home.social

#ai-2 — Public Fediverse posts

Live and recent posts from across the Fediverse tagged #ai-2, aggregated by home.social.

fetched live
  1. #Ai2 has released #MolmoWeb, an #openweight #visualwebagent that operates from browser screenshots and executes actions like clicking, typing, and scrolling. It comes with #MolmoWebMix, a dataset of 30,000 human task trajectories and 2.2 million screenshot question-answer pairs, making it the largest publicly released collection of human web-task execution. venturebeat.com/data/ai2-relea #tech #media #news

  2. #Ai2 has released #MolmoWeb, an #openweight #visualwebagent that operates from browser screenshots and executes actions like clicking, typing, and scrolling. It comes with #MolmoWebMix, a dataset of 30,000 human task trajectories and 2.2 million screenshot question-answer pairs, making it the largest publicly released collection of human web-task execution. venturebeat.com/data/ai2-relea #tech #media #news

  3. New Bolmo architecture lets you train language models directly on bytes, cutting tokenization overhead and boosting multilingual performance. Built on Ai2’s Dolma 3 and Olmo pipelines, it promises faster, cheaper LLM research. Curious how byte‑level training reshapes AI? Dive into the details. #Bolmo #ByteLevel #Ai2 #Multilingual

    🔗 aidailypost.com/news/bolmo-arc

  4. AI2 just dropped Olmo 3.1 (32B), a new RL‑trained model that jumps +5 points on AIME and +4 on ZebraLogic. Early IFEval results show stronger reasoning and code skills, and the weights are open‑source for the community to build on. Dive into the benchmarks and see what’s next for open‑source LLMs! #AI2 #Olmo3_1 #AIME #ZebraLogic

    🔗 aidailypost.com/news/ai2-relea

  5. The #AllenInstitute for AI (#Ai2) released #Olmo3, a collection of #openlanguage models designed to compete with industry leaders. Olmo 3 outperforms fully open models like #Stanford’s #Marin and commercial open-weight models like #Meta’s #Llama3.1, offering improved performance, efficiency, and transparency. geekwire.com/2025/ai2-releases #AIagent #AI #ML #NLP #LLM #GenAI

  6. The #AllenInstitute for AI (#Ai2) released #Olmo3, a collection of #openlanguage models designed to compete with industry leaders. Olmo 3 outperforms fully open models like #Stanford’s #Marin and commercial open-weight models like #Meta’s #Llama3.1, offering improved performance, efficiency, and transparency. geekwire.com/2025/ai2-releases #AIagent #AI #ML #NLP #LLM #GenAI

  7. AI2's new Olmo 3 family pushes the frontier, rivaling Qwen and Llama while adding open-reasoning tools and full data transparency with OlmoTrace and Dolma 3. Discover how this open-source-focused LLM could reshape research and deployment. #Olmo3 #Ai2 #Qwen #Llama

    🔗 aidailypost.com/news/ai2s-olmo

  8. #AI2 Skylight app opens parts of the platform for public use app.skylight.earth/, anyone can explore AI powered detections of vessel activity at sea #ai4good

  9. Galileo, an open-source multimodal model from #AI2 designed to process many kinds of earth observations at once—driving applications such as mapping agricultural land, detecting floods, and monitoring marine pollution nasaharvest.org/news/galileo-i #ai4good #tech4good

  10. 🚀 #NSF and #NVIDIA just handed #Ai2 a mountain of cash 💰 to build an "open AI #ecosystem," whatever that means. Meanwhile, the rest of us are still trying to figure out how to open a PDF. 😅 Looks like AI is going green 🌍, because it’s sure as hell not going open source.
    allenai.org/blog/nsf-nvidia #OpenAI #AIInnovation #HackerNews #ngated

  11. 🚀 #NSF and #NVIDIA just handed #Ai2 a mountain of cash 💰 to build an "open AI #ecosystem," whatever that means. Meanwhile, the rest of us are still trying to figure out how to open a PDF. 😅 Looks like AI is going green 🌍, because it’s sure as hell not going open source.
    allenai.org/blog/nsf-nvidia #OpenAI #AIInnovation #HackerNews #ngated

  12. Allen Institute for AI (Ai2) researchers are developing a type of LLM called FlexOlmo that enables data owners to control how data is used to train models, and it is possible to extract the data out of the LLM later on.

    This will provide data owners with greater flexibility to manage and balance data privacy and control over where data is used in LLM training. link.wired.com/public/40617345 #AI #LLM #LLMTraining #Ai2 #FlexOlmo #DataPrivacy ##CopyRight #LLMData

  13. Allen Institute for AI (Ai2) researchers are developing a type of LLM called FlexOlmo that enables data owners to control how data is used to train models, and it is possible to extract the data out of the LLM later on.

    This will provide data owners with greater flexibility to manage and balance data privacy and control over where data is used in LLM training. link.wired.com/public/40617345 ##CopyRight

  14. #SciArena is a crowdsourced platform built by the #AllenInstituteForAI (#Ai2, @allenai) for testing #AI tools on scientific research tasks.
    sciarena.allen.ai/

    You can find its latest results in a July 1 blog post. If you use any AI tools for research purposes, note that the SciArena tool rankings differ by discipline.
    allenai.org/blog/sciarena

  15. #SciArena is a crowdsourced platform built by the #AllenInstituteForAI (#Ai2, @allenai) for testing #AI tools on scientific research tasks.
    sciarena.allen.ai/

    You can find its latest results in a July 1 blog post. If you use any AI tools for research purposes, note that the SciArena tool rankings differ by discipline.
    allenai.org/blog/sciarena

  16. Meet Atlantes, allenai.org/blog/atlantes, the first AI-powered GPS model for real-time maritime intelligence #AI2 #AI4Good

  17. Recent breakthroughs like #OlympicCoder outperform #Claude3.7 on coding tasks with just 7B parameters, while #AI2's #OLMo2 models match #OpenAI's o1-mini performance.

  18. #AI2 releases OLMo 2 32B, trained on 6T tokens with #Tulu3.1 post-training. Matches or exceeds GPT3.5 Turbo while using just 1/3 the compute of #Qwen2.5 32B. Complete open recipe includes data, code, weights and training methodology.