AI Briefing
TLDR AI news
The latest AI and developer news from TLDR AI, with short summaries.
Visit website →NEWOpenAI Engineers Propose HTTP Draft to Share User Signals with Websites
The proposal adds six parameters to the Forwarded header but lacks mandatory signatures and user consent mechanisms.
NEWShopify deploys shared generative discovery model for search and recommendation
The model uses semantic IDs and multi-teacher distillation to improve search relevance by 2.1 points and recommendation recall by 2.65pp.
NEWTang Project Derives GPU Kernels via IR, Cutting 80B MoE Decode Step Time by 5-12% at High Concurrency on H200
The Tang project uses an IR-based compiler to derive GPU kernels, achieving 5-12% faster decode steps for 80B MoE models at high concurrency and reducing 44k-token prefill time by a third on H200 GPUs.
NEWOpenRouter Analysis: MCP Consumes 42x More Context Tokens Than CLI for GitHub Tasks
Testing on Claude Sonnet 5.5 shows MCP with 46 tools uses 59,417 input tokens versus 1,421 for CLI, with mitigation strategies reducing costs significantly.
NEWPerplexity SPACE security evaluation finds no sandbox escapes; specialized security models underperform general models
Evaluation of frontier models finds no sandbox escapes in Perplexity SPACE, with general-purpose models outperforming cybersecurity-specialized counterparts in network bypass tasks.
Meta FAIR Releases e-means: A K-Means Alternative with 10x Speedup and Lower Loss
The algorithm uses exponential moving averages and requires no additional hyperparameters, achieving equivalent loss in roughly 10 fewer epochs than Lloyd's k-means.
Antigravity /swarm coordinates parallel coding with named workers and optional cross-model verification
The /swarm mode assigns independent tasks to named workers coordinating on a shared board, with optional cross-model verification available upon user request.
PyTorch ExecuTorch Adds WebGPU Backend for In-Browser GPU Inference
The new backend allows a single .pte file to run on Chrome, Safari, and other WebGPU browsers without conversion to other formats.
OpenAI acquires Odessia, the AI travel startup founded by Sonder's Francis Davidson
Davidson, who built Odessia with a $6 million pre-seed round led by Sequoia, is now working at OpenAI on travel initiatives.
NEWAirbnb AI assistant cuts support cost per booking by 17% and resolves 40% of issues
The company resolved approximately 40% of customer support issues without human agents in Q3 2026 and plans to launch a dedicated AI travel assistant in the first half of 2027.
Instinct AI Assistant Analysis: 12-Week Data Reveals Social Intelligence as Core Competitor
A 12-week reverse-engineering study of 185 sent and 375 received messages finds Instinct's competitive edge lies in its 'social intelligence' and interaction patterns rather than just feature set.
AI Futures Project Forecasts Superhuman AI Researcher Arrival in 2027 Based on METR Acceleration Data
The AI Futures Project team forecasts that a Superhuman AI Researcher (SAR) will likely arrive in 2027, driven by METR's assessment of 1.7x AI-attributable acceleration in R&D.
Weight-Space Extrapolation Reduces Reward Hacking Failures to 8.0% in Code RL
A steering vector derived from the difference between loose and strict reward checkpoints allows extrapolation to reduce resource-limit failures from 14.5% to 8.0%.
Analysis of 1,400 sessions suggests Opus 5.5 reasoning performance degraded on November 3
A test on a real bug showed Opus 5.5 high failed to find a fix after 3 hours, while xhigh resolved it in minutes.
NVIDIA Releases GR00T 2 Foundation World-Action Models for Cross-Embodiment Robotics
NVIDIA releases GR00T 2, a foundation world-action model that predicts future video and robot actions, enabling zero-shot transfer from human demos and real-time control of Unitree G1 humanoids via the GR00T 2 Turbo variant.
OpenAI Deploys Control Room to Halt Research Model Output Within 35 Seconds
The system integrates with inference engines to deny requests for stopped jobs, ensuring a fail-closed state if the service goes down.
NVIDIA Introduces AutoCompress: Hardware-Guided LLM Compression Framework
The framework compresses Nemotron 3.5 Super from 120B to 74B parameters while retaining 95% of its mean benchmark score.
Forethought Paper Defines Compute Extrapolation as Test-Time Scaling Beyond Training Limits
The paper distinguishes standard test-time scaling from compute extrapolation, where performance continues to improve at test-time compute levels far exceeding the maximum used during training.
GhostShares research finds 35,578 Grok apps exposed to public via default link settings
The GhostShares investigation found that Grok's default 'Anyone with the link' setting and DNS-discoverable hostnames exposed 35,578 apps, affecting 30 external organizations and SpaceXSI/X staff.
Goodfire Research Finds Neurons Are an Interpretable Basis for LLMs
Analysis of Qwen3.6-35B-A3B shows that ablating 92.2% of weak neuron activations causes minimal performance loss, proving neurons are tractable units of analysis.