LlmWikis.org
A non-normative reference layer for large multimodal models, AI interoperability, benchmarks, implementation patterns, and source-backed explainers.
Launch boundary
LlmWikis explains and catalogs. UAIX.org remains canonical for UAI-1 specifications, schemas, registries, validators, roadmap, governance, and Project Handoff guidance.
Turn a blank domain into a useful, measurable knowledge surface.
The first site pass follows the strategy reports: restore a working public surface, define the scope, seed the taxonomy, create UAIX bridge pages, and keep contribution quality bounded until policy and review loops exist.
Complementary first
Launch around UAIX-adjacent explainers and AI interoperability topics before deciding whether to expand into a broad standalone sector wiki.
Structured taxonomy
Use repeatable page types for models, benchmarks, protocols, implementations, comparisons, sources, timelines, and glossary records.
Human review
AI-assisted drafts are allowed only as drafts. Public pages need human review, citations, and a clear evidence status.
The first wiki map is built around high-value technical work.
LMM Model Registry
Profiles for architectures, modalities, training signals, strengths, limits, hardware needs, and source trails.
Evaluation Hub
A blueprint for benchmark pages, score provenance, contamination notes, update cadence, and comparison paths.
UAIX Implementation Guides
Non-normative tutorials and implementation notes that point back to UAIX for canonical UAI-1 records.
Prompting and workflows
Patterns for multimodal prompts, agent handoffs, UI understanding, documents, audio, video, and spatial tasks.
Hardware and deployment
Edge, workstation, cloud, acceleration, latency, memory, and operational tradeoffs for multimodal systems.
Governance and sourcing
Contributor intake, source rules, moderation, legal readiness, and transparent support boundaries.
A wiki launch needs trust signals before scale.
Not live yet
Live benchmark dashboards, automated arXiv ingestion, public MCP access, open editing, memberships, grants, and multilingual support are planned candidates. They should stay described as planned until implemented and reviewed.