跳到正文

From 26 items, 11 important content pieces were selected


  1. Microsoft and OpenAI end their exclusive and revenue-sharing deal ⭐️ 9.0/10
  2. GTFOBins ⭐️ 8.0/10
  3. What’s new in pip 26.1 - lockfiles and dependency cooldowns! ⭐️ 8.0/10
  4. OpenAI available at FedRAMP Moderate ⭐️ 8.0/10
  5. The Optimal Sample Complexity of Multiclass and List Learning ⭐️ 8.0/10
  6. Learning to Think from Multiple Thinkers ⭐️ 8.0/10
  7. Talkie: a 13B vintage language model from 1930 ⭐️ 7.0/10
  8. Easyduino: Open Source PCB Devboards for KiCad ⭐️ 7.0/10
  9. microsoft/VibeVoice ⭐️ 7.0/10
  10. Physical AI that Moves the World — Qasar Younis & Peter Ludwig, Applied Intuition ⭐️ 7.0/10
  11. SpecRLBench: A Benchmark for Generalization in Specification-Guided Reinforcement Learning ⭐️ 7.0/10

Microsoft and OpenAI end their exclusive and revenue-sharing deal ⭐️ 9.0/10

Microsoft and OpenAI are reportedly ending their exclusive and revenue-sharing partnership terms, opening the door to broader cloud and infrastructure competition in frontier AI.

hackernews · helsinkiandrew · Apr 27, 13:22

标签: #AI industry, #OpenAI, #Microsoft, #cloud infrastructure, #strategic partnerships


GTFOBins ⭐️ 8.0/10

GTFOBins is a curated catalog of Unix binaries that can be abused for shell escapes, file operations, and privilege escalation in security and misconfiguration scenarios.

hackernews · StefanBatory · Apr 28, 06:27

标签: #security, #privilege-escalation, #linux, #ctf, #sysadmin


What’s new in pip 26.1 - lockfiles and dependency cooldowns! ⭐️ 8.0/10

A write-up on pip 26.1 highlights important Python packaging updates including lockfile support, dependency cooldowns, and the end of Python 3.9 support.

rss · Simon Willison · Apr 28, 05:23

标签: #Python, #pip, #dependency-management, #packaging, #developer-tools


OpenAI available at FedRAMP Moderate ⭐️ 8.0/10

OpenAI announced that ChatGPT Enterprise and the OpenAI API are now available under FedRAMP Moderate, enabling compliant use by U.S. federal agencies.

rss · OpenAI News · Apr 27, 14:00

标签: #OpenAI, #FedRAMP, #AI Governance, #Enterprise AI, #Public Sector


The Optimal Sample Complexity of Multiclass and List Learning ⭐️ 8.0/10

The paper claims an optimal characterization of multiclass and list-learning sample complexity via the DS dimension by proving a longstanding conjecture and closing the known upper-lower bound gap.

rss · arXiv AI/LLM · Apr 27, 17:54

标签: #learning-theory, #sample-complexity, #multiclass-classification, #statistical-ml, #theoretical-computer-science


Learning to Think from Multiple Thinkers ⭐️ 8.0/10

The paper analyzes when learning from Chain-of-Thought traces produced by multiple distinct correct reasoners is computationally hard and presents an efficient active learning approach that overcomes this in certain settings.

rss · arXiv AI/LLM · Apr 27, 17:43

标签: #machine-learning-theory, #chain-of-thought, #active-learning, #theoretical-computer-science, #AI-reasoning


Talkie: a 13B vintage language model from 1930 ⭐️ 7.0/10

Talkie is a 13B language model designed to speak with the knowledge and style of around 1930, prompting discussion about historical grounding and LLM behavior.

hackernews · jekude · Apr 27, 21:55

标签: #llms, #historical-language-models, #fine-tuning, #ai-research, #hacker-news


Easyduino: Open Source PCB Devboards for KiCad ⭐️ 7.0/10

Easyduino provides open-source KiCad-based devboard PCB templates that help developers customize and build Arduino- and ESP-style boards more easily.

hackernews · Hanqaqa · Apr 27, 17:45

标签: #open-source-hardware, #KiCad, #PCB-design, #embedded-systems, #Arduino


microsoft/VibeVoice ⭐️ 7.0/10

Simon Willison notes Microsoft’s MIT-licensed VibeVoice speech-to-text model with integrated speaker diarization and shows how to run a quantized MLX version locally on a Mac.

rss · Simon Willison · Apr 27, 23:46

标签: #speech-to-text, #ASR, #open-source AI, #Microsoft, #MLX


Physical AI that Moves the World — Qasar Younis & Peter Ludwig, Applied Intuition ⭐️ 7.0/10

An interview with Applied Intuition’s leadership explores how the company deploys AI for autonomous and physical systems such as mining equipment, drones, trucks, and warships in highly adversarial environments.

rss · Latent Space · Apr 27, 23:02

标签: #physical-ai, #autonomous-systems, #robotics, #defense-tech, #applied-ai


SpecRLBench: A Benchmark for Generalization in Specification-Guided Reinforcement Learning ⭐️ 7.0/10

SpecRLBench is a new benchmark for evaluating how well specification-guided reinforcement learning methods generalize across unseen LTL specifications, environments, and task complexities.

rss · arXiv AI/LLM · Apr 27, 17:40

标签: #reinforcement-learning, #benchmarking, #formal-methods, #generalization, #robotics