{"assertions":[{"assertion_index":0,"assertion_text":"Earlier numbers point the same direction: the company has said Claude wrote more than 80 percent of its merged code as of May, and lines merged in 2026 run about 8x the 2021 to 2024 average Anthropic Engineering Nate Jones Import AI.","citation_check":null,"claim_ids":[1350,72446,77235,77236,77237],"link_only_urls":[]},{"assertion_index":1,"assertion_text":"Changes are getting larger as well as more frequent, and a 50-million-line Ruby migration that Stripe and Anthropic say took one day instead of an estimated two-plus months is exactly the kind of output someone still has to check Simon Willison's Weblog AI Engineer.","citation_check":null,"claim_ids":[7455,72340],"link_only_urls":[]},{"assertion_index":2,"assertion_text":"Meta pitches the pair as direct rivals to the leaders, and its own tests put them level with Fable 5.1 and GPT-6 Astra AI Daily Brief TechCrunch AI.","citation_check":null,"claim_ids":[76930,76931,76932,76934,76935,76939,77642],"link_only_urls":[]},{"assertion_index":3,"assertion_text":"TechCrunch's report has Wonder Full reserved for Muse Code subscribers and absent from the API (the interface other software uses to call a model), while Latent Space lists both models as reachable through the Meta Model API and Muse Code Latent Space.","citation_check":null,"claim_ids":[77305,77306,77307],"link_only_urls":[]},{"assertion_index":4,"assertion_text":"On September 30 Google limited Gemini 4 Argon, its newest top model, to government users and trusted cyber defenders in its Fairwind Program, and said it would widen access to developers, enterprises, and consumers as soon as it had feedback from those early testers Google DeepMind Blog Latent Space.","citation_check":null,"claim_ids":[70573,70574,70576,70592,71798,71801,71812],"link_only_urls":[]},{"assertion_index":5,"assertion_text":"Argon is now open to every developer through the API in AI Studio, and to Pro and Ultra subscribers in the Gemini app and AI Mode in Search Google DeepMind Blog.","citation_check":null,"claim_ids":[77451,77452,77453],"link_only_urls":[]},{"assertion_index":6,"assertion_text":"The one outside datapoint, Artificial Analysis's 50 percent score on its AA-Omniscience test (an independent benchmark firm's knowledge test), is a single score with no comparison in this evidence, so it cannot tell you how Argon ranks on your work; it is a reason to test rather than assume Latent Space.","citation_check":null,"claim_ids":[71798,71801,71812],"link_only_urls":[]},{"assertion_index":7,"assertion_text":"By Apple's accounting, the added language is worth as much as two to three times the scarce target text when measured by validation loss (a standard error score on held-out text) and two to thirteen times when measured by downstream accuracy (how well the model does on real tasks afterward) Apple Machine Learning Research.","citation_check":null,"claim_ids":[77260,77261,77262],"link_only_urls":[]},{"assertion_index":8,"assertion_text":"This fits Apple's earlier finding that scarce target text can be reused 15 to 20 times inside a mixture, and Cohere's data-mixing recipe for in-language reasoning Apple Machine Learning Research Cohere Blog.","citation_check":null,"claim_ids":[49528,49530,49531,73616],"link_only_urls":[]},{"assertion_index":9,"assertion_text":"- Meta's developer documentation is the place to settle whether Wonder Full reaches the API or stays inside Muse Code. TechCrunch AI","citation_check":null,"claim_ids":[76930,76931,76932,76934,76935,76939],"link_only_urls":[]},{"assertion_index":10,"assertion_text":"- Independent evaluations of Argon beyond Artificial Analysis's 50 percent AA-Omniscience score are the numbers to set against Google's own claims. Latent Space","citation_check":null,"claim_ids":[71798,71801,71812],"link_only_urls":[]},{"assertion_index":11,"assertion_text":"- Scion is in-house today; whether it ever becomes available outside Anthropic is an open question, and if it did, large engineering teams would have a new tooling option to weigh. Anthropic Engineering","citation_check":null,"claim_ids":[77235,77236,77237],"link_only_urls":[]},{"assertion_index":12,"assertion_text":"- Ollama's 0.40.4 release lets you run decision models (a newer kind of model built to pick between set answers) straight from the terminal with one command, and separately adds Kev decision models to its Apple-silicon runner, which makes a local-versus-cloud comparison cheap