Model Follower.

Up to date

Rebuilt
2026-10-25 06:54 UTC
Release radar
117 on watch
Hub snapshot
2026-10-25
Re-checked
daily · every number sourced and dated

Links / Past days

Monday, 14 September 2026

The AI links the most newsletters and link sources had shared by the end of the day (UTC), each source counted once.

As of 2026-09-14 23:59 UTC Status sealed Rebuilt from issues on file

Reconstructed after the day from the issues on file (dates as published). An issue counts on the date it carries, even if it was collected later.

That day 36-hour window

  1. Shared by 7

    Dario Amodei

    Anthropic CEO Dario Amodei calls for slowing AI capability advancement to let safety measures catch up. Details a three-step plan including embedded third-party evaluators, democratic coordination on safety standards, and global government coordination. cited

    Reported byExponential View, TheSequence, The Rundown AI, Superhuman AI, TLDR AI, AI Breakfast, Turing Post

    Where it was shared
    darioamodei.com announcement · 17 min read First shared 2026-09-13
  2. Shared by 2

    and NASA released the NASA-IBM Lunar Foundation Model, an open-source model combining decades of lunar data from US and…

    “The model helps researchers decide where to investigate. Its predictions about ice still need to be distinguished from direct evidence of ice, especially when they inform exploration plans. Release.” via Turing Post

    Reported byData Points, Turing Post

    How 1 source put it
    • Data Points2026-09-14issue ↗
    • Turing Post2026-09-14issue ↗ The model helps researchers decide where to investigate. Its predictions about ice still need to be distinguished from direct evidence of ice, especially when they inform exploration plans. Release.
    research.ibm.com First shared 2026-09-14
  3. Shared by 2

    ToolGrad: Efficient tool-use dataset generation with textual “gradients”

    “Google Research flipped how you make tool-use training data: ToolGrad builds a verified API chain first, then writes the user question, instead of inventing a request and hoping an agent finds a working path. That answer-first loop hit a 99.8% success rate on 16,000 real…” via TLDR AI

    Reported byTLDR AI, Turing Post

    How 1 source put it
    • TLDR AI2026-09-14issue ↗ Google Research flipped how you make tool-use training data: ToolGrad builds a verified API chain first, then writes the user question, instead of inventing a request and hoping an agent finds a working path. That answer-first loop hit a 99.8% success rate on 16,000 real…
    • Turing Post2026-09-14issue ↗
    research.google First shared 2026-09-14
  4. Shared by 3

    Real-SWE: Benchmarking AI models on private, real-world, enterprise codebases

    “The new Real-SWE benchmark tests AI models on complex tasks using private enterprise codebases, reflecting real software engineering conditions. With a maximum resolution rate of 38.8%, agents face challenges such as proprietary systems and business-specific coding conventions.” via TLDR AI

    Reported byTLDR AI, AI Breakfast

    Discussed onHacker News 275 pts · 158 comments

    How 1 source put it
    • TLDR AI2026-09-14issue ↗ The new Real-SWE benchmark tests AI models on complex tasks using private enterprise codebases, reflecting real software engineering conditions. With a maximum resolution rate of 38.8%, agents face challenges such as proprietary systems and business-specific coding conventions.
    • AI Breakfast2026-09-14issue ↗
    withspecific.com First shared 2026-09-12
  5. Shared by 2

    Altman ruled out a 2026 OpenAI IPO, calling it an "ill-advised moment" given safety concerns

    “OpenAI CEO Sam Altman told Fortune that the company won’t IPO in 2026, saying right now would be an “ill-advised moment” to go public, given the AI safety concerns.” via The Rundown AI

    Reported byThe Rundown AI, AI Breakfast

    How 1 source put it
    • The Rundown AI2026-09-14issue ↗ OpenAI CEO Sam Altman told Fortune that the company won’t IPO in 2026, saying right now would be an “ill-advised moment” to go public, given the AI safety concerns.
    • AI Breakfast2026-09-14issue ↗
    fortune.com First shared 2026-09-14
  6. Shared by 1

    A single firm is behind OpenAI, Anthropic, and Meta hacking scandals

    Discussed onHacker News 694 pts · 255 comments

    effort.news First shared 2026-09-14
  7. Shared by 2

    released GPT-Live-1, a voice model that listens and speaks simultaneously rather than chaining separate speech-to-text,…

    “is a full-duplex voice model that listens and speaks at once and hands hard reasoning to a backend model like Astra. $0.05 per minute, 12 new voices, and roughly 30 points better than GPT-Realtime-2.1 on Full Duplex Bench.” via AI Breakfast

