AI News & Research: Live Feed

News

23 items

OAI

2h ago

GPT-5.5 released with 2M token context and native video understanding

breaking model · OpenAI

3h ago

Runway ML raises $450M Series E at $7B valuation

funding · TechCrunch

INF

4h ago

Anthropic in talks to raise at $180B valuation, sources say

funding hot · The Information

ANT

5h ago

Claude Skills marketplace launches for enterprise teams

product · Anthropic

RTR

6h ago

China's DeepSeek releases V4 model, matches GPT-5 on reasoning benchmarks

china model · Reuters

BLM

7h ago

Nvidia unveils GB300 chip; 2.5x inference speed vs H200

hardware · Bloomberg

WSJ

8h ago

US restricts export of advanced AI chips to 42 additional countries

policy · WSJ

CUR

8h ago

Cursor 2.0 ships with multi-repo agents and background execution

product · Cursor

MSFT

9h ago

GitHub Copilot Pro adds autonomous agent tier at $39/mo

product · Microsoft

20h ago

Alibaba's Qwen3 gains ground in Southeast Asian enterprise market

china enterprise · Financial Times

22h ago

Perplexity offers free year of Pro to verified students and developers

deal · TechCrunch

VRG

23h ago

Meta launches Llama 4 Nano — 3B parameters, runs on consumer laptops

model open-source · The Verge

RTR

yesterday

Moonshot AI (Kimi) raises $600M at $4.5B valuation from Alibaba, Tencent

china funding · Reuters

BLM

yesterday

Google DeepMind's Gemini 3 tops MMLU with 94.2%, edges out GPT-5.5

model · Bloomberg

AXS

yesterday

EU AI Act enforcement begins for foundation model providers

policy hot · Axios

yesterday

Suno raises $200M Series C for AI music generation platform

funding · TechCrunch

INF

yesterday

ElevenLabs approaches $6B valuation in secondary market

funding · The Information

WSJ

2d ago

OpenAI signs $10B compute deal with Oracle Cloud Infrastructure

hardware · WSJ

RTR

2d ago

ByteDance's Doubao surpasses 200M monthly active users in China

china · Reuters

BLM

2d ago

Amazon integrates Claude into Alexa+ for all Prime members

product · Bloomberg

2d ago

Character.AI launches interactive agent SDK for developers

product · TechCrunch

2d ago

Zhipu AI (GLM) reaches $3B valuation with $400M funding round

china funding · Financial Times

AXS

3d ago

Adobe Firefly 4 launches with commercial-safe video generation

product · Axios

Research & Blogs

24 items

1h ago

Notes on prompting GPT-5.5: what's changed in the reasoning layer

analysis · Simon Willison

2h ago

Scaling Laws for Mixture-of-Experts Models at 1T+ Parameters

research · arXiv

3h ago

The Rise of Small Language Models: Why 3B is the new 70B

analysis · Latent Space

4h ago

Podcast: Demis Hassabis on Gemini 3 and AGI timelines

podcast · Dwarkesh

OAI

5h ago

Building safer agents: our latest safety evaluations for GPT-5.5

safety · OpenAI

6h ago

Efficient Long-Context Attention via Learned Sparse Patterns

research · arXiv

7h ago

AI's second act: the death of the general-purpose chatbot

analysis · Ben Evans

8h ago

Community round-up: 47 new open models released this week

open-source · Hugging Face

DPM

8h ago

Reinforcement learning from human feedback: five years later

research · DeepMind

9h ago

Import AI 412: RL breakthroughs, China's chip catch-up, robotics

newsletter · Import AI

ANT

20h ago

Deep dive: how Claude Skills work under the hood

engineering · Anthropic

22h ago

Building an AI news aggregator in 120 lines of Python

tutorial · Simon Willison

23h ago

DeepSeek-V4 Technical Report: 671B params, 37B active

research china · arXiv

yesterday

Cursor vs Windsurf vs Copilot: which agent actually ships code?

review · Latent Space

STR

yesterday

The AI Aggregators: How OpenAI became the Google of AI

analysis · Stratechery

yesterday

Fine-tuning Llama 4 for domain-specific tasks: a practical guide

tutorial · Hugging Face

yesterday

Podcast: Dario Amodei on AI safety in the age of agents

podcast · Dwarkesh

yesterday

Constitutional AI at Scale: Lessons from 100M+ Interactions

research · arXiv

2d ago

Chinese AI: what Western analysts keep getting wrong

china analysis · Ben Evans

2d ago

Import AI 411: Small models eat the world, sovereign AI trend

newsletter · Import AI

OAI

2d ago

Scaling test-time compute: our latest results with o5-preview

research · OpenAI

DPM

2d ago

AlphaFold 4: predicting protein interactions at atomic resolution

research · DeepMind

2d ago

Interview: Perplexity's Aravind Srinivas on the search wars

interview · Latent Space

STR

3d ago

The Compute Trap: Why AI startups fail at scale

analysis · Stratechery

Videos

7 items

Research update: what we're seeing in model safety evals

Anthropic · 1d ago

Livestream replay: latest product updates

OpenAI · 2d ago

Inside Google DeepMind: latest Gemini research

Google DeepMind · 2d ago

This new paper made researchers say "wow"

Two Minute Papers · 3d ago

Breaking down this week's biggest model release

AI Explained · 4d ago

Policy & Safety

6 items

TPP

3h ago

US and EU regulators compare notes on frontier-model oversight

· Tech Policy Press

7h ago

Why most AI safety incidents still go unreported

· AlgorithmWatch

CAI

1d ago

This week in alignment research

· CAIS AI Safety Newsletter

Quantum computing is a separate field from AI. It sits here because the two keep getting mentioned together, and because AI labs and chip makers are watching it. Headlines come from quantum-focused outlets; nothing is mixed into the AI columns.

Quantum headlines

live

Loading quantum headlines…

Where it stands

Oct 2026

peer-reviewedGoogle reported its 105-qubit Willow chip cutting logical error rates as the error-correcting code grew, a result published in Nature (Dec 2024). APS Physics
company claimIBM and the University of Chicago say they demonstrated quantum advantage using 70 logical qubits (July 2026). This is the companies' own claim; independent verification is still to come. IBM
company-reportedQuantinuum says its 98-qubit Helios system reached 94 logical qubits in one entangled state (Nov 2025). Not independently verified. The Quantum Insider
roadmapIBM's roadmap targets a 200-logical-qubit machine called Starling in 2029. A plan, not a delivered product. The Quantum Insider
analyst viewAnalysts note there is no accepted quantum speed-up for training or running AI models on ordinary data; loading the data into a quantum machine is a major obstacle. PostQuantum.com
expert surveyA Global Risk Institute survey of 26 experts put the chance of a code-breaking quantum computer within 10 years at 28 to 49 percent. A survey of opinion, not a measurement. Global Risk Institute

The AI industry, as it happens.

News

Research & Blogs

Videos

Policy & Safety

Quantum headlines

Where it stands

10X AI Assistant

The AI industry, as it happens.

News

Research & Blogs

Videos

Policy & Safety

Quantum headlines

Where it stands

New AI tools, by email.

10X AI Assistant