to run. Ollama Releases","citation_check":null,"claim_ids":[77769,77770],"link_only_urls":[]},{"assertion_index":13,"assertion_text":"- Nate Jones points to an odd split in OpenAI's unreleased model: it could reconstruct a roughly 16,000-line program just by watching how it behaved, yet it botched a request to adapt an existing library. Test agents on unfamiliar customization work before trusting them with it. Nate Jones","citation_check":null,"claim_ids":[77370,77372,77373],"link_only_urls":[]},{"assertion_index":14,"assertion_text":"Anthropic's merge counts are self-reported, and the material available here gives no definition of a change's size and no defect figure; the 80 percent authorship number reaches this brief through a third party Nate Jones.","citation_check":null,"claim_ids":[72446],"link_only_urls":[]}],"brief_date":"2026-10-14","bundle_id":"evidence-bundle-v2-5b41e325a1a9122072c51cc0e8fff59fe62bf0ba02df12fbe788f035b6f9eebd","claims":[{"assertion_indexes":[0],"citation_authority_claim_id":1350,"citation_checks":[],"claim_id":1350,"claim_text":"Anthropic observed an 8x increase in lines of code merged in 2026 compared to the average of years 2021-2024.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-06-09T06:41:07.708466+00:00","source_name":"Import AI","source_url":"https://importai.substack.com/p/import-ai-460-reward-hacking-society"},{"assertion_indexes":[0,14],"citation_authority_claim_id":72446,"citation_checks":[],"claim_id":72446,"claim_text":"Anthropic reports that Claude has authored more than 80% of the code being merged into its code base as of May.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-06T06:49:10.335386+00:00","source_name":"Nate Jones","source_url":"https://www.youtube.com/watch?v=osZZjdMZVvA"},{"assertion_indexes":[0,11],"citation_authority_claim_id":77235,"citation_checks":[],"claim_id":77235,"claim_text":"Anthropic averaged approximately 2,500 code changes merged per day in May 2026.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:45:53.835911+00:00","source_name":"Anthropic Engineering","source_url":"https://www.anthropic.com/engineering/redesigning-source-control"},{"assertion_indexes":[0,11],"citation_authority_claim_id":77236,"citation_checks":[],"claim_id":77236,"claim_text":"As of September 2026, Anthropic merged over 30,000 code changes per day.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:45:53.835911+00:00","source_name":"Anthropic Engineering","source_url":"https://www.anthropic.com/engineering/redesigning-source-control"},{"assertion_indexes":[0,11],"citation_authority_claim_id":77237,"citation_checks":[],"claim_id":77237,"claim_text":"Anthropic built an in-house version control system named Scion.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:45:53.835911+00:00","source_name":"Anthropic Engineering","source_url":"https://www.anthropic.com/engineering/redesigning-source-control"},{"assertion_indexes":[1],"citation_authority_claim_id":7455,"citation_checks":[],"claim_id":7455,"claim_text":"Coding agents are constructing increasingly large and sophisticated changes.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-07-02T20:16:25.381864+00:00","source_name":"Simon Willison's Weblog","source_url":"https://simonwillison.net/2026/Jul/2/understand-to-participate/#atom-everything"},{"assertion_indexes":[1],"citation_authority_claim_id":72340,"citation_checks":[],"claim_id":72340,"claim_text":"Stripe and Anthropic reported migrating a 50-million-line Ruby codebase in a single day using AI, a task they had estimated would take a team over two months.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-06T06:45:50.820287+00:00","source_name":"AI Engineer","source_url":"https://www.youtube.com/watch?v=_mi3alkqy4s"},{"assertion_indexes":[2,9],"citation_authority_claim_id":76930,"citation_checks":[],"claim_id":76930,"claim_text":"Meta\u2019s Superintelligence Lab released two new models named Muse Wonder and Muse Wonder Full.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-13T06:49:39.417334+00:00","source_name":"TechCrunch AI","source_url":"https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/"},{"assertion_indexes":[2,9],"citation_authority_claim_id":76931,"citation_checks":[],"claim_id":76931,"claim_text":"The Muse Wonder and Muse Wonder Full models represent Meta's first attempt at creating a frontier model.