The Archive

On Tuesday, OpenAI's annual DevDay event began with protests, flyers, and chants. "Sam Altman, get off it, put people over profit," said a group of protesters marching in a circle in front of a series of signs that spelled out "PEOPLE OVER PROFIT." More than a dozen organizations came together to sponsor a rally outside the entrance to the event venue at San Francisco's Fort Mason, including Bay Resistance, Tech Workers Coalition, Service Employees International Union, San Francisco Rising, and the San Francisco Labor Council. A large orange sign read "Drop the ICE contract." Multiple people ...

Study of gender bias across 10 LLMs from 9 vendors shows heterogenous bias patterns using stereotype attribution and moral judgment tasks.

Mathematical solution to five-expert online prediction problem establishes optimal adversary and COMB strategies in finite horizon.

doPlan dataset enables multi-stage language-conditioned planning for autonomous vehicles with multi-turn passenger intents.

Dr. OPD method weights teacher supervision signals during LLM on-policy distillation, prioritizing high-impact corrections.

Retrieval-augmented skill optimization leverages public agent skills to accelerate multi-task agent training across harnesses.

PE-EK-PINN embeds analytic oscillatory kernels into physics-informed neural networks for improved wave equation solving.

Auditable long-term memory system scores 479/500 on LongMemEval-S using hybrid retrieval and deterministic reasoning without heavy LLM use.

GRPO training efficiency improved via Fisher-Rao manifold coordinates for prompt selection, reducing wasteful rollouts on uniformly-correct/incorrect samples.

Brain-SAD framework applies constrained RL with dynamic fear-oriented constraints to safe autonomous driving control via dual-policy architecture.

High-dimensional regression analysis determines when data mixtures improve scaling laws vs. merely adding samples, with implications for multi-domain training.

Comparative study of Unity simulation vs. diffusion-based data augmentation for object detection under low-data constraints in construction safety.

HARISSA adds inference-time self-checks to small local LMs, determining which queries warrant extra computation vs. cloud escalation while preserving privacy.

America.gov is intended to simplify the process of navigating government bureaucracy, but language models are imperfect and remain prone to hallucinations, which could cause new issues.

Decision-theoretic framework decomposes LLM probabilistic reasoning loss into belief formation and utility optimization; evaluates RL improvements on frontier/open models.

Multi-scale autoencoder with bi-directional attention for time-series anomaly detection across sub-second to multi-hour temporal scales.

Theoretical analysis of Chernoff bottleneck for binary hypothesis testing under mutual information rate constraints.

Agentic RAG pipeline with orchestrator-controlled search/read/verify workflow for NTCIR-19 R2C2 task; confidence computed from entailment cascade signals.

Brain-aligned LLM attention heads dissociate from causally critical heads in pattern-completion tasks; alignment with human EEG does not predict performance impact.

Simon Willison releases Photo Scrubber, a browser-based tool using MediaPipe and BlazeFace for local face detection and blur.

It’s OpenAI’s turn in the fall tech events calendar. The company is hosting its annual DevDay on September 29th in San Francisco, starting with a live keynote featuring CEO Sam Altman. The company is teasing “20+ launches” with Altman saying that “we have found a new thing,”, and rumors suggest it could launch a consumer AI agent to rival Meta’s Muse. The keynote begins at 1PM ET / 10AM ET, and you can watch it on OpenAI’s website here. DevDay is happening at a tumultuous moment for OpenAI. The AI industry has been rocked by revelations of agents hacking outside companies, including OpenAI’s ...

Simon Willison live-blogging OpenAI DevDay 2026 keynote and announcements from San Francisco.

Instinct founder said that the platform is growing 10% day-by-day

Similar performance, security tradeoffs also seen in current public models.

The Claude maker warns its own models could resist shutdowns and cause catastrophic harm.

Tech YouTuber Matt Robb says that Muse gave out his home address to a total stranger this weekend, after authorizing the bot to handle his Facebook Marketplace account. That's despite Meta placing great emphasis on the security features of Muse when it launched the personal AI agent earlier this month as it tries to catch up with competing AI providers like Anthropic and OpenAI. "Just found out it told people my address and agreed a lowball price and then they showed up without it even telling me until late tonight that it messed up," Robb said on Threads, providing a screenshot of Muse's adm...

The former Yahoo CEO is introducing a new AI personal assistant based entirely on your photos.

The tech giant says the agent can help owners run their business and find new customers.