Viable Media, LLC

company

AI & ML interests

LLM evals, red teaming, jailbreaks, RAG ergonomics, small-model baselines, content provenance/C2PA, deepfake detection, consent & identity wallets, hallucination audits, adversarial inputs, latency & UX, explainability (pragmatic), slop detection, safety vs usability

Organization Card

Community About org cards

VibeAxis

Receipts, not vibes. We break AI toys in public—tiny tools, stress-tests, and proof that “looks real” isn’t evidence.

Writer-engineers shipping small, fast demos; red-teaming LLMs; and building provenance-first UX so performance theater has nowhere to hide.

What we ship

Spaces (demos): quick, client-side prototypes you can poke. If it melts, good—now we know where.
Playbooks: reproducible tests, failure cases, and “prove-it” pipelines.
Receipts: posts, write-ups, and teardown notes that link back to the working demo.

Live Spaces (pinned)

Slop Index Lite — on-page slop detector (clichés, hedges, repetition, sentence bloat).
Client-side only. No upload. No storage.
⮕ Demo: https://huggingface.co/spaces/Vibeaxis/Slop-Index-Lite · Write-up: https://vibeaxis.com/the-ai-slop-index-grade-any-page-in-60-seconds/
Fake-Fast Detector Lite — catch “performance theater” (TTFB, cache flags, script bloat) in the wild.
⮕ Demo: https://huggingface.co/spaces/vibeaxis/fake-fast-lite · Notes: https://vibeaxis.com/fake-fast-detector/

Want to see something specific here? Open a discussion with a short repro or a sketch. Weird edge cases welcome.

How we work (the receipts doctrine)

Prove-it is the default. If media or claims can’t travel with provenance (device signing / content credentials / chain-of-custody), they travel with a warning label.
Small > big. Fast demos beat half-built platforms. Ship, measure, iterate.
Humans first. UX over leaderboard cosplay. If it’s not useful under thumb, it’s theater.
Document failure. Bad outputs are feature requests in disguise. We publish them.

Contribute / collaborate

Issues & ideas: use Discussions to pitch a micro-demo, submit a failure mode, or request a test.
PRs: keep them surgical—one feature/fix per PR, with a 60-second demo video or GIF if possible.
Attribution: we keep model/data/tool credits visible; you should too.

No spam, no “DM for promo,” no stealth tracking. This is a lab, not a funnel.

Reuse & license

Code in Spaces: MIT unless noted in the Space.
Content (write-ups, docs): CC BY-SA 4.0 unless otherwise specified.
Third-party assets: credited and licensed upstream; check each Space’s footer.