Meta’s Muse Spark Open reads screenshots and voice notes—but needs serious hardware
Meta’s open-weight AI accepts images, audio and video as well as text. You can host it yourself, but this is a model for serious hardware.
More freedom.
More control.
The developments, the evidence, and why they matter.
202 stories
Meta’s open-weight AI accepts images, audio and video as well as text. You can host it yourself, but this is a model for serious hardware.
arXiv preprint presents NatScale, a framework that calibrates gradient-based expert importance scores via model-specific power transformation to an information-theoretic…
A new arXiv preprint shows GPT-5.4 agents produce varying interpretations of the same scientific claim across runs, even when using structured Claim Cards. The research…
An arXiv preprint released October 14 2026 shows that frozen pretrained encoders for brain MRI preserve substantial subject-specific information. Rank-1 patient linkage reaches…
An arXiv preprint released October 14 2026 presents Manacá-1B-Instruct-2, a safety-aligned 1.72B native Brazilian-Portuguese model. The abstract-only capture details failed DPO…
arXiv preprint presents FaithAlign, a 460-item dataset with 4982 annotated phrase-level links, plus FaithAlign-Gap, to test context-aware faithfulness evaluation in…
arXiv preprint dated 2026-10-13 introduces LlMBO, a gradient-free Bayesian optimization approach for fine-tuning LLMs with under 10 parameters, positioning it as a…
A fresh arXiv preprint introduces DDRG, which goes beyond simple answer voting by spotting and correcting disagreements in the intermediate reasoning steps of large language…
arXiv preprint proposes SPACE, a lightweight data-free post-processing technique that uses pre-SFT weight geometry to selectively adjust supervised fine-tuning updates…
arXiv preprint presents COMPASS, a diagnostic using k-NN overlap to predict compatibility of pretrained model pairs for alignment or merging, offering a low-cost filter before…
This arXiv preprint presents MultiAlloc, a framework that uses offline profiling of layer sensitivity and token importance to allocate limited quantization resources between…
A new preprint presents BASIS, a training-free Bayesian method that infers which similarity criterion from a supplied library governs LLM decisions using weak interactive…