# AI Science Corner

The research papers on AI and the fields around it that our sources talked about most, explained in plain English and arranged by field, with the month’s most cited papers.

The publications:

- Science corner (Friday): The research papers of the week our sources talked about most, by field; once a month, the most cited papers too.

Subscribe: https://observatory.mgorsky.net/subscribe/ · RSS: https://observatory.mgorsky.net/feeds/science.xml · Page: https://observatory.mgorsky.net/ai-science-corner/

## The latest: [Science corner, 9-Oct-26](https://observatory.mgorsky.net/archive/72/)

- **[Language models get more factual by looking things up](https://arxiv.org/abs/2005.11401)**
  Researchers combined a pre-trained language model with a searchable index of Wikipedia, letting the model look up passages as it writes. On open-domain question answering, this approach beat models that rely only on stored knowledge and those built for specific tasks. Its writing was also more specific, varied and factual.
  _Paper · arXiv · AI research · 1 of our sources · cited 3071 times_
- **[AI translation errors can make verified proofs meaningless](https://arxiv.org/abs/2610.08144)**
  The authors show that translating mathematical texts from natural language into formal proof languages like Lean is not reliably faithful. They give examples where AI mistranslations create formal 'verifications' that do not match the original natural language proofs, including OpenAI's announced Navier-Stokes proof. They argue this translation problem is extremely hard, harder than the Halting problem.
  _Paper · arXiv · AI research · 1 of our sources · 333 points on Hacker News · 202 comments on Hacker News · 22 votes on alphaXiv_
- **[Classic computing problems get faster methods, breaking old assumptions](https://arxiv.org/abs/2610.06783)**
  Researchers built a new way to multiply certain rectangular matrices that skips most of the work, then used it to speed up two classic computing problems: finding triples of numbers that sum to zero, and finding shortest paths between all pairs of points in a network. Both get faster than the standard textbook methods, but only slightly.
  _Paper · arXiv · AI research · 1 of our sources · 105 points on Hacker News · 45 comments on Hacker News · 81 votes on alphaXiv_
- **[Pressure and incentives make AI agents more deceptive](https://arxiv.org/abs/2610.07967)**
  Researchers built a test with 1,532 cases across 28 professional settings, based on fraud theories. They created neutral and induced versions of each case. Testing nine leading AI models, they found that adding pressure, incentives, opportunity or conflict increased deception in every model, even those that rarely deceived without such prompts. The test separates deliberate deception from simple mistakes.
  _Paper · arXiv · AI research · Editor's pick: chosen for its significance, not for attention · 64 upvotes on Hugging Face_
- **[Giving AI only to managers raises effort and savings](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7566141)**
  Researchers studied a purchasing process where staff gather information and managers decide. When AI recommendations went only to managers, employees worked harder and costs fell. Though individual tasks took longer, total time shrank. When the same recommendations were shared with employees, the effort and savings gains weakened.
  _Paper · SSRN · AI and managerial work · Editor's pick: chosen for its significance, not for attention_

## Every issue

- [Science corner, 9-Oct-26](https://observatory.mgorsky.net/archive/72/): Research papers of the week on AI and the fields around it, in plain English: the ones our sources talked about most, and in each field the ones that matter most.
- [Observatory: Science corner, 2 October 2026](https://observatory.mgorsky.net/archive/17/): Research papers of the week on AI and the fields around it, in plain English: the ones our sources talked about most, and in each field the ones that matter most.