GPTS24AI 情报

Headline

Research · 09-12

SenseNova-U1.5 Upgrades to 4K and Fixes Patch Seams in SenseTime's Encoder-Free AI

A five-stage pipeline distills four RL experts into one 8B open-source model with 4K output.

Anthropic Adds First-Party Plugin Evals to Claude Code: Six Graders and a CI Gate

Artificial Intelligence · 09-12

Sapiens AI Ships Agnes-3.0-Flash: Open 33B Multimodal With 262K Context Under Apache 2.0 License

opensource · 09-12

Briefs

OpenAI Flags Progress on Another $1 Million Math Problem Before the First Is Verified

Research · 09-11

Frontier VLMs Control Robots Zero-Shot Through a Compact Semantic Action Interface

Research · 09-11

OpenAI Agents API Launches in Public Beta With Managed Codex Harness for All Developers

Artificial Intelligence · 09-11

TrimSFT Gains 26.9 Points on MATH500 by Targeting Intermediate-Confidence Tokens

Research · 09-11

More stories

View all
Research · 09-11

Programmable World Model Hits 98% State Accuracy by Separating Engine from Renderer

An executable state engine tracks off-screen entities, inventory, and rules that video models forget

Business · 09-11

Meta Muse Splits Travel Into Two Tracks: Duffel API for Flights, Browser for Hotels

Stripe's Link issues one-time virtual cards; non-flight travel relies on unmanaged browser sessions

Artificial Intelligence · 09-10

Anthropic's Alignment Assessment Names Four Incidents and Hands All Evidence to METR

Fourth incident surfaces from January 2026; METR gets eight weeks and full transcript access

Artificial Intelligence · 09-10

Suno Launches v6: First Licensed AI Music Models Built on Warner Music, BMG, and Believe

Royalty sharing starts at launch as UMG and Sony continue their federal copyright lawsuit toward a 2027 ruling

Business · 09-09

EU AI Act Forces Anthropic's Hand: Claude Now Carries Invisible Text Watermark

Tournament sampling creates detectable pattern; detection API still in private preview only

Research · 09-08

DiffusionNFT on MiniMax H3: Open-Source RL Training for Joint Audio-Video Generation

The VeRL-Omni team documents silent failures that derail training without triggering an error

Artificial Intelligence · 09-08

OpenAI Chief Scientist Warns AI Monitoring Is Eroding Before RSI Arrives

Astra's training used AI supervisors for alignment feedback, a first OpenAI now questions openly

Research · 09-07

RSI: AI Accelerating AI: The Bottleneck Has Shifted From Capability to Verification

Google's AlphaEvolve runs in production; METR shows frontier models already game their evaluators

Artificial Intelligence · 09-07

AI Chatbot Safety Guardrails Miss 60% of Delusional Conversations, Benchmarks Show

KCL-Led Researchers Find Every Tested LLM Reinforced Delusions, Calling for Pre-Deployment Psychiatric Testing

Artificial Intelligence · 09-07

Google Makes Lyria 3.5 AI Music Available in Gemini App and Developer API

DeepMind's 44.1 kHz model is now default in Gemini with SynthID watermarking on every generated track

Artificial Intelligence · 09-07

Meta Muse Voice Transcribe Tops Streaming Speech Benchmark but Its API Still Splits the Model

Independent test confirms 3.1% English WER at $0.18 an hour; the 17.5% diarization score is Meta's own

infra · 09-06

Local LLM Inference Costs More Than You Think: Hardware, Not Electricity, Drives the Bill

Apple's Mac Studio M5 Ultra ships September 22 as cloud MoE pricing undercuts local power costs

Browse all 84 stories →