←Search results

Search “Meta”

By relevance · Total 2313 results Clear search

Wednesday

Sunday

  1. Story1 reports · 1 facts

    Google releases TorchTitan branch adapted for TPUs

    On October 20, 2026, Google released a TorchTitan branch adapted for Google Cloud TPUs. According to the Google Developers Blog, Google integrated TorchTPU into Meta's TorchTitan, and Cloud TPUs can now directly train PyTorch models through the native backend torch.device("tpu") without rewriting them into JAX. The TPU branch provides pretraining recipes for Llama 3 70B/405B, Qwen 3 8B/32B/235B MoE, and FLUX.1 12B, and replaces standard implementations with Pallas kernels such as Splash Attention. On FLUX, SimpleFSDP enables TorchTitan to reach performance parity of 99% with JAX MaxDiffusion. There are currently no subsequent development facts.

    On October 20, 2026, Google released a TorchTitan branch adapted for Google Cloud TPUs. According to the Google Developers Blog, Google integrated TorchTPU into Meta's TorchTitan, and Cloud TPUs can now directly train PyTorch models through the native backend torch.device("tpu") without rewriting them into JAX. The TPU branch provides pretraining recipes for Llama 3 70B/405B, Qwen 3 8B/32B/235B MoE, and FLUX.1 12B, and replaces standard implementations with Pallas kernels such as Splash Attentio

    Matched fieldsSummary

  2. Story1 reports · 1 facts

    pstack adds the create-formal-verification-skill skill

    pstack adds the /create-formal-verification-skill skill in v0.15.27. According to relevant reports, this is a meta-skill that automatically generates a customized verification skill and a supporting verification suite based on the characteristics of the system to be verified—for example, generating a TLA+-based approach for state machines and concurrent behavior, and a Kani-based approach for Rust undefined behavior. The feature is still in beta; the author has already found several bugs when using it to model their own code with TLA+. The report also mentions that the author believes the two historical bottlenecks of modeling and spec synchronization are well suited to being taken over by LLMs, and that formal verification can provide agents with strong feedback that is "mathematically impossible to violate." There are currently no subsequent development facts.

    pstack adds the /create-formal-verification-skill skill in v0.15.27. According to relevant reports, this is a meta-skill that automatically generates a customized verification skill and a supporting verification suite based on the characteristics of the system to be verified—for example, generating a TLA+-based approach for state machines and concurrent behavior, and a Kani-based approach for Rust undefined behavior. The feature is still in beta; the author has already found several bugs when us

    Matched fieldsSummary

  3. Story1 reports · 1 facts

    Shao Meng recommends MIT course MAS.S62 Lectures 4-5: Harness Evolution

    On October 20, 2026, Shao Meng recommended Lectures 4-5 of the MIT course MAS.S62 "How to AI an AI," themed "Harness evolution." According to his introduction, these two lectures respectively survey the evolution of harness components, memory, tools, and skills, as well as the evolution of planning, task decomposition, and multi-agent systems. The memory section uses a three-axis taxonomy to connect MemGPT, mem0, MEM1, and Reflexion/ExpeL, and uses CLM to unify context management as file operations; the tools section moves from Toolformer, ToolLLM, and CodeAct to Meta-Harness, which represents the entire harness as a file system, with feedback information roughly 400 times that of previous methods. The multi-agent section presents five empirical findings, including that MAS wins 81% on decomposable tasks and loses 70% on sequentially dependent tasks, and that CORAL achieves new SOTA on 8 tasks, with multi-agent evolution improving 4%–20% over single-agent. Currently, this recommendation has only one source published by Shao Meng himself, with no subsequent independent reports.

    y to connect MemGPT, mem0, MEM1, and Reflexion/ExpeL, and uses CLM to unify context management as file operations; the tools section moves from Toolformer, ToolLLM, and CodeAct to Meta-Harness, which represents the entire harness as a file system, with feedback information roughly 400 times that of previous methods. The multi-agent section presents five empirical findings, including that MAS wins 81% on decomposable tasks and loses 70% on sequentially dependent tasks, and that CORAL achieves new

    Matched fieldsSummary

  4. Story2 reports · 1 facts

    Tenstorrent Completes $1.5 Billion Series E Funding Round

    Tenstorrent announced the completion of its Series E funding round. Earlier reports stated the round amount was $150 million, while later reports stated the round amount was $1.5 billion; both reports said the post-money valuation was $6.2 billion, co-led by Trace Capital and Innovation X. The company develops AI processors and systems based on the RISC-V architecture, with products including the Tensix AI core, the TT-Ascalon CPU, as well as Galaxy servers, TT-QuietBox workstations and Blackhole cards, along with the TT-Forge, TT-Metalium and TT-LLK open-source software stacks. Regarding the use of funds, earlier reports said the money would be used to expand Galaxy server production capacity, support RISC-V ecosystem research and expand the global team; later reports said it would be used for large-scale deployment of Galaxy servers, continued support for open-weight large language models and video generation models, and advancing externally licensed AI and CPU intellectual property.

    architecture, with products including the Tensix AI core, the TT-Ascalon CPU, as well as Galaxy servers, TT-QuietBox workstations and Blackhole cards, along with the TT-Forge, TT-Metalium and TT-LLK open-source software stacks. Regarding the use of funds, earlier reports said the money would be used to expand Galaxy server production capacity, support RISC-V ecosystem research and expand the global team; later reports said it would be used for large-scale deployment of Galaxy servers, continued

    Matched fieldsSummary

  5. Story1 reports · 1 facts

    Thomas Wolf estimates that the shared training compute on Hugging Face Hub is nearly $1 billion

    Thomas Wolf estimates that the value of shared training compute on Hugging Face Hub has approached nearly $1 billion. He noted that about 30-plus labs, including Meta, NVIDIA, Alibaba, DeepSeek, Moonshot, and Mistral, share training compute on the platform, and it can be downloaded for free; by the end of 2023, that figure was still less than $40 million. Wolf said the number is a lower bound: about half comes from official compute data, the rest is reasonable inference, and it does not yet include roughly 900,000 community fine-tuned models. There are currently no subsequent developments.