    Reported byAI Breakfast, Data Points

    How 1 source put it
    • AI Breakfast2026-09-11issue ↗ is a full-duplex voice model that listens and speaks at once and hands hard reasoning to a backend model like Astra. $0.05 per minute, 12 new voices, and roughly 30 points better than GPT-Realtime-2.1 on Full Duplex Bench.
    • Data Points2026-09-14issue ↗
    openai.com First shared 2026-09-11
  8. Shared by 2

    A Misalignment of AI in Mathematics

    Twenty-five Fields Medalists sign a declaration warning that AI benchmarks misalign with the mathematical community's goals. Details how rushing AI solutions destroys the human transmission chain needed to integrate new methods into the mathematical canon. cited

    Reported byData Points

    Discussed onHacker News 151 pts · 10 comments

    terrytao.wordpress.com announcement First shared 2026-09-11
  9. Shared by 1

    DeepSeek reports it beats Anthropic’s Opus-5 and OpenAI’s GPT-5.6 Sol on Terminal-Bench 2.1 (90.6 vs 89.1 and 88.8) and…

    “…, high, and max reasoning tiers. DeepSeek reports it beats Anthropic’s Opus-5 and OpenAI’s GPT-5.6 Sol on Terminal-Bench 2.1 (90.6 vs 89.1 and 88.8) and DeepSWE v1.1, though it trails both on Terminal-Bench 4.0 and GPQA Diamond. The smaller KV cache makes long-context, agentic workloads cheaper to serve; the MIT license makes it self-hostable.…” via Data Points

    Reported byData Points

    AboutDeepSeek-V4-Flash-Vision-Exp, DeepSeek-V4-Flash-DSpark

    marktechpost.com First shared 2026-09-14
  10. Shared by 1

    Both are available through an OpenAI-compatible API; no open weights, and the service excludes the EU/EEA.

    “…g versus 27.3 for Opus 5, and 74.3 on DeepSWE. Both are available through an OpenAI-compatible API; no open weights, and the service excludes the EU/EEA. The routing approach lets developers get near-frontier results without locking into one vendor, though several cited benchmarks (including SWEFish) are Sakana’s own and unverified externally.…” via Data Points

    Reported byData Points

    marktechpost.com First shared 2026-09-14
  11. Shared by 1

    The response format replaces flat chunks with a document-page-block tree carrying stable IDs, standardized markdown,…

    “…an cost more than before. The response format replaces flat chunks with a document-page-block tree carrying stable IDs, standardized markdown, and grounding data down to bounding boxes and word-level confidence, enabling coordinate-based redaction and citation-backed extraction. (Disclosure: LandingAI is part of the DeepLearning.AI ecosystem.)…” via Data Points

    Reported byData Points

    marktechpost.com First shared 2026-09-14
  12. Shared by 2

    Anthropic's Mythos 5 spent hundreds of pages fighting CAPTCHA

    “Anthropic says Claude Mythos 5 reached the live internet during a security test and uploaded a malicious package to PyPI, the public index for Python software. The agent spent roughly 150 pages of internal reasoning trying to pass CAPTCHAs before 15 systems downloaded the package. The Senate Opens a Probe Into…” via Forward Future

    Reported byForward Future, TLDR AI

    How 1 source put it
    • Forward Future2026-09-11issue ↗ Anthropic says Claude Mythos 5 reached the live internet during a security test and uploaded a malicious package to PyPI, the public index for Python software. The agent spent roughly 150 pages of internal reasoning trying to pass CAPTCHAs before 15 systems downloaded the package. The Senate Opens a Probe Into…
    • TLDR AI2026-09-14issue ↗
    techcrunch.com First shared 2026-09-11
  13. Shared by 2

    AI researchers debate how close we are to recursive self-improvement

    “This post features a transcript of a podcast with Beren Millidge, the CTO of Zyphra, John Schulman, the chief scientist at Thinking Machines and a co-founder of OpenAI, and Charlie O'Neill, head of model training at Baseten. The episode uncovers the details of what's…” via TLDR AI

    Reported byTLDR AI

    Discussed onHacker News 121 pts · 116 comments

    dwarkesh.com First shared 2026-09-11
  14. Shared by 2

    OpenAI Pushes Its IPO Beyond 2026

    “Sam Altman said OpenAI would not go public in 2026, arguing that current AI safety concerns made an IPO ill-advised. The company had previously filed confidentially and was reportedly considering a 2027 listing instead.” via TLDR AI

    Reported byTLDR AI

    Discussed onHacker News 105 pts · 66 comments

    techcrunch.com First shared 2026-09-12
  15. Shared by 1

    Cursor Projects lets users take on larger bodies of work.

