Подтвердите e-mail

Для публикаций, комментариев, реакций и сообщений подтвердите адрес.

←

Публикация

Aaron Tay

Last thread i promise. 1/ Powerful AI makes the inside view extraordinarily seductive. Ask an LLM for the strongest case for one side Ask it to expose methodological weaknesses. Ask it to rebut every objection. Soon you may feel that you have mastered the debate.

Aaron Tay

18/ Information literacy should teach students to question authority without pretending that all authorities are interchangeable. “Authority is constructed” must not become “expertise is optional” - end for now

Обсуждение

1 прямых ответов · 10 сообщений

Aaron Tay

2/ But the model may be far better at generating arguments than the student is at evaluating them. It can make a fringe position sound sophisticated. It can make a genuine consensus look fragile.

Ответ для Aaron Tay
Aaron Tay

3/ It can also do the reverse. It may flatten a real scientific dispute into a bland statement that “experts generally agree”. LLMs can manufacture both false balance and false certainty.

Ответ для Aaron Tay
Aaron Tay

4/ The danger is not only hallucination. An LLM can give a novice the experience of having investigated a debate deeply without giving them the expertise needed to recognise which arguments are misleading.

Ответ для Aaron Tay
Aaron Tay

5/ The machine can generate objections faster than the learner can acquire the knowledge needed to assess them. This is not merely an accuracy problem. It is a rhetorical asymmetry problem.

Ответ для Aaron Tay
Aaron Tay

6/ Call the result epistemic learned overconfidence. The explanations are clear. The objections are answered. The user feels increasingly competent. Responsiveness is mistaken for reliability. Persuasion is mