NAll news

All news

Thursday30 items

  1. Phoenix TechResearch
    AI score75

    Anthropic publishes Claude protein design study: 310,000 designs generated in 64 hours, 17 validated experimentally

    Anthropic has published a biology study in which Claude autonomously invoked 12 protein design tools within 64 hours, generating 318,484 candidate protein designs targeting 5 scorpion venom neurotoxins, and screened out 211 to be handed over to Adaptyv Bio and Twist Bioscience laboratories for validation, with 17 ultimately meeting the preset binding criteria. Among them, the dissociation constants KD of Cn2-B1 and Cn2-B2 targeting the Cn2 toxin reached single-digit nanomolar levels, but the experiments only validated binding ability and have not yet proven that they can neutralize the toxin or provide protection in vivo. The research team set the goals and screening rules; after Claude ran autonomously for 24 consecutive hours, the researchers added 4 additional prompts; the computational design phase consumed a total of about 10,000 GPU hours; and all 211 candidate designs and experimental data have been made public.

    #Papers/Research#Agents#Anthropic

  2. LMSYS BlogFirst-partyResearch
    AI score57

    LMSYS uses Miles to train a SWE agent on MI355X, SWE-bench Verified…

    LMSYS used Miles, SGLang, Megatron-LM, and Harbor to perform fully asynchronous agent reinforcement learning on Qwen3-30B-A3B-Instruct on AMD Instinct MI355X, raising pass@1 from 7.3% to 27.5% on a 100-task stratified subset of SWE-bench Verified. Training was conducted in two stages: first SFT on approximately 4,000 public trajectories from Open-SWE-Traces, then GRPO on approximately 1,350 SWE-smith tasks. After fixing the model and RL recipe, scaling inference from 1 node to 3 8-GPU nodes, switching to E2B cloud sandboxes, and removing the concurrency limit increased rollout throughput from approximately 860 to approximately 3,480 trajectories/hour; limiting trajectory length on the same 8-GPU node reduced single-step training time from 850 seconds to 284 seconds.

    Agentic RL loop: SGLang generates actions, a CPU sandbox executes and grades them, and Megatron-LM trains on the resulting trajectories.

    #Papers/Research#Agents#English text is being prepared.

  3. Sierra：BlogFirst-partyResearch
    AI score58

    Sierra launches 𝜏-i18n multilingual voice agent evaluation benchmark

    Sierra releases 𝜏-i18n, extending 𝜏-Voice to five languages: Spanish, Brazilian Portuguese, Hindi, Korean, and Mandarin, retaining the same airline, retail, and telecom tasks, policies, tools, and success criteria to enable cross-language comparison. Across 2,700 main calls, task completion rates for Spanish, Hindi, and Portuguese lag English by an average of 2.4, 3.8, and 6.4 percentage points, while Korean and Mandarin lag by 22.6 and 9.1 percentage points respectively; Korean is the weakest language among the three agents. In addition to task completion, 𝜏-i18n separately reports interaction metrics (response timing, selectivity, tool-call efficiency) and generation quality (naturalness, speech fidelity), and has open-sourced the codebase and the language pack framework LanguageFactory.

    Task completion, interaction, and generation

    #Evaluations / benchmarks#Speech#Multimodal

  4. Unite.AIResearch
    AI score76

    OpenAI, PerturbAI, and Scripps use GPT-Rosalind agents to build a mouse brain reasoning atlas

    On October 21, 2026, OpenAI announced that, with PerturbAI and Scripps Research, it used thousands of AI agents driven by GPT-Rosalind to analyze existing experimental data from millions of mouse brain cells within 32 minutes and aggregated the reports of each agent into a public reasoning atlas. The method is called distributed biological reasoning, with each agent responsible for one gene perturbation, analyzing cellular responses, retrieving literature, and proposing explanations. The team says equivalent manual analysis and literature review would take months to a year; under fixed models and resources, a single agent handling the maximum workload cited only 11 papers, while distributed agents cited more than 700, and in blind evaluation human experts preferred the distributed agent reports in 75% of cases. A companion preprint was released on bioRxiv on October 16, 2026, and has not yet undergone peer review; the reasoning atlas and the Distributed Agents framework are publicly available on GitHub under the MIT license, and the dataset is the whole-brain CRISPR Perturb-seq atlas from PerturbAI on Hugging Face.

    Same story · 5 reports

    #Papers/Research#Agents#English text is being prepared.

  5. Unite.AIResearch
    AI score84

    Microsoft researchers and collaborators prove the 31-year-old KLS conjecture

    Microsoft Research announced that senior researcher Xinzhi Zhang and collaborator Zhao Song, through a human-led agentic AI research workflow, proved the KLS conjecture proposed in 1995; the paper "An O(1) Bound for the KLS Constant" was submitted to arXiv on October 1 and revised on October 4. During the research process, AI agents explored more than 100 candidate paths, but when the authors gave multiple leading models the same token budget for unguided attempts, they failed to achieve the goal. In the same period, there were also independent proofs by Bizeul, Klartag, and Lehec, as well as the Lean 4 formalization project KLS500, which completed machine verification.