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-13T06:49:39.417334+00:00","source_name":"TechCrunch AI","source_url":"https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/"},{"assertion_indexes":[2,9],"citation_authority_claim_id":76932,"citation_checks":[],"claim_id":76932,"claim_text":"According to Meta's testing, Muse Wonder and Muse Wonder Full achieve benchmark scores comparable to Fable 5.1 and GPT-6 Astra.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-13T06:49:39.417334+00:00","source_name":"TechCrunch AI","source_url":"https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/"},{"assertion_indexes":[2,9],"citation_authority_claim_id":76934,"citation_checks":[],"claim_id":76934,"claim_text":"Muse Wonder Full is exclusively available to Muse Code subscribers and is not served through the API.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-13T06:49:39.417334+00:00","source_name":"TechCrunch AI","source_url":"https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/"},{"assertion_indexes":[2,9],"citation_authority_claim_id":76935,"citation_checks":[],"claim_id":76935,"claim_text":"Both Muse Wonder and Muse Wonder Full models will be used to power the Muse agent.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-13T06:49:39.417334+00:00","source_name":"TechCrunch AI","source_url":"https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/"},{"assertion_indexes":[2,9],"citation_authority_claim_id":76939,"citation_checks":[],"claim_id":76939,"claim_text":"The Muse Wonder and Muse Wonder Full models are the first frontier model releases from Meta since Alexandr Wang restructured the company's AI efforts in August of the previous year.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-13T06:49:39.417334+00:00","source_name":"TechCrunch AI","source_url":"https://techcrunch.com/2026/10/12/meta-releases-its-first-frontier-model-muse-wonder/"},{"assertion_indexes":[2],"citation_authority_claim_id":77642,"citation_checks":[],"claim_id":77642,"claim_text":"Meta announced a new series of Frontier AI models named Wonder, positioning them to compete directly with leading industry models.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:59:03.113658+00:00","source_name":"AI Daily Brief","source_url":"https://podcasters.spotify.com/pod/show/nlw/episodes/Why-Muse-Wonder-Is-Dividing-AI-Users-e3qf7l6"},{"assertion_indexes":[3],"citation_authority_claim_id":77305,"citation_checks":[],"claim_id":77305,"claim_text":"Meta launched the Muse Wonder model family, consisting of Muse Wonder and Muse Wonder Full, described by the source as codenamed \"Watermelon.\"","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:48:09.873867+00:00","source_name":"Latent Space","source_url":"https://www.latent.space/p/ainews-meta-muse-wonder-and-wonder"},{"assertion_indexes":[3],"citation_authority_claim_id":77306,"citation_checks":[],"claim_id":77306,"claim_text":"Meta Superintelligence Labs (MSL) was founded approximately 14 months prior to the October 2026 launch of Muse Wonder.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:48:09.873867+00:00","source_name":"Latent Space","source_url":"https://www.latent.space/p/ainews-meta-muse-wonder-and-wonder"},{"assertion_indexes":[3],"citation_authority_claim_id":77307,"citation_checks":[],"claim_id":77307,"claim_text":"Muse Wonder and Muse Wonder Full are available through the Meta Model API and the Muse Code platform.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:48:09.873867+00:00","source_name":"Latent Space","source_url":"https://www.latent.space/p/ainews-meta-muse-wonder-and-wonder"},{"assertion_indexes":[4],"citation_authority_claim_id":70573,"citation_checks":[],"claim_id":70573,"claim_text":"Google DeepMind announced the release of its new frontier model, Gemini 4 Argon.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-01T06:55:07.814163+00:00","source_name":"Google DeepMind Blog","source_url":"https://deepmind.google/blog/gemini-4-argon-our-next-era-of-frontier-intelligence/"},{"assertion_indexes":[4],"citation_authority_claim_id":70574,"citation_checks":[],"claim_id":70574,"claim_text":"Google DeepMind is rolling out Gemini 4 Argon to trusted cyber defenders through its Fairwind Program.