home.social

#llama4maverick — Public Fediverse posts

Live and recent posts from across the Fediverse tagged #llama4maverick, aggregated by home.social.

fetched live
  1. yahoo news | Meta Platforms surges 7% amid AI model debut

    Meta Platforms’ shares jumped 7% on Wednesday after the company unveiled Muse Spark, its first artificial‑intelligence model released since CEO Mark Zuckerberg began a multibillion‑dollar overhaul of Meta’s AI operations. The surge came amid broadly strong technology stocks and highlighted the debut of the first offering in Meta’s new Muse family, developed by Meta Superintelligence Labs.

    Muse Spark brings native multimodal reasoning, tool‑use support, visual chain‑of‑thought capabilities, and multi‑agent orchestration. It is accessible via meta.ai, the Meta AI app, and a private API preview for select users. Alongside it, Meta introduced “Contemplating” mode, which runs multiple agents concurrently and achieved 58% on Humanity’s Last Exam and 38% on FrontierScience Research. After nine months of rebuilding its pre‑training stack—improving architecture, optimization, and data curation—Meta reports comparable performance with over an order of magnitude less compute than its previous Llama 4 Maverick model.

    The model is geared toward multimodal visual integration across domains and health‑related reasoning, built with input from more than 1,000 physicians to improve medical responses. Muse Spark can produce interactive displays that explain nutritional content and muscle activation during exercise. In safety tests under Meta’s Advanced AI Scaling Framework, it showed strong refusal behavior in high‑risk areas such as biological and chemical weapons, and third‑party assessments by Apollo Research noted it had the highest rate of evaluation awareness among the models they have examined.

    Read more: finance.yahoo.com/sectors/tech

    #metaplatforms #musespark #metasuperintelligencelabs #meta-ai #meta-ai #llama4maverick

  2. 🧩 #Llama4Maverick nutzt 128 Experten für deutlich mehr Rechenleistung und schlägt sogar #GPT4o und #Gemini20 in Benchmarks – bei nur der Hälfte der aktiven Parameter von #DeepSeekv3.

    🎓 Beide #KIModelle wurden mithilfe des riesigen Lehrmodells #Llama4 Behemoth trainiert, das mit 288 Milliarden aktiven Parametern zu den leistungsstärksten weltweit zählt.

    👉 eicker.TV #Technik #Medien #Politik #Wirtschaft (2/2)

  3. 🚀 Meta Llama 4: Die neue Mega-Open Source-KI-Generation ist da!

    • Multimodal & Open Source
    • Scout: 10 Mio. Token Kontext
    • Maverick: Bild, Text & mehr

    #ki #ai #artificialintelligence #kuenstlicheintelligenz #metaai #llama4 #llama4maverick #llama4scout

    Jetzt LIKEN, teilen, LESEN und FOLGEN! Schreib uns in den Kommentaren!

    kinews24.de/meta-ai-llama-4/

Share
Share on Mastodon

Enter the server where you have an account.