A fluid gallery for ML experiments

Virginia lawmakers propose an AI-in-schools pilot (SB 394) and a rule that schools can’t require or encourage chatbot use for coursework (HB 1186).

AI-news grab bag: world models (Solaris), video reasoning, DreamID-style identity control, LavaSR upscaling, and a Doc-to-LoRA workflow idea.

A quick survey of how states are structuring AI task forces: membership, scope, duration, deliverables, and limits—plus a Kansas HB 2592 case study.

A Turkish encoder-decoder model that operates at the character level (Charformer-style), skipping tokenizers. Potentially useful for noisy text and spelling variations, but still marked as in-development.

A 229M-parameter sparse MoE model trained on Witcher lore and synthetic instruction data. Not compatible with vanilla Transformers; it runs via the author’s `cirilla` Python package.

An early DualTowerVLM checkpoint: separate vision and text towers, fused later for multimodal generation. Interesting if you’re experimenting with VLM architecture and representation fusion.

Cosmos Dataset Search is a semantic search platform for large video datasets, supporting text-to-video and video-to-video queries by generating embeddings with Cosmos-embed and running fast similarity search in Milvus (cuVS-accelerated).

An NVIDIA blueprint for building a video analytics agent that can summarize and answer questions over large video libraries by ingesting frames + audio, generating dense captions, and indexing the results for context-aware RAG.

A BioNeMo blueprint for designing peptide/protein binders from a target sequence using models like RFDiffusion, ProteinMPNN, and AlphaFold2 (including multimer), producing candidate binders and predicted complex structures in PDB format.