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-01T06:55:07.814163+00:00","source_name":"Google DeepMind Blog","source_url":"https://deepmind.google/blog/gemini-4-argon-our-next-era-of-frontier-intelligence/"},{"assertion_indexes":[4],"citation_authority_claim_id":70576,"citation_checks":[],"claim_id":70576,"claim_text":"Google DeepMind intends to make Gemini 4 Argon available to developers, enterprises, and consumers as soon as possible after gathering feedback from early testers.","correction_role":null,"corroborates_eligible":true,"label":"forecast","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-01T06:55:07.814163+00:00","source_name":"Google DeepMind Blog","source_url":"https://deepmind.google/blog/gemini-4-argon-our-next-era-of-frontier-intelligence/"},{"assertion_indexes":[4],"citation_authority_claim_id":70592,"citation_checks":[],"claim_id":70592,"claim_text":"Google DeepMind plans to release Gemini 4 Argon to paid API customers and Google AI Ultra subscribers before broader availability.","correction_role":null,"corroborates_eligible":true,"label":"forecast","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-01T06:55:07.814163+00:00","source_name":"Google DeepMind Blog","source_url":"https://deepmind.google/blog/gemini-4-argon-our-next-era-of-frontier-intelligence/"},{"assertion_indexes":[4,6,10],"citation_authority_claim_id":71798,"citation_checks":[],"claim_id":71798,"claim_text":"Google DeepMind introduced the Gemini 4 Argon model for coding, enterprise knowledge work, and cyber defense.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-04T06:36:44.096982+00:00","source_name":"Latent Space","source_url":"https://www.latent.space/p/ainews-gemini-4-argon-gdms-answer"},{"assertion_indexes":[4,6,10],"citation_authority_claim_id":71801,"citation_checks":[],"claim_id":71801,"claim_text":"Gemini 4 Argon is initially accessible only to government users and trusted cyber defenders in the Fairwind Program.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-04T06:36:44.096982+00:00","source_name":"Latent Space","source_url":"https://www.latent.space/p/ainews-gemini-4-argon-gdms-answer"},{"assertion_indexes":[4,6,10],"citation_authority_claim_id":71812,"citation_checks":[],"claim_id":71812,"claim_text":"Artificial Analysis reports that Gemini 4 Argon has 50% accuracy on AA-Omniscience.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-04T06:36:44.096982+00:00","source_name":"Latent Space","source_url":"https://www.latent.space/p/ainews-gemini-4-argon-gdms-answer"},{"assertion_indexes":[5],"citation_authority_claim_id":77451,"citation_checks":[],"claim_id":77451,"claim_text":"Gemini 4 Argon is globally available for developers and enterprises via the API in AI Studio and the Gemini platform.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:52:45.197566+00:00","source_name":"Google DeepMind Blog","source_url":"https://deepmind.google/blog/introducing-gemini-4-argon/"},{"assertion_indexes":[5],"citation_authority_claim_id":77452,"citation_checks":[],"claim_id":77452,"claim_text":"Gemini 4 Argon is available for Google AI Ultra subscribers in Google Antigravity.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:52:45.197566+00:00","source_name":"Google DeepMind Blog","source_url":"https://deepmind.google/blog/introducing-gemini-4-argon/"},{"assertion_indexes":[5],"citation_authority_claim_id":77453,"citation_checks":[],"claim_id":77453,"claim_text":"Gemini 4 Argon is available for Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:52:45.197566+00:00","source_name":"Google DeepMind Blog","source_url":"https://deepmind.google/blog/introducing-gemini-4-argon/"},{"assertion_indexes":[7],"citation_authority_claim_id":77260,"citation_checks":[],"claim_id":77260,"claim_text":"In data-constrained bilingual pre-training settings, mixing in data from a high-resource auxiliary language yields larger improvements in validation loss and downstream task accuracy compared to aggressive hyperparameter tuning.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:46:10.291343+00:00","source_name":"Apple Machine Learning Research","source_url":"https://machinelearning.apple.com/research/mix-dont-tune"},{"assertion_indexes":[7],"citation_authority_claim_id":77261,"citation_checks":[],"claim_id":77261,"claim_text":"The performance gap between auxiliary data mixing and hyperparameter tuning grows as model size increases.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:46:10.291343+00:00","source_name":"Apple Machine Learning Research","source_url":"https://machinelearning.apple.com/research/mix-dont-tune"},{"assertion_indexes":[7],"citation_authority_claim_id":77262,"citation_checks":[],"claim_id":77262,"claim_text":"Auxiliary data mixing boosts performance by an amount equivalent to 2\u20133 times the unique target data on validation loss and 2\u201313 times on downstream task accuracy.