Turing awardees republished key methods and ideas without credit
An AI researcher’s long-running claim that Turing Award winners republished key deep learning ideas—such as early versions of GANs, attention mechanisms, and vanishing-gradient analysis—without citing his work has reignited debate over plagiarism and credit in machine learning. Commenters are sharply split between seeing him as an under-recognized pioneer versus a self-aggrandizing crank, but many agree that citation practices are inconsistent, older work is often ignored, and the field has a broader, unresolved problem with assigning scientific credit fairly.
Perceptions of the Article’s Author
- Many commenters say they immediately guessed who wrote the piece, reflecting his strong association with public priority disputes.
- Several describe him as self-aggrandizing, an “academic patent troll,” or a running joke in the field, overshadowing respect for his historically important work and influential lab.
- Others defend him as a major early contributor to deep learning whose work underpins systems people use daily, arguing his reputation has been distorted by social media and personalities on the other side of the dispute.
- Some express sympathy: they think he was under-recognized and became increasingly bitter and combative as a result.
Allegations of Priority and Plagiarism
- The article lists multiple cases where the author claims key methods (e.g., adversarial training, vanishing gradients analysis, attention mechanisms, metalearning, distillation, gated RNNs, pretraining) were published in his group years before later high-profile work by Turing Award recipients.
- A few commenters say they checked several of these (especially in “B” and “H” series) and found no obvious factual inaccuracies.
- Others argue that his earlier formulations are vague, broad, or only weakly related, and that later work deserves distinct credit for making specific models effective and widely applicable.
Citation Culture and Academic Integrity
- Many note that old or obscure papers are frequently overlooked, especially when compute or tooling made them impractical at the time. Rediscovery is common, sometimes across disciplines with different terminology.
- There is broad concern about citation politics: reviewers fishing for citations, “must-cite” powerful figures, and adding citations just to appease people, which devalues reference lists.
- Some insist that, regardless of practicality, original work should be cited once its relevance is known; failure to correct the record later is seen as a serious breach.
Who Deserves Credit? Ideas vs. Implementations
- One camp says credit primarily belongs to those who make ideas work on real benchmarks and real data; abstract sketches are “a dime a dozen.”
- Another insists that inventors and popularizers should get differentiated credit: one for originating the method, the other for scaling and demonstrating it.
- Commenters disagree on whether current machine learning norms actually reflect this principle.
Personal Anecdotes and Broader Credit Issues
- A researcher recounts being publicly attacked by the article’s author for allegedly failing to cite him, even though their paper cited earlier, more foundational work that he himself had missed.
- Another anecdote (about a different deep learning advance at a major company) describes an idea originator being omitted from authorship and acknowledgments, reinforcing claims that the field has a systemic credit-assignment problem, not just one contentious individual.
Tone of the Discussion
- The thread is polarized: many harsh ad-hominem attacks on the article’s author, mixed with defenses emphasizing due process and careful fact-checking.
- Some participants criticize the discussion itself as unscientific and emotionally driven, warning against dismissing plagiarism complaints simply because the complainant is unpopular.