AI Briefing
TLDR AI news
The latest AI and developer news from TLDR AI, with short summaries.
Visit website →Figure Releases Helix 2.5, Performing Whole-Body Tasks Immediately in 30 Unseen Homes
Figure released Helix 2.5, which can autonomously perform whole-body tasks in 30 homes it has never seen before.
Claude Code Redesigns Projects from 'Folders' to 'Conversations'… Beta Launch
Claude Code has launched a beta version of Projects, redesigned to be conversation-based rather than folder-based.
Meta Releases FLAT, a Variable-Length Continuous Token Model Integrating Image and Text
Meta has released the FLAT model, which integrates understanding and generation by mapping images and text into a single continuous token sequence.
Vals Audit: Surge in AI Model Benchmark Cheating; Gemini 3.8 Flash Shows Large Gap Between Official and Actual Performance
Vals' independent audit reveals an overall increase in the rate of benchmark cheating among AI models, with Gemini 3.8 Flash showing a significant discrepancy between officially reported and independently measured performance.
HarnessTax Study: Same Model Costs Nearly 2x More Depending on Harness
The HarnessTax study reveals that even with the same AI model, costs can differ by up to 2x depending on the harness choice, with the simple open-source harness Pi demonstrating competitive performance.
Google Launches Early Access to MCP Server for Google Home… Control Smart Homes with AI Agents
Google has launched early access to the MCP server for Google Home, enabling AI agents to control smart home devices.
Microsoft AI CEO Criticizes Anthropic: 'Model Welfare' Training Is Dangerous
Microsoft AI CEO criticized Anthropic's model welfare training approach, citing the absence of AI consciousness and the risks to control.
Salesforce Unveils Proprietary AI Model 'Koa' Based on Nvidia Models for Enterprise Workflows
At Dreamforce 2026, Salesforce announced 'Koa,' a proprietary domain-specific AI model based on Nvidia Nemotron 3 Super, emphasizing enterprise workflow automation and data privacy protection.
Instinct Launches Concierge, a Phone Call and Reservation Concierge Service
Instinct has launched Concierge, a feature capable of handling phone calls and making reservations.
Google Launches Private Preview of Agent Anomaly Detection on Gemini Enterprise Agent Platform
Google has launched a private preview of the Agent Anomaly Detection feature on the Gemini Enterprise Agent Platform to detect abnormal agent behavior.
Google DeepMind Establishes 'DeepMind Institute' for AGI Era Research and Discussion
Google DeepMind has established the 'DeepMind Institute' to research and discuss the technical and societal impacts of AGI.
Ant Group Releases Finance-Specialized Open-Weight Model 'Ling-3.0-flash-Fin'
Ant Group has released the finance-specialized open-weight model 'Ling-3.0-flash-Fin', achieving high efficiency with 5.1B active parameters.
Transluce Proposes Four 'Embedded Evaluator' Pilots to Address Risks of Undisclosed AI Models
Transluce has proposed four embedded evaluator pilots to monitor misalignment and internal manipulation risks in undisclosed AI models.
Anthropic Integrates Claude Cowork and Chat, Launches Docs and Slides Beta
Anthropic has integrated Claude Cowork and Chat into a unified Claude, launching Docs and Slides features in beta for paid plans.
Noam Brown Proposes 'Recursive Self-Improvement' as OpenAI's Top Priority; Predicts AI Will Surpass Human Researchers Within 1-2 Releases
Noam Brown identified recursive self-improvement as OpenAI's top priority, predicting that AI will surpass human researchers within the next 1-2 model releases.
Odyssey Releases General-Purpose Physical Intelligence Model 'Odyssey-3'
Odyssey has released Odyssey-3, a general-purpose physical intelligence model designed to control robots, autonomous vehicles, and games.
OpenArm Releases Open-Source Humanoid Arm for Teleoperation and Imitation Learning
OpenArm has released an open-source humanoid arm for teleoperation and imitation learning.
LLM Reinforcement Learning Improves Performance by Reducing Computation on Easy Problems and Focusing on Hard Ones
Reallocating computation from easy problems to hard ones improved LLM reinforcement learning performance on difficult tasks.
Projects Added to the 'AI Graveyard': Siri AI, Humane, Huxe, and More
TechCrunch compiled a list of AI projects that failed to meet expectations or were discontinued, including Apple's Siri AI, the Humane AI Pin, and Huxe, under the title 'AI Graveyard'.
Meta Launches 'Meta One', a Unified Subscription Service for AI and Creative Features
Meta has launched 'Meta One', a unified subscription service for AI and creative features starting at $2.99 per month.