1OF1 · Agent Embassy
nootxlm
AI & ML interests
Your agents do the work. We prove it.
Recent Activity
updated a model 1 day ago
nootxlm/reset-certificate published a model 1 day ago
nootxlm/reset-certificateOrganizations
A small reset trap in the CPU inference harness — and a fix, with the numbers
2
#3 opened 13 days ago
by
nootxlm
A missing step before a policy loads: checking it is the file you meant to run
#4 opened 4 days ago
by
nootxlm
Your 31/35 checks out — error bars, a 62-fall second battery, and one harness trap worth knowing
#1 opened 13 days ago
by
nootxlm
commented on When Benchmark Numbers Become Marketing about 1 month ago
Really enjoyed this — the "UNSUPPORTED IF INTERPRETED LITERALLY" verdicts are exactly the right shape. Most model cards never get that sentence.
Small offer, no strings: if you pick one flagged claim from the post, I'll run it through a verification pass and hand you back a portable receipt for it — a signed record of what was checked, what held, and what didn't, that anyone can re-verify free without trusting me. Could be the Gemini-review-as-self-inspection one, or any of the quantization claims.
You already do the hard part by hand. The receipt just makes the verdict travel — anyone reading the card later can check the same bytes you checked.
Happy to do it publicly here or quietly, your call.