Chart says AI skill predicts robot skill, minus Gemini
Robocurve PBC's plot only holds together once the Gemini models come out of it.
Robocurve PBC posted an analysis arguing that general AI capability predicts robot capability, with one group of exceptions. Gemini models, the account says, are strong robotics outliers.
General AI capability predicts robot capability, except for Gemini models, which are strong robotics outliers.
The number doing the work is the correlation coefficient. With every model in the plot it sits at 0.20, which is close to no relationship at all. Take the Gemini models out and it jumps to 0.96, which is close to a straight line.
Two ways to read it
One reading is flattering to Google. Gemini models score better on robots than their general capability would suggest, so they sit above the trend rather than on it. That would make them genuinely unusual, not noise.
The other reading is less kind to the chart. A correlation that only appears after you delete a specific family of models is a correlation that depends on which points you are willing to call outliers. The models being removed are the ones Google has been pushing hardest into robotics, including the Gemini 4 Argon work on full-body humanoid control.
Robocurve PBC did not frame it as a problem. The post presents the exception as the finding, which is a defensible way to do it if you think Gemini really is doing something different. The thread was flagged with a chart and a thread marker, and the headline claim is the one line above.
Who noticed
Keerthana Gopalakrishnan replied to the post with a heart emoji. That is the whole reaction, so read nothing more into it than approval.
What the sources do not show is the underlying data. There is no list here of which models were scored, which robotics benchmark was used, or how many points are in the plot. A jump from 0.20 to 0.96 is dramatic, and how dramatic depends entirely on how many models are on the chart and how many were pulled off it. Nobody has posted that detail yet.
So treat it as a claim worth arguing about rather than a measurement. Either Gemini is the robotics anomaly of the current model crop, or the trend line is being held up by the points that got excluded.
Sound like you were at the demo
One email a day. Under three minutes. The demos, the deployments and the part the press release left out.