    Reported byTLDR AI

    cursor.com First shared 2026-09-14
  16. Shared by 1

    SoftBank Gets Upsized $11.9 Billion Loan in OpenAI Funding Push

    “SoftBank just borrowed nearly $12 billion from about 20 banks to keep funding OpenAI, beating the $10 billion it first sought. Son is still aiming near $65 billion into OpenAI by October even as Altman shelves a 2026 IPO over safety and SoftBank shares sank as much as 13% Monday…” via TLDR AI

    Reported byTLDR AI

    japantimes.co.jp First shared 2026-09-14
  17. Shared by 1

    A cache hit is not proof that you skipped the work

    “A cache hit can be true and still fail to prove that work was skipped. A cache event becomes evidence when the independent oracle expects the prefix, the engine attests it, the prompt path skips it, the output stays identical, the evaluator passes, and the verifier binds those facts to the…” via TLDR AI

    Reported byTLDR AI

    siddhantkhare.com First shared 2026-09-14
  18. Shared by 1

    How Good Are Frontier Models at Physics? Expert Re-Grading Reveals Broken Evaluations and Near-Saturation of Leading Benchmarks

    “Those scary physics scores were often the test's fault. Experts rechecked six popular benchmarks and found wrong answer keys, fuzzy questions, and grader bugs behind most model “fails.” Clean them up and frontier…” via TLDR AI

    Reported byTLDR AI

    arxiv.org First shared 2026-09-14
  19. Shared by 1

    Sakana: Fugu Ultra v2

    “Fugu Ultra v2 is the higher-performance model in Sakana AI's Fugu family. It uses a language model trained to route tasks across a fixed pool of open and specialized models and to recursively call instances of itself. The model prioritizes answer quality on complex multi-step reasoning, autonomous research, and…” via TLDR AI

    Reported byTLDR AI

    openrouter.ai First shared 2026-09-14
  20. Shared by 1

    The Recurrent Looped Transformer combines a causal encoder with a recurrent decoder that carries its final hidden state…

    Reported byTLDR AI

    yifanzhang-pro.github.io First shared 2026-09-14

The week to that day two or more sources

  1. Shared by 7

    Dario Amodei

    Anthropic CEO Dario Amodei calls for slowing AI capability advancement to let safety measures catch up. Details a three-step plan including embedded third-party evaluators, democratic coordination on safety standards, and global government coordination. cited

    Reported byExponential View, TheSequence, The Rundown AI, Superhuman AI, TLDR AI, AI Breakfast, Turing Post

    Where it was shared
    darioamodei.com announcement · 17 min read First shared 2026-09-13
  2. Shared by 5

    Countering misuse of AI: September 2026 / Anthropic

    “Anthropic's Threat Intelligence team has identified and disrupted several operations in the past several months where threat actors tried to use Claude for malicious activity. This report shares case studies from those operations and describes how malicious use of Claude has evolved since…” via TLDR AI

    Reported byThe Rundown AI, Superhuman AI, TLDR AI, AI Breakfast, Interconnects

    How 1 source put it
    anthropic.com 156 min read First shared 2026-09-11
  3. Shared by 3

    Real-SWE: Benchmarking AI models on private, real-world, enterprise codebases

    “The new Real-SWE benchmark tests AI models on complex tasks using private enterprise codebases, reflecting real software engineering conditions. With a maximum resolution rate of 38.8%, agents face challenges such as proprietary systems and business-specific coding conventions.” via TLDR AI

    Reported byTLDR AI, AI Breakfast

    Discussed onHacker News 275 pts · 158 comments

    How 1 source put it
    • TLDR AI2026-09-14issue ↗ The new Real-SWE benchmark tests AI models on complex tasks using private enterprise codebases, reflecting real software engineering conditions. With a maximum resolution rate of 38.8%, agents face challenges such as proprietary systems and business-specific coding conventions.
    • AI Breakfast2026-09-14issue ↗
    withspecific.com First shared 2026-09-12
  4. Shared by 4

    OpenAI Launches the Agents API

    “OpenAI introduced the Agents API in public beta, giving developers access to the managed agent harness and infrastructure behind Codex. It handles context, tools, subagents, persistent execution, files, and code environments for agents that can run for extended periods.” via TLDR AI

