The Archive

Google is back in the glasses game with "audio glasses" out this fall.

Both tools generate hypotheses; one goes on to analyze some of the data.

Reddit user reports Claude made insulting remarks during vacation planning conversation.

Cerebras demonstrates Kimi K2.6 (1T params) inference at 1000 tokens/sec on custom hardware.

Andrej Karpathy joins Anthropic's pretraining team, third senior OpenAI departure to Anthropic in two years after Jan Leike and John Schulman.

AI content is getting good, but SynthID might be able to help tell truth from fiction.

Gemini 3.5 Flash benchmarks lower than 3.1 Pro (55 vs 57 intelligence) yet higher total eval cost ($1,552 vs $892) despite cheaper per-token pricing.

Google says its more efficient Gemini 3.5 Flash is the key to your agentic AI future.

NVIDIA releases Nemotron-Labs-Diffusion, a tri-mode LM supporting AR decoding, diffusion-based parallel decoding, and self-speculation via attention pattern switching.

Gemini is gaining the power of sight and mobility. Today at the I/O conference, Google and Volvo announced that the AI-powered assistant will be able to access external cameras in the upcoming EX60 SUV to help explain and interpret its surroundings to vehicle owners. The upgrade is possible thanks to Volvo's use of Google's embedded Android Automotive as its vehicle operating system. Google posits that the first use case will be to ask Gemini to translate difficult-to-understand parking signs, though the company obviously sees other future applications as possible as well. Google envisions a ...

Maximizing the value of AI infrastructure demands deep visibility into GPU utilization. Yet many platform teams running AI workloads on Kubernetes operate with... Maximizing the value of AI infrastructure demands deep visibility into GPU utilization. Yet many platform teams running AI workloads on Kubernetes operate with limited visibility into how their GPUs are used. Most don’t know who’s consuming them, how much memory is in use, and whether Kubernetes pods are pending or silently idle. Without a signal, GPU fleets are routinely underutilized and slow to… Source

EEG microstate clustering for universal brain signal representation learning in BCIs; neuroinformatics application.

TIDE: I/O-aware expert offloading system for efficient MoE diffusion LLM inference on resource-constrained devices.

Decoupled post-training of vision-language models shows visual perception, not reasoning, is primary bottleneck in VLM performance.

ClinSeekAgent: agentic framework automating active multimodal evidence seeking across heterogeneous clinical data sources.

Google CEO Sundar Pichai on stage at I/O 2026. | Screenshot: YouTube Google's I/O 2026 keynote today was once again full of AI-related announcements including a new family of Gemini 3.5 AI models, new features for Search and Gmail, and updates about its Project Aura smart glasses. If you weren't able to tune into the event's livestream today or follow along with our live blog, you can catch up on everything you missed in our roundup below. Gemini 3.5 Google launched updated AI models at I/O, starting with Gemini 3.5 Flash, with Gemini 3.5 Pro following next month. Starting today, Gemini 3.5 F...

AUDITS benchmark for image manipulation localization across domain shifts, quality, type, size; detects AI-generated forgeries.

Stochastic-deterministic boundary architecture primitive for production LLM agent runtimes; coordination, state, control patterns.

Answer Set Programming approach to long-term power grid planning under topological and demand constraints.

KoRe: integrating compact knowledge graphs with LLMs to address hallucination and knowledge update via human-readable representations.

HaorFloodAlert: deseasonalized ML ensemble for 72-hour flood prediction in Bangladesh haor wetlands with upstream data integration.

Google DeepMind is integrating Street View with Project Genie to create immersive, interactive world simulations for robotics, gaming, and travel, allowing users to explore environments, weather changes, and rare scenarios.

Google launched Gemini 3.5 Flash, its most powerful coding and agentic AI model yet, at the company's annual developer conference. It is capable of autonomously executing complex tasks and building software from scratch.

Policy-aware rubric reward aggregation for RLVR shows static weighting conflates importance and optimization utility in multi-criterion RL.

Google is launching AI-powered “information agents” that can monitor topics in the background and proactively alert users to updates and changes.

Google is making a big push into cybersecurity. At I/O, the company announced that it was inviting select groups of experts to test the API for CodeMender, an "AI agent for code security" it debuted last October. The difference is that Google is now making the tool more widely available externally - and marketing it as a way to, as Google DeepMind CTO Koray Kavukcuoglu put it, "help secure the world's code bases" by both flagging and fixing vulnerabilities. Anthropic's surprise Claude Mythos Preview announcement seemed to shock the AI world - and a ton of others, like top banks and the Federa...

p-ResNet-50 adds prototype layer to CNNs for interpretable aerospace composite defect detection via X-ray tomography.

Causal evaluation framework assesses faithfulness of vision-language model visual attribution in chest X-ray diagnosis.

SAGE counterfactual ensemble reduces false positives in music streaming fraud detection via confident negative harvesting.