Malik asks if Astra is plagiarizing the papers it learned from
A junior researcher's blog post on credit assignment pulled in Jitendra Malik, Chris Paxton and Chen Tessler within hours.
Jitendra Malik says AI models like Astra should arguably be accused of plagiarism, because they do not cite the research they are built on. He says he co-authored a couple of in-hand rotation papers that might have been used in the Astra pen-spinning demo.
If certain kinds of creative work stops because incentives are disrupted, it is like farmers eating their seed corn.
This is not a claim that the results are fake. Malik calls the progress amazing, and says it is built on the work of generations of scientists. His point is about the machinery around the result. Patent disclosures and paper citations were how society handled credit assignment, and a model that synthesizes prior work into a new demo skips all of it.
Where it started
The thread began with Toru (@toruo_o), who wrote a blog post as a junior researcher after reading recent thoughts from Malik, Michael Black, Ken Goldberg and Phillip Isola. Toru calls the recent Astra dexterous manipulation demos truly impressive, then asks two questions academia has no answer for. How should credit be assigned when an agent synthesizes many prior works into a new research result. And how can academia attract and keep talent when so much of its incentive structure runs on credit.
Malik also pointed at an open letter by 25 Fields Medalists responding to the Navier Stokes result, which complained about disruption to the mathematics research process. He says some people read that as turf protection, and that the point was more subtle.
The pushback
Chen Tessler (@chentessler) says this is not an LLM problem at all. Researchers have often ignored prior work knowingly, he says, and he has had people with early access to his code and papers build on them without proper credit. There is no way to report it and no penalty. The incentive structure is bi-directional, he argues. You want credit assigned to you, and your work looks less impressive when you assign too much of it to others.
Chris Paxton (@chris_j_paxton) came at it from a different angle, saying it feels harder and harder to do meaningful work without the resources of a massive, well-funded institution.
This lands in the same week Michael Black said Astra made his ECCV paper obsolete before he presented it. The argument is no longer about whether the dexterity demos hold up. It is about who gets named when they do.
There are no checks-and-balances, no way to "report" such a work, no penalty for doing so.