2026-10-24 的 AI 资讯

OpenAI says it will no longer train its AI models on the live internet, citing “significant risks”

AI devours a fifth of companies' IT budgets

Hottest toy for Christmas? Nvidia CEO plays with adorable Microduck robots, 20,000 sell

AI is devastating higher education, but the system was broken

OpenAI now trains its frontier AI models with no internet access

Securing our frontier model training environment

Show HN: Corgi, a helpful {agent, terminal} multiplexer

Jack Dorsey’s Block made an AI model that talks to you like an awkward new friend

New AI Is More Than Twice as Good at Diagnosing Diseases as Doctors

Dessa: the most authentic full-duplex speech model ever built

From Toast to Bessemer: Who founders will hear from at TechCrunch Founder Summit 2026

The iPhone Duo should have been an iPad

Rapidly scaling online storage to serve over 1 billion ChatGPT users (Part II)

Now Google says it might use your Gmail and Google Photos data to target ads

Meta delays new business terms after publisher pushback on AI scraping

Retro gaming is being overrun with AI

Uber rival inDrive eyes 2027 robotaxi tests in Central Asia with Pony AI and Nvidia

Left alone with nothing to do, AI agents started playing

OpenAI and Poland’s NASK partner to advance AI safety and security

OpenAI working on new $30B raise at $1.25T valuation ahead of IPO

Canva’s first AI image model plans designs before generating them

Ask HN: If the models are so good, why doesn't OpenAI just start 100 companies?

Meta’s AI could still scrape news sites, legal experts say

Meta turns Muse sharing back on

Sure, let’s smoosh SpaceX and Tesla together, who cares

1Password launches a way for AI agents to log in on your behalf without seeing your password

Build more responsive apps with the Decisions API

Canva releases its first image generation model

OpenAI makes its Jev competitor widely available as decision model competition heats up

Microsoft’s Mustafa Suleyman says its AI outperforms doctors on hard cases as he chases ‘medical superintelligence’

Sensorimotor Alignment: A Platonic Proxy for Embodied Policy Success

Direct Pass@$k$ Optimization for Thought Pruning: Balancing Quality and Diversity in Test-Time Reasoning

Phase Transition from Random Features to Fully Trainable Networks

Reliable AI-Assisted Decisions via Selective Auditing

SpectralCanary: Embarrassingly Simple Low-Frequency Traces for T2I Dataset-Use Auditing

Cost-Aware Abstention for LLM-Based IT Ticket Classification

METTA: Memory-Enhanced Test-Time Adaptation for Meta-Reinforcement Learning

EpiSEM-Bench: Evaluating Safety and Evidence Recovery in LLM Agents under Missing Evidence

Low-Bit N-Gram Embeddings for Language Models

A Fix for Softmax and Cross-Entropy

Causal Chain of Thought: Mitigating Attention Leakage via Context Isolation

Can an Agent Debug a Web Corpus? Agentic Auditing of Slovak FineWeb2

BeNice! Closing Looping's Inference-Memory Gap via KV-Sharing

Questioning Instrumental Convergence in RLVR: Checkpoint Evidence from OLMo-3

Bilingual Language Models Capture Cross-Linguistic Influence in Second-Language Reading

OP-Print: Tracing Membership Imprints in On-Policy Self-Distillation

Efficient and Optimal Learning of Discrete Distributions with Person-Level Privacy

How Ground Truth Shapes Retrieval Scores: Evidence-Aware Evaluation of a Retrieval-Augmented Generation System for Breast Lesions

CRANE: Post-Hoc Counterfactual Retrieval for Multimodal Classifiers

Simulated Control Evals

Beyond Best-of-$k$: Prior-Consistent DAPS for Ambiguous Nonlinear Inverse Problems

VERA: A Responsible AI Dashboard for Assessing LLM Deployments in Light of the EU AI Act

Would an LLM Match into a Residency? Uncertainty, Reasoning and Image Use of Open Language Models on the Spanish MIR Examination

Action Learning for Scientific Simulations

Who Is Sycophantic to Whom? Confident LLMs, Human Deference, and the Collapse of Trust Calibration

What Makes a Good Unmasking Policy? A Controlled Study of Learned Unmasking in Diffusion Language Models

Rethinking Sequence Compression with Test-time Regression

Social-HUMINT: An LLM-Augmented Unified Framework for Measuring Social Media Exposure

On the Token Collapse Phenomenon in Textual Graph Soft Prompting

When to Follow the Teacher: Advantage-Based On-Policy Distillation