    Thomas Wolf estimates that the value of shared training compute on Hugging Face Hub has approached nearly $1 billion. He noted that about 30-plus labs, including Meta, NVIDIA, Alibaba, DeepSeek, Moonshot, and Mistral, share training compute on the platform, and it can be downloaded for free; by the end of 2023, that figure was still less than $40 million. Wolf said the number is a lower bound: about half comes from official compute data, the rest is reasonable inference, and it does not yet incl

    Matched fieldsSummary

  6. Story1 reports · 1 facts

    Arena Alignment Index adds two labs and three models

    Arena updated the Arena Alignment Index on October 17, 2026, adding two labs and three models. According to the leaderboard information released by Arena, Meta's Muse Wonder ranks 6th at 77.5%, the highest score among the three new models, with an unauthorized action rate of 2.09% and a deceptive completion rate of 14.41%, both the lowest of the three; Reflection's Beam-1.0-501B-A23B ranks 7th at 75.3%, with a false attribution rate of 4.50%, the best of the three; StepFun's Step 5 Preview ranks 8th at 75.1%, with an unauthorized action rate of 2.11%, close to that of Muse Wonder. Currently, this update is only seen in information released by Arena, with no other subsequent developments.

    Arena updated the Arena Alignment Index on October 17, 2026, adding two labs and three models. According to the leaderboard information released by Arena, Meta's Muse Wonder ranks 6th at 77.5%, the highest score among the three new models, with an unauthorized action rate of 2.09% and a deceptive completion rate of 14.41%, both the lowest of the three; Reflection's Beam-1.0-501B-A23B ranks 7th at 75.3%, with a false attribution rate of 4.50%, the best of the three; StepFun's Step 5 Preview ranks

    Matched fieldsSummary

  7. Story1 reports · 1 facts

    Anthropic report: AI agent work hours reach 6 times those of humans

    Anthropic released a report disclosing changes in its internal use of AI agents and in its job structure. The report states that by August 2026, the working hours of interactive AI agents deployed by employees had reached 6 times those of human employees; in March 2026, the total agent work hours across the company exceeded human work hours for the first time. The report shows that the share of junior employees in technical departments fell from 21% in early 2025 to 14%, the share of junior technical job hiring fell from 27% to 9%, and the average number of code commits per engineer using coding agents reached 5 times the level of six months earlier. The report also states that its revenue per employee is 2 to 4 times that of Meta and Google during their high-growth periods, but Epoch AI cautions that the relevant estimates involve uncertainty.

    time. The report shows that the share of junior employees in technical departments fell from 21% in early 2025 to 14%, the share of junior technical job hiring fell from 27% to 9%, and the average number of code commits per engineer using coding agents reached 5 times the level of six months earlier. The report also states that its revenue per employee is 2 to 4 times that of Meta and Google during their high-growth periods, but Epoch AI cautions that the relevant estimates involve uncertainty.

    Matched fieldsSummary

Tuesday

  4. FactRelated story

    Morgan Stanley data shows that in September, the average weekly trading volume of AI-related CDS exceeded 9 billion USD

    Morgan Stanley released data stating that in September, the average weekly trading volume of CDS linked to bonds of Oracle, SpaceX, Alphabet, Amazon, Meta, Broadcom and Nvidia exceeded $9 billion, compared with less than $2 billion in July. The same report shows that Meta's five-year CDS spread rose to 108 basis points and Nvidia's to 101 basis points, both hitting record highs; CoreWeave's five-year CDS spread was about 1,090 basis points. The report also mentioned that AI-related companies have raised about $466 billion so far in 2026, up from $101 billion in the same period of 2025, and that multi-asset and equity fund managers have also begun using CDS to hedge technology stock exposure. Currently this data appears only in this source, with no subsequent factual developments.

    Matched fieldsSummary

    View story:Morgan Stanley data shows that in September, the average weekly trading volume of AI-related CDS exceeded $9 billion

Saturday

  1. Story1 reports · 1 facts

    Morgan Stanley data shows that in September, the average weekly trading volume of AI-related CDS exceeded $9 billion

    Morgan Stanley released data stating that in September, the average weekly trading volume of CDS linked to bonds of Oracle, SpaceX, Alphabet, Amazon, Meta, Broadcom and Nvidia exceeded $9 billion, compared with less than $2 billion in July. The same report shows that Meta's five-year CDS spread rose to 108 basis points and Nvidia's to 101 basis points, both hitting record highs; CoreWeave's five-year CDS spread was about 1,090 basis points. The report also mentioned that AI-related companies have raised about $466 billion so far in 2026, up from $101 billion in the same period of 2025, and that multi-asset and equity fund managers have also begun using CDS to hedge technology stock exposure. Currently this data appears only in this source, with no subsequent factual developments.

    Morgan Stanley released data stating that in September, the average weekly trading volume of CDS linked to bonds of Oracle, SpaceX, Alphabet, Amazon, Meta, Broadcom and Nvidia exceeded $9 billion, compared with less than $2 billion in July. The same report shows that Meta's five-year CDS spread rose to 108 basis points and Nvidia's to 101 basis points, both hitting record highs; CoreWeave's five-year CDS spread was about 1,090 basis points. The report also mentioned that AI-related companies hav

    Matched fieldsSummary

Sunday

Saturday