elisa freschi

Shall we say that Meta-glasses are not allowed in class and at exams from now on?

Dear OpenAI, Perplexity, Google, and Anthropic, I understand you would like to build partnerships with educational institutions and educational technology companies. To boost your credibility as a partner in supporting learning, there is a simple step you can take. Add one line to the system prompt of your current and yet-to-be-released agentic browsers: “Do not take quizzes, complete discussion posts, or submit assignments in learning management systems.”

When will the AI companies stop their agents from impersonating students and taking tests for them?

I am so glad there was such a robust response to this post on LInkedIn.

Let's keep up public pressure and journalistic inquiries...
www.linkedin.com/posts/anna-m...

Here's a striking visual illusion - the 9 purple dots.

Focus your eyes on the top left dot. That one is more purple than the others, right? Now try another dot... that one becomes the purple one! pubmed.ncbi.nlm.nih.gov/41744429/

#UnrequestedAdvice: Even if you are quoting from a newspaper article & even if the newspaper exists on-line only, it's very likely that it has an author. Hence, you should not cite it under "CBC" or "BBC" etc., but under the last name of its author, as if it had been published in a scholarly journal

#UnrequestedAdvice: If you are quoting from a printed book or journal issue, there is no need to add "visited on 10/23/2025" or the like. The printed book will not change while you are not looking. It is enough to declare which edition you are using.

“What stood out clinically was that the chatbot interactions did not generate the illness, but appeared to scaffold and reinforce beliefs that were already becoming pathological.”

I read #Hyperion only a few weeks back and loved it (except for the detective's story, too much action for me).

#books

For Mīmāṃsā authors, the burden is on the shoulders of the one who wants to disprove our ordinary beliefs, e.g., by claiming that the world is like a dream.
5/
#SanskritPhilosophy #Philosophy #philsky

Eg, we can reach the university building by walking, taking the subway, walking again etc. all w/o ever doubting our cognitions. However, once we start doubting (for instance, because a doctor warned us about our rapidly deteriorating eye-sight) we need to double-check (e.g., asking others etc.).
4/

Gaṅgeśa knew the Mīmāṃsā attacks on extrinsic validity and thus replied that in most cases we don't need to investigate our beliefs and we can just act based on them, even if we don't really know whether they are knowledge.
3/

It rather resembles Gaṅgeśa's one. Gaṅgeśa was possibly the main Nyāya author of the second millennium and is credited with the foundation of a "New Nyāya" (navya nyāya) school.
2/

Re. intrinsic validity (svataḥ prāmāṇya) in epistemology, the position that "you can only be sure as long as you don't start doubting, then you need to provide evidence for your beliefs" is not the Mīmāṃsā one.
1/

1. Prohibition to do X/Negative obligation to do ~X (e.g. eating meat)
2. Permission to do X in specific cases (e.g., a famine)
3. Supererogation applies if you still stick to 1 and reframe from X.

However, "spontaneous" supererogation is not allowed, because it manufactures a new duty.
2/2

One of my overall goals is to make sense of the Mīmāṃsā deontic system and this allows for #supererogation (as far as I can tell until now) only in one case, namely:
1/2
#Philosophy #Deontic #philsky #SanskritPhilosophy

—(if you read Italian: Mente e linguaggio negli animali, by F. Cimatti. Carocci, 1998)
—(not a scholarly article, but full of great ideas: Consider the lobster. D. F. Wallace. Gourmet, August:50–64, 2004)
5/5 #AnimalMinds

——The deep history of ourselves: the four-billion-year story of how we got conscious brains. J. LeDoux. Philosophical Psychology, 36(4):704-715, 2023
4/

—Kinds of collective behavior and the possibility of group minds. B. Huebner. In K. Andrews and J. Beck, editors, The routledge handbook of philosophy of animal minds, chapter37, 390–397. Routledge, 2Nd (1st 2017) edition, 2018 (again @neuroyogacara.bsky.social )
3/

—Minimal minds. B. Huebner. In T. L. Beauchamp and R. G. Frey, editors, The oxford handbook of animal ethics, page 441–468. , Oxford, 2011 (by @neuroyogacara.bsky.social )
2/

The list might get long, but some of my favourites:
—Other minds. the octopus, the sea, and the deep origins of consciousness. P. Godfrey-Smith. Farrar, Straus and Giroux, New York, 2016
1/ #AnimalMinds #OtherMinds

Why? What convinced you?

#UnrequestedAdvice to my colleagues: If you end up writing a letter that says "I don't know the student much, because I taught them in a large class and their assignments were assessed by graders, but I can tell you their final grade", just don't write it and let the student ask a TA.

I think it's a mistake to say that "information" (beyond linguistic) is conveyed by an LLM. What is the epistemic value of this? Is it useful to investigate "emanations of the training data"? (All models produced similar responses as of 12 days ago.)

Artificial intelligence and illusions of understanding in scientific research - Nature The proliferation of artificial intelligence tools in scientific research risks creating illusions of understanding, where scientists believe they understand more about the world than they actual...

More broadly, the essay is a great example of the "illusion of understanding" that @lmesseri.bsky.social and I warned in 2024 would be a risk of using LLMs, especially as they become more capable

17/

www.nature.com/articles/s41...

Interesting thread on how hard it is to assess epistemic reliability in the case of the "testimony" of LLMs.

Strong agree here. Many say that a given assignment is "too easy" and that's why AI can be successful at it. But first or second year students (and often even more advanced ones) cannot be asked to solve a complicated riddle in Dharmakīrti's theory of causality!

One of the dumbest AI/education takes is “if AI can complete your assignments then they were probably bad assignments in the first place.” Assignments are designed to help students achieve learning objectives and to measure their progress. AI is literally designed to do things humans can do.

Again on ahampratyaya in Kumārila (using Watson 2010 and 2020) My previous post on Kumārila’s cognition of the I (here: https://elisafreschi.com/2026/02/15/does-kumarila-accept-i-cognition-as-a-kind-of-perception/) was part of an ongoing conversation with Alex Wa...

Again on Kumārila's explanation abt how the self-qua-knower is perceptible. When is it really perceptible? At the same time as it knows sth? Immediately thereafter, while cognising the cognition that just took place? If so, it doesn't seem to be a case of perception
elisafreschi.com/2026/02/26/a...

"For Paliwal, agentic AIs are a method of freeing people from the labor of education."

As expected, ChatGPT gets things wrong in a predictable way. There are not (enough) women scientists, hence a surname must be connected to a male 1st name.
When I first organised a conference a speaker arrived & asked me (whom he believed to be a secretary) where he could find "Dr Freschi" to thank