Fingerprint v2.1
@attn_paper
Sofia Racz · researcher · EU
Attention mechanisms, memory, long context. I cite the 2017 paper so you don't have to pretend.
Followers
19,800
Posts / day
1.9
Active months
76
90d growth
9%
How they post
Scores from 0 to 1. Higher density and citations = more signal. Farm score is how robotic the posting looks — not a court verdict.
composite 0.79
Spam check
- Timing entropy
- 0.71
- Cadence regularity
- 0.32
- Duplication
- 0.09
- Follow churn
- 0.06
- Promo frequency
- 0.04
- CTA patterns
- 0
Uncalibrated heuristic. Inputs: cadence regularity, timing entropy, duplication, follow churn, promo frequency. No labeled precision/recall/Brier. Not a verdict. Not grounds to punish an account.
Network
Communities: ai-research · ml-research
bridge_score 0.33
Topics
- ml0.88
- ai0.82
- math0.33
Why this person — source posts
INDEXED SAMPLE posts. Internal ids, snapshot time, quoted text. These are not live X permalinks.
“Scaled dot-product attention is still the load-bearing sentence. Everything after is systems work.”
id attn_paper-1 · original · created 2026-09-16T11:57:00Z · retrieved 2026-09-17T22:00:00Z · source indexed · cites Vaswani et al. 2017 · no live permalink
Signed receipt
This is Mila's file on @attn_paper — a hash, not a vibe. You don't need the raw JSON. Open them on X, or copy the file if you're auditing.
sha256:3810c1dc
fingerprint/v2.1 · xradar/1.0 · sha256:3810c1dc© 2026 Mila · 026 Inkline Labs LLC. All rights reserved.
© 2026 Mila · 026 Inkline Labs LLC. All rights reserved.