Monday, October 12, 2026
22 stories · 157 sources38호
industry · Hacker News · 2건Korean coverage · GeekNews

Nvidia in talks to acquire US open-model startup Reflection AI

Nvidia in talks to acquire US 'open' model startup Reflection AI

Nvidia is in talks to acquire or increase its investment in Reflection AI, a US developer of open-weight AI models that the US government sees as a competitor to low-cost Chinese models. The talks are at an early stage, and options beyond a full acquisition include an acqui-hire of staff, a technology licensing deal, and additional equity investment.

If you are weighing US open-weight models as an alternative to cheaper Chinese ones, a deal could change who owns Reflection AI's models and on what license terms, so watch it before you commit to a dependency.

Today's 72–7

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

    500B Tokens Later: AI Agents Decompile a First-Person Shooter

    500B Tokens Later: Letting AI Agents Decompile a First-Person Shooter

    Over about three months, AI agents reconstructed a game in C++, recovering 99% of all functions and reaching byte-for-byte matches with the original in 83%. In the first month the game could run and load maps, but readable-looking code hid wrong function signatures, types and struct layouts.

    Agent-written code that runs and reads well can still have wrong signatures, types and struct layouts, so reverse-engineering workflows need automated checks like byte-level matching built into the design.

  2. industry · TechCrunch AI · 2건

    Nadella: AI models need an 'emergency brake,' should be assumed compromised

    Microsoft’s Satya Nadella says AI models need an ‘emergency brake’

    In a Saturday post on X, Microsoft CEO Satya Nadella wrote that it's time "to step back and assess the trust architecture" of AI. He rejected treating AI as a "set of nested black boxes" and called for systems where models can be contained, observed, and leave behind "tamper-proof human readable evidence."

    Developers shipping AI models should design for the assumption that a model may be compromised, building in containment, observability and incident-disclosure processes from the start.

  3. other · Simon Willison

    Dwarf Fortress now uses version control

    Dwarf Fortress uses version control now

    Simon Willison notes that Dwarf Fortress, long known for not using version control, appears to have adopted it. Tarn Adams said in March 2013 he disliked code being committed into a "black box thingy with no immediate upside," and in March 2016 he still said he didn't use version control.

    Even a famously complex codebase eventually moved to version control, which is a prompt to check whether your own projects have commit history and rollback in place.

  4. release · Simon Willison

    Python 3.15.0 added to actions/python-versions

    Python 3.15.0 added to actions/python-versions

    Simon Willison reports that the stable Python 3.15.0 has landed in the actions/python-versions repository, which he had been waiting on for a couple of days. You can now add "3.15" to a GitHub Actions testing matrix to run tests against the new release.

    Adding "3.15" to your GitHub Actions matrix is now enough to check your project against Python 3.15.0 in CI.

  5. policy · The Verge AI

    Silicon Valley falls in line with Trump's push to rename AI

    Silicon Valley is falling in line with Trump’s attempt to rename AI

    President Trump announced plans to rebrand artificial intelligence as "super intelligence," though most of the broader population is sticking with AI. Some of Silicon Valley's most powerful are going along with it, with Elon Musk promising to rename SpaceXAI to SpaceXSI.

    Since the White House is pressing on terminology, developers selling to government may need to check how product and feature names are worded in procurement documents and proposals.

  6. industry · TechCrunch AI

    Can the AI industry win over data center foes by dropping NDAs?

    Can the AI industry persuade data center opponents by getting rid of NDAs?

    On TechCrunch's Equity podcast, the hosts debate the impact of data center moratoriums. They also ask whether the AI industry has a communication problem.

    If data center moratoriums spread, they could constrain cloud capacity and site planning, which matters to teams weighing region choices and capacity commitments.

Releases today1

  1. open-webui v0.12.0OpenAI Realtime 음성 모델로 대화하는 실시간 음성 통화가 추가됐다트래커·GitHub

Rising on GitHubstars gained in a day

  1. morluto/reaReverse engineer anything with agents, from app behavior to native binaries.+25,585
  2. storytold/artcraftAn intentional crafting engine for artists, designers, and filmmakers.+2,956
  3. cathrynlavery/diagram-designDiagram design skill with 42 diagram types for Claude Code and other coding agents, in self-contained HTML + SVG.+1,083
This week's repositories →

Also2

  1. These execs think voice AI hasn’t reached its ChatGPT moment yet
    TechCrunch AI·10/11 23:00
  2. Learning to use local AI is exciting, overwhelming, and frustrating
    The Verge AI·10/11 20:00

Korea13

  1. 구글, 업무용 Gemini 에이전트 발표 — 장시간 작업과 기억 유지, Claude 지원
    GeekNews·10/11 15:27
  2. 최신 AI 모델들, 인간의 알고리듬 혁신에 필적하는 성과를 내는 데 어려움 겪어
    GeekNews·05:43
  3. 앤트로픽, '클로드 스타트업' 무료 혜택 일시 중단…신청 폭증에 전면 재검토
    AI타임스·04:46
  4. 앤트로픽, 내부 AI 평가서 실시간 인터넷 차단…“안전장치 검증 때까지”
    AI타임스·03:23
  5. 버그를 환자처럼, 코딩 에이전트를 의료팀처럼 다룬 5개월
    GeekNews·10/11 19:34
  6. “AI와 직장 동료처럼 대화하라”…클로드 코드 개발자의 프롬프트 조언
    AI타임스·04:15
  7. 중학생이 AI(Claude)와 함께 '자기 자신을 통과하지 못하는 입체' 추측을 증명
    GeekNews·10/11 23:35
  8. Show GN: source-to-explainer-video - PR·책·함수를 3분 설명 영상으로 만들어 주는 에이전트 스킬
    GeekNews·10/11 23:07
  9. AI에 관해 ‘불가피하다’고 할 만한 것은 거의 없음
    GeekNews·10/11 23:04
  10. Show GN: 해줘 로컬 - 크레덴셜 금고가 붙은 로컬 AI 에이전트 오케스트레이터
    GeekNews·10/11 23:00
  11. Show GN: OpenGoal - 마크다운만으로 AI 코딩 에이전트의 현재 목표를 관리하는 명세 기반 하네스
    GeekNews·10/11 22:56
  12. Show GN: buttonmasher – 중복 결제와 재시도 버그를 찾아보는 Claude Code 스킬
    GeekNews·10/11 20:54
  13. 할리우드, AI 불안 속 저커버그·머스크·알트먼 영화 잇따라 선보여
    AI타임스·02:54