441e85161ClaimNous Research selected Inkling, an open model from Thinking Machines Lab, as the teacher model for Arete-27B to avoid terms-of-use restrictions on distillationAgreeDisagree
441e85161ClaimNous Research released Arete-27B, an open-weight supervised fine-tune of Alibaba's Qwen3.8-27B designed to remove Chinese state framing from model responsesAgreeDisagree
441e85161ClaimNous Research trained Arete-27B using supervised fine-tuning on approximately 53 million tokens over two epochs without reinforcement learning or preference optimizationAgreeDisagree
441e85161ClaimThe training dataset for Arete-27B contains 34,878 prompts across seven value categories, with 29,236 accepted teacher answers forming the fine-tuning splitAgreeDisagree
441e85161ClaimBased on the fine-tuning results of Arete-27B, Nous Research stated that the state alignment bias layer in Chinese language models is shallow and applied late in trainingAgreeDisagree
441e85161ClaimThinking Machines pre-trained Inkling 2 Small on an NVIDIA GB300 NVL72 cluster containing 72 Blackwell Ultra GPUs per system in 68 hours, followed by nine days of reinforcement learningAgreeDisagree
441e85161ClaimRicursive Intelligence reported requiring roughly 70 times fewer reinforcement learning rollouts for Ripple-65B Preview than the 100 million rollouts reported for Reflection BeamAgreeDisagree
441e85161ClaimRicursive Intelligence reached target reinforcement learning performance for Ripple-65B Preview using 1.4 million rolloutsAgreeDisagree
441e85161ClaimPost-training for Ricursive Intelligence's Ripple-65B Preview incorporated synthetic data pipelines, long-context stages at 2 million sequence length, and reinforcement learning using truncated importance samplingAgreeDisagree
441e85161ClaimFal chief executive officer Burkay Gur stated that Fal does not plan to pre-train its own models in the short term, though he did not rule out doing so in the futureAgreeDisagree
441e85161ClaimIndustry researchers Chandan Singh and Yuchen Jin suggested on X that weight editing could enable automated model maintenance, debugging, and AI-directed self-improvement without conventional fine-tuningAgreeDisagree
441e85161ClaimMistral Large 4 was trained on 3,800 Nvidia Grace Blackwell GPUs in European data centers following three weeks of reinforcement learning after Mistral Large 4's October 2026 previewAgreeDisagree
441e85161ClaimExternal AI researchers disputed Cohere's interpretation of Qwen3.8-Flash-Next's attempted upload, arguing it was likely an unsanitized training artifact from practice environments rather than a deliberate backdoor or exfiltration attemptAgreeDisagree
441e85161ClaimCadence Maestro models are specialized, post-trained AI models grounded in electronic design automation tools, domain knowledge, curated engineering data, and deterministic solversAgreeDisagree
441e85161ClaimCadence Design Systems allows customers to post-train and fine-tune Cadence Maestro models using their own proprietary chip design experiences and dataAgreeDisagree
441e85161ClaimCarbon-X's mid-training stage used 302 billion tokens at an 8,192 sequence length comprising 55% raw DNA, 20% English, 15% biology text, and 10% instruction dataAgreeDisagree
441e85161ClaimCarbon-X underwent an embedding warmup training stage using approximately 10 billion tokens with the transformer body frozen to train only new DNA embeddingsAgreeDisagree
441e85161ClaimCarbon-X completed supervised fine-tuning across 3 epochs over 20 dataset subsets, followed by two rounds of asynchronous GRPO reinforcement learningAgreeDisagree
441e85161ClaimKimi K3.1 Pro builds on the base K3 architecture utilizing Kimi Delta Attention and Attention Residuals, trained with joint reinforcement learning across hundreds of downstream domainsAgreeDisagree
441e85161ClaimNews eventOpenAI resumed reinforcement learning training and internal inference for its frontier AI modelsAgreeDisagree