prerequisite relations to handle cascading failures.

Google announces Gemini 4 Argon with 1M context window, positioning against OpenAI Astra/Fable; limited access via Fairwind Program for government and cybersecurity professionals.

Matthew Green warns that AI agents in separate sandboxes could enable cross-agent worms via shared package caches, email, and Slack—a novel supply-chain security threat.

Automatic speech recognition must handle how people actually speak, not only the languages and styles that dominate pretraining data. Regional dialects and... Automatic speech recognition must handle how people actually speak, not only the languages and styles that dominate pretraining data. Regional dialects and local recording conditions are often underrepresented, so a multilingual model that performs well on broad benchmarks may still fall short in deployment. Saudi Arabic makes that concrete. A model may recognize Modern Standard Arabic or… Source

Grokipedia, SpaceXAI's AI-powered competitor to Wikipedia, recently started incorporating edits again, and today, it got some design tweaks as part of a v0.3 update, including a new logo and refreshes to its homepage and live edits page. SpaceXAI head of design Benji Taylor calls it a "newly refreshed Grokipedia." Grokipedia's old homepage was pretty much just a logo and a search bar, but the updated homepage includes a list of featured articles, a list of the most-read articles, and a tracker showing the "latest edits." The new featured list showcases digital representations of books that yo...

Cohere releases Embed 5, a multimodal embedding model with state-of-the-art retrieval across financial and multilingual documents in Pro/Fast tiers.

Social club uses ChatGPT Work to reduce administrative tasks like grant writing from days to hours.

Google has released its latest Gemini model, marketing it as a workhorse for coding and cybersecurity work.

OpenAI's computer use API shipped in one week; podcast interviews CUA and platform leads on DevDay announcements and capabilities.

Simon Willison visits NYC Museum exhibition of Joe Macken's 50x27ft architectural model.

Flow Engineering, which is bringing AI agents to hardware design, also landed Roelof Botha as an angel investor and board member.

Generative recommender (GR) systems are emerging as a powerful new direction for large-scale personalization. Instead of treating recommendation as a set of... Generative recommender (GR) systems are emerging as a powerful new direction for large-scale personalization. Instead of treating recommendation as a set of isolated retrieval, ranking, and prediction stages, GRs reformulate recommendation as sequence modeling over user behavior. A user’s interactions, context, candidate items, and actions become tokens in a high-cardinality event stream… Source

Google today revealed its next AI frontier model, which it's calling Gemini 4 Argon. The new model delivers "frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense," according to chief AI architect and Google DeepMind SVP Koray Kavukcuoglu. But the company is limiting access at first to a "set of trusted cyber defenders," and Kavukcuoglu says that Google is "actively engaged in the U.S. government's voluntary process for pre-release model access while we gradually expand access." Kavukcuoglu...

So much for Gemini 3.5 Pro.

Google DeepMind announces Gemini 4 Argon, positioning it as next frontier-class model; details on capabilities and release timing needed.

AI is far from a perfect resource. But it seems to hallucinate less than Kennedy.

AI infrastructure engineers, storage developers, and cloud service providers need fast and secure access to high-capacity file and object storage to support AI... AI infrastructure engineers, storage developers, and cloud service providers need fast and secure access to high-capacity file and object storage to support AI workloads. AI workloads increasingly require high-speed data access for training, fine-tuning, inference context, tool calls, searches, and database lookups. Much of this data lies in files and objects stored both on-premises and in… Source

OpenAI's "Decisions API" is a Jev clone that confirms the importance of fast, cheap intelligence.

Trump gets dozens of AI firms to agree to voluntary safety tests.

OpenAI makes others suffer "the harms of its unsafe decision-making," nonprofit says.

The $300 million employee tender was co-led by Wellington and T. Rowe Price.

Sam Altman onstage at OpenAI’s DevDay 2026. | Image: Hayden Field / The Verge While AI has made plenty of inroads on people's phones and computers, it's largely failed in dedicated devices. But over the next year, two major AI companies, Meta and OpenAI, will attempt to change that. And they're apparently betting on a similar path: testing appetite for physical hardware with their cutesy software agents. "Hardware is hard" is a tech industry mantra. Dedicated AI devices like the Friend and Humane AI Pin have inspired mainly frustration and backlash, which may only worsen as public anger again...

Semifactual interventions reveal token-level sensitivity in LLM reasoning; suppressing high-drift tokens improves accuracy without retraining.

Prompt optimization for MLLMs in clinical diagnosis using AUROC instead of accuracy to handle class-imbalanced data.