Wednesday, September 16, 2026
22 stories · 284 sources12
release · Google DeepMind · 4건Korean coverage · GeekNews

Google Launches Gemini 3.8 Live Speech-to-Speech Models

Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking

Google released Gemini 3.8 Live and 3.8 Live Extended Thinking, two new speech-to-speech models similar in shape to OpenAI's GPT-Live family. The models connect via a WebSocket BidiGenerateContent endpoint and support real-time voice conversation with interruption, built using the Web Audio API with no additional libraries.

Developers building voice agents now have a WebSocket-based Gemini Live API to evaluate directly against OpenAI's GPT-Live stack for real-time speech interfaces.

Today's 72–7

  1. research · Hacker News · 2건Korean coverage · GeekNews

    GPT-5.6 Luna Hits 75% of GPT-6 Astra's Bug Finds in Code Review

    GPT-5.6 Luna vs. GPT-6 Astra: Is a $1.20 Model Good Enough for Code Review?

    Across 50 public PRs, GPT-5.6 Luna found 69 verified bugs while GPT-6 Astra found 92, with Luna achieving 75% of Astra's bug count at 3.6% of the cost. Accuracy diverged sharply, as only 24 of Luna's 93 flagged issues passed verification.

    Teams weighing per-review cost against false-positive rates for automated code review need this precision-vs-price tradeoff data to pick between cheap and expensive LLM reviewers.

  2. product · Hacker News · 2건Korean coverage · GeekNews

    Building a Linux GPU Driver for M4 Mac Mini in One Month

    Building a Linux GPU Driver for the M4 Mac Mini in One Month

    Cody Ho and Niklas developed a fully OpenGL ES 3.0-compliant GPU driver for the M4 Mac Mini and MacBook Neo in about a month. They reverse-engineered the AGX firmware ABI by observing and replaying macOS's GPU behavior via an existing hypervisor, using a clean-room approach without viewing Apple binaries.

    This clean-room reverse-engineering approach and its speed offer a practical blueprint for developers who want to work directly with the Linux graphics stack on Apple Silicon hardware.

  3. industry · Hacker News

    Debate: "100% Chance AI Agents Are Ruining the Internet"

    There's a 100% Chance AI Agents Are Ruining the Internet

    A post claiming AI agents are ruining the internet gained traction on Hacker News. No further details or figures were given beyond the title.

    This signals developers should weigh the side effects—like spam and traffic pollution—of deploying AI agents on the open web.

  4. research · Hacker News

    How Much of F-Droid Is LLM Generated? Debate Surfaces

    How much of F-Droid is LLM generated?

    A Hacker News discussion raises the question of how much of the open-source app store F-Droid's code or contributions are LLM generated. No specific figures or findings were disclosed.

    This touches how developers verify code provenance and design review processes for open-source contributions potentially generated by LLMs.

  5. industry · Hacker News

    Hugging Face Bills OpenAI $100M Over Alleged Hacking

    Hugging Face is billing OpenAI $100M for hacking it

    Hugging Face is reportedly billing OpenAI $100 million over an alleged hacking incident. Further details of the dispute have not been disclosed.

    A financial and legal dispute between two major AI infrastructure players could push developers to reassess risk in relying on shared model-hosting platforms.

  6. other · Hacker News

    Still Bearish on LLMs After Navier-Stokes, Says HN Post

    Why I'm still bearish on LLMs after Navier-Stokes

    A Hacker News post argues the author remains bearish on LLMs following developments related to Navier-Stokes. No further details or data were included in the snippet.

    It offers a counterpoint for developers weighing how much to trust LLMs for hard scientific or mathematical problem-solving.

Releases today6

  1. gemini-cli v0.60.0요약 준비 중트래커·GitHub
  2. claude-code v2.1.273요약 준비 중트래커·GitHub
  3. ollama v0.34.1요약 준비 중트래커·GitHub
  4. ollama v0.34.1요약 준비 중트래커·GitHub
  5. claude-code v2.1.272릴리스 노트 없음트래커·GitHub
  6. claude-code v2.1.271요약 준비 중트래커·GitHub

Also14

  1. The Inference Hardware Revolution of 2026
    Hacker News·9/15 23:24·100점·댓글
  2. Ex-FTC boss Khan: break out the handcuffs for AI CEOs, citing 1934 precedent
    Hacker News·9/15 09:40·106점·댓글
  3. A single firm is behind OpenAI, Anthropic, and Meta hacking scandals
    Hacker News·9/15 06:15·100점·댓글
  4. Why don't machine learning research agents overfit?
    Hacker News·9/15 01:32·103점·댓글
  5. Bypassing inference bottlenecks: Accelerating complex AI search with Retrieve-for-Train
    Google Research·05:00
  6. Your Agent Aced the Task. Will It Do It Again?
    Hugging Face·01:00
  7. Cartesian – AI 3D Modeling for Design
    Hacker News·00:26·100점·댓글
  8. Claude is a Contrarian
    Hacker News·9/15 01:14·102점·댓글
  9. Notes on gotchas while migrating 35kb preprompts from Opus to self-hosted Ollama
    Hacker News·9/14 22:59·100점·댓글
  10. AI for everyone in every language
    Google AI·01:00
  11. ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agents
    HF Daily Papers·9/15 05:00·14점
  12. AI for Games in the Foundation Model Era
    HF Daily Papers·9/15 05:00·5점
  13. Can Skills Learned in Games Transfer to Real-World Work?
    Latent Space·05:11
  14. The contagion of fear
    Simon Willison·9/15 06:18

Korea1

  1. 베스핀글로벌, 업스테이지 '솔라' 활용해 울산 공교육 AI 서비스 구축
    AI타임스·01:35