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:46:10.291343+00:00","source_name":"Apple Machine Learning Research","source_url":"https://machinelearning.apple.com/research/mix-dont-tune"},{"assertion_indexes":[8],"citation_authority_claim_id":49528,"citation_checks":[],"claim_id":49528,"claim_text":"Scarce target corpora can be reused 15\u201320 times in mixture training, with the optimal number of repetitions depending on target data size, compute budget, and model scale.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-21T06:26:38.196445+00:00","source_name":"Apple Machine Learning Research","source_url":"https://machinelearning.apple.com/research/scaling-laws-mixture-pretraining"},{"assertion_indexes":[8],"citation_authority_claim_id":49530,"citation_checks":[],"claim_id":49530,"claim_text":"Mixture training tolerates much higher repetition rates than single-source training.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-21T06:26:38.196445+00:00","source_name":"Apple Machine Learning Research","source_url":"https://machinelearning.apple.com/research/scaling-laws-mixture-pretraining"},{"assertion_indexes":[8],"citation_authority_claim_id":49531,"citation_checks":[],"claim_id":49531,"claim_text":"Too little target data in a mixture underexposes the model to the target domain, while too much target data yields diminishing returns and eventual overfitting due to excessive repetition.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-21T06:26:38.196445+00:00","source_name":"Apple Machine Learning Research","source_url":"https://machinelearning.apple.com/research/scaling-laws-mixture-pretraining"},{"assertion_indexes":[8],"citation_authority_claim_id":73616,"citation_checks":[],"claim_id":73616,"claim_text":"Tiny Aya L2-Thinker was trained using a data mixing strategy of English reasoning data, multilingual reasoning data, and multilingual non-reasoning data.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-07T06:50:25.654580+00:00","source_name":"Cohere Blog","source_url":"https://cohere.com/blog/building-multilingual-bridges"},{"assertion_indexes":[12],"citation_authority_claim_id":77769,"citation_checks":[],"claim_id":77769,"claim_text":"The Ollama v0.40.4 release, published on 2026-10-14, enables the use of decision models directly from the terminal via the `ollama run` command.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T07:04:03.269478+00:00","source_name":"Ollama Releases","source_url":"https://github.com/ollama/ollama/releases/tag/v0.40.4"},{"assertion_indexes":[12],"citation_authority_claim_id":77770,"citation_checks":[],"claim_id":77770,"claim_text":"Kev decision models are now supported by the MLX runner on Apple silicon devices.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T07:04:03.269478+00:00","source_name":"Ollama Releases","source_url":"https://github.com/ollama/ollama/releases/tag/v0.40.4"},{"assertion_indexes":[13],"citation_authority_claim_id":77370,"citation_checks":[],"claim_id":77370,"claim_text":"OpenAI released more than 700 mathematical research manuscripts generated by an internal, unreleased model, including results supported by machine-checked proofs.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:49:31.083179+00:00","source_name":"Nate Jones","source_url":"https://natesnewsletter.substack.com/p/narrow-superintelligence-ai"},{"assertion_indexes":[13],"citation_authority_claim_id":77372,"citation_checks":[],"claim_id":77372,"claim_text":"The same internal AI model that rebuilt a roughly 16,000-line toolkit from its behavior produced an unusable result on a customization task.","correction_role":null,"corroborates_eligible":true,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:49:31.083179+00:00","source_name":"Nate Jones","source_url":"https://natesnewsletter.substack.com/p/narrow-superintelligence-ai"},{"assertion_indexes":[13],"citation_authority_claim_id":77373,"citation_checks":[],"claim_id":77373,"claim_text":"The author forecasts that AI capabilities may become more uneven, with exceptional abilities improving much faster than weaker ones, due to training feedback concentrating gains where it is cheap and plentiful.","correction_role":null,"corroborates_eligible":true,"label":"forecast","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-10-14T06:49:31.083179+00:00","source_name":"Nate Jones","source_url":"https://natesnewsletter.substack.com/p/narrow-superintelligence-ai"}],"version":"evidence-bundle-v2"}