    Reported byTLDR AI, AI Breakfast, TheSequence, Turing Post

    How 1 source put it
    • TLDR AI2026-09-11issue ↗ OpenAI introduced the Agents API in public beta, giving developers access to the managed agent harness and infrastructure behind Codex. It handles context, tools, subagents, persistent execution, files, and code environments for agents that can run for extended periods.
    • AI Breakfast2026-09-11issue ↗
    • TheSequence2026-09-13issue ↗
    • Turing Post2026-09-14issue ↗
    openai.com First shared 2026-09-11
  5. Shared by 4

    OpenAI Launches ChatGPT for Financial Services

    “OpenAI launched ChatGPT for Financial Services, an edition of ChatGPT Work with PitchBook, Crunchbase, and LSEG data baked in for valuation models and pitch decks.” via The Rundown AI

    Reported byThe Rundown AI, Superhuman AI, TLDR AI, AI Breakfast

    How 1 source put it
    openai.com First shared 2026-09-11
  6. Shared by 3

    OpenAI puts Pro subscriptions on hold due to Astra demand

    “OpenAI has paused subscriptions for its $200-per-month Pro plan. The company's Astra model is now rolling out to Pro, Plus, Enterprise, and Business accounts. The model promises a major leap forward in reasoning, coding, and computer use. OpenAI says the model is the beginning of the…” via TLDR AI

    Reported byTLDR AI, AI Breakfast, Forward Future

    How 1 source put it
    • TLDR AI2026-09-11issue ↗ OpenAI has paused subscriptions for its $200-per-month Pro plan. The company's Astra model is now rolling out to Pro, Plus, Enterprise, and Business accounts. The model promises a major leap forward in reasoning, coding, and computer use. OpenAI says the model is the beginning of the…
    • AI Breakfast2026-09-11issue ↗
    • Forward Future2026-09-14issue ↗
    techcrunch.com First shared 2026-09-11
  7. Shared by 2

    ToolGrad: Efficient tool-use dataset generation with textual “gradients”

    “Google Research flipped how you make tool-use training data: ToolGrad builds a verified API chain first, then writes the user question, instead of inventing a request and hoping an agent finds a working path. That answer-first loop hit a 99.8% success rate on 16,000 real…” via TLDR AI

    Reported byTLDR AI, Turing Post

    How 1 source put it
    • TLDR AI2026-09-14issue ↗ Google Research flipped how you make tool-use training data: ToolGrad builds a verified API chain first, then writes the user question, instead of inventing a request and hoping an agent finds a working path. That answer-first loop hit a 99.8% success rate on 16,000 real…
    • Turing Post2026-09-14issue ↗
    research.google First shared 2026-09-14
  8. Shared by 2

    Altman ruled out a 2026 OpenAI IPO, calling it an "ill-advised moment" given safety concerns

    “OpenAI CEO Sam Altman told Fortune that the company won’t IPO in 2026, saying right now would be an “ill-advised moment” to go public, given the AI safety concerns.” via The Rundown AI

    Reported byThe Rundown AI, AI Breakfast

    How 1 source put it
    • The Rundown AI2026-09-14issue ↗ OpenAI CEO Sam Altman told Fortune that the company won’t IPO in 2026, saying right now would be an “ill-advised moment” to go public, given the AI safety concerns.
    • AI Breakfast2026-09-14issue ↗
    fortune.com First shared 2026-09-14
  9. Shared by 2

    OpenAI Pushes Its IPO Beyond 2026

    “Sam Altman said OpenAI would not go public in 2026, arguing that current AI safety concerns made an IPO ill-advised. The company had previously filed confidentially and was reportedly considering a 2027 listing instead.” via TLDR AI

    Reported byTLDR AI

    Discussed onHacker News 105 pts · 66 comments

    techcrunch.com First shared 2026-09-12
  10. Shared by 2

    Anthropic's Mythos 5 spent hundreds of pages fighting CAPTCHA

    “Anthropic says Claude Mythos 5 reached the live internet during a security test and uploaded a malicious package to PyPI, the public index for Python software. The agent spent roughly 150 pages of internal reasoning trying to pass CAPTCHAs before 15 systems downloaded the package. The Senate Opens a Probe Into…” via Forward Future

    Reported byForward Future, TLDR AI

    How 1 source put it
    • Forward Future2026-09-11issue ↗ Anthropic says Claude Mythos 5 reached the live internet during a security test and uploaded a malicious package to PyPI, the public index for Python software. The agent spent roughly 150 pages of internal reasoning trying to pass CAPTCHAs before 15 systems downloaded the package. The Senate Opens a Probe Into…
    • TLDR AI2026-09-14issue ↗
    techcrunch.com First shared 2026-09-11