    Same story · 3 reports

    #Papers/Research#Agents#Reasoning

  7. Greg BrockmanResearch
    AI score58

    Greg Brockman shares new lead from GPT-Rosalind in drug discovery

    Greg Brockman shared OpenAI's progress on GPT-Rosalind: the model has surfaced a new lead in drug discovery, namely whether restoring cholesterol transport could protect neurons from degeneration. OpenAI says DCTN1 is a gene associated with Perry syndrome, and that disrupting it in mouse brain data increased the expression of genes involved in cholesterol synthesis; on this basis, GPT-Rosalind hypothesized that cholesterol may not be reaching where neurons need it. In collaboration with PerturbAI and Scripps Research, thousands of GPT-Rosalind agents worked in parallel, linking gene expression results from millions of mouse brain cells with scientific literature within 32 minutes, and OpenAI says it is expanding this collaboration to find new drug targets.

    Quoted postOpenAI · @OpenAI

    GPT-Rosalind surfaced a new lead for drug discovery: could restoring cholesterol transport protect neurons from degeneration? DCTN1 is a gene linked to Perry syndrome, a rare neurodegenerative disorder. In mouse-brain data, disrupting it increased expression of genes involved in making cholesterol. GPT-Rosalind suggested a possible explanation: cholesterol may not be reaching where neurons need it. That gives scientists a specific mechanism to test and, if confirmed, a direction for developing new treatments for neurodegenerative disorders. With @perturbai_tx and @scrippsresearch, a team of thousands of GPT-Rosalind agents worked in parallel, connecting gene-expression results from millions of mouse brain cells with scientific literature in just 32 minutes. We’re now expanding this collaboration to identify new drug targets. https://t.co/oHdF3ELsMj

    Same story · 5 reports

    #English text is being prepared.#Papers/Research#Agents

  8. Microsoft ResearchFirst-partyResearch
    AI score78

    Microsoft Research proves the 31-year-old unsolved KLS conjecture

    Microsoft Research researcher Xinzhi Zhang and co-authors proved the KLS conjecture proposed in 1995, giving a dimension-independent fixed bound; the previous best result only allowed the bound to grow slowly with dimension. The proof came from a human-guided agent search process, in which the team explored more than 100 candidate paths, with researchers evaluating, pruning, and redirecting them; an independent team also gave a complete proof at the same time, followed by Lean 4 formal verification. The authors say earlier unguided end-to-end attempts with multiple frontier models failed to achieve the goal.

    Same story · 3 reports

    #Papers/Research#Reasoning#Microsoft

  9. Google ResearchFirst-partyResearch
    AI score64

    Google Research uses AI to assist blind-sweep ultrasound assessment of gestational age, with error comparable to experts

    Google Research published a study in JAMA Network Open; its AI gestational age assessment model based on blind-sweep ultrasound achieved a mean absolute error of 4.18 days in prospective testing in Chicago (n=192) and Nairobi (n=193), reaching non-inferiority compared with expert sonographers' 4.48 days. The model was fine-tuned only on data from 120 participants in Chicago and did not use Nairobi data; the errors in the two locations were 4.10 days and 4.27 days, respectively. The study used a Clarius C3 handheld probe, with standardized blind sweeps performed by operators without professional ultrasound training.

    Bar chart comparing the mean absolute error between the AI model and expert clinical standards across multiple cohorts.

    #Papers/Research#Image generation#Google

Wednesday22 items

  4. Rohan PaulResearch
    AI score76

    OpenAI and others use 1,946 agents to interpret a mouse whole-brain CRISPR atlas in 32 minutes

    OpenAI, PerturbAI, and Scripps Research used 1,946 GPT-Rosalind-5.5 agents to complete interpretation of a mouse whole-brain CRISPR atlas in 32 minutes; OpenAI estimates that manual work would take months. Each agent analyzed only the effect of a single gene knockout and retrieved literature, and did not access the atlas paper published after the model's December 2025 knowledge cutoff, yet still matched or refined 48 of 55 published findings supportable by the expression data. The core of the method is to split one giant analysis into nearly 2,000 single-gene small tasks executed in parallel; a single agent processing 893 genes cited only 11 papers, while after splitting the citation count was 711 times that, and blinded PhD reviewers preferred the split reports in 75% of cases.

    Quoted postOpenAI · @OpenAI

    GPT-Rosalind surfaced a new lead for drug discovery: could restoring cholesterol transport protect neurons from degeneration? DCTN1 is a gene linked to Perry syndrome, a rare neurodegenerative disorder. In mouse-brain data, disrupting it increased expression of genes involved in making cholesterol. GPT-Rosalind suggested a possible explanation: cholesterol may not be reaching where neurons need it. That gives scientists a specific mechanism to test and, if confirmed, a direction for developing new treatments for neurodegenerative disorders. With @perturbai_tx and @scrippsresearch, a team of thousands of GPT-Rosalind agents worked in parallel, connecting gene-expression results from millions of mouse brain cells with scientific literature in just 32 minutes. We’re now expanding this collaboration to identify new drug targets. https://t.co/oHdF3ELsMj

    Same story · 5 reports

    #Papers/Research#Agents#OpenAI

  10. 36Kr · AIResearch
    AI score66

    Mass General Brigham team publishes FAHR-Face study in The Lancet Digital Health, using facial photos to predict mortality risk in cancer patients

    The U.S. Mass General Brigham health care system developed the FAHR-Face model, which identifies differences in mortality risk among cancer patients using only a single facial photograph; the study was published on October 20 in The Lancet Digital Health. Among 3,820 cancer patients who did not participate in training, 64 of the 122 people rated by the model as extremely high risk died within 90 days after the photo was taken, while only 20 of the 975 people with scores between 0 and 0.1 died, a rate of 2.1%. The study also conducted external validation using 4,892 patients at the Maastro radiotherapy center in the Netherlands and obtained similar results, but the model cannot yet be used directly in clinical practice.

    #Papers/Research#Multimodal#English text is being prepared.