Loading activity…
@azaliamirh
Founder @RicursiveAI, Asst. Prof. of CS at Stanford. Prev: DeepMind, Anthropic, Brain. Co-Creator of MoEs, AlphaChip, Test Time Scaling.
AI
Its creators say model and accelerator choices can yield several times more value per joule on the same task.
AI
It reportedly tops U.S. open-weight models on TerminalBench 2.1 and runs locally on a MacBook Pro or DGX Spark.
AI
Samaya says Sterling scored 58.9% versus Claude Fable 5.1’s 55.9%, at $0.61 versus $6.89 per question.
AI
Its creators say an LLM writes optimized Nvidia PTX directly, bypassing intermediate compiler layers.
AI
Zhan says Sequoia, Jeff Dean, Sachin Katti and Philip Wong will be among the fund’s limited partners.
AI
In the announcement, @bneyshabur places Mirendil-Base in the 100B–500B-parameter range and promises independent evaluations.
AI
TechCrunch reports its founders are developing AI to learn from each chip-design iteration and improve later designs.
AI
Its creators also report 10–21× faster unconditional generation on three datasets, with comparable or better generative perplexity.
XAI
A post introducing the method reports that it matches the performance of synchronous updates on long-horizon agent tasks.
AI
A post describes a model that translates source code such as Triton directly into low-level assembly such as PTX, with a verifier checking functional correctness, race conditions and deadlocks.
AI
The Turbo-dLLM team reports DFlash2 speculative-decoder training speedups of 2.48x at 512K context length and 7.59x at 1M context length on eight H100 GPUs.
AI
Discussing a Stanford University and Together AI paper, the user reports an 18-fold rise in accuracy per joule over 16 months: 5.9-fold from better accelerators versus 3-fold from better models.
Technology
Azalia Mirhoseini and Anna Goldie receive TIME100 AI recognition after founding Ricursive.
AI
Sunny Madra and Elon Musk comment on AI intelligence per joule efficiency.
AI
She recalls their time at Google Brain and looks ahead to his new chapter.
