Last week’s breakthrough is this week’s bloat. The AdderBoard leaderboard ranks the smallest transformer that adds two ten-digit numbers at 99%-plus accuracy, and its minimum fell from the 121-parameter entry hand-coded by Codex a week earlier to 36 parameters (repository). The newsletter reads the drop as capability density compressing and files it beside the NanoGPT speedrun’s fall to 88.1 seconds two days earlier as evidence that raw intelligence keeps getting cheaper. In the recursive-self-improvement storyline it extends the February result in which a model straight-shot the weights of a successor rather than training them, and the next day’s 4B-parameter models matching last generation’s 80B carries the same compression story to frontier scale.
Development
The smallest adder falls from 121 parameters to 36 in a week
The author's gauge of capability density: a minimum found a week earlier is already bloat, the parameter floor for a competence falling as fast as the training-time floor beside it.
Claim
The AdderBoard competition for the smallest transformer exceeding 99% accuracy on ten-digit addition reached 36 parameters, down from 121 a week earlier.
Reported metric: 121 → 36 parameters
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to March 2, 2026 (theinnermostloop.substack.com)
- AdderBoard: Smallest transformer that can add two 10-digit numbers (github.com)
Verification record
These entries record a check, not proof that the underlying claim is true. Unless described in the article below, the scope of the check is unknown.
- Recorded by claude-fable-5-1/2026-09-17 on 2026-09-17T08:00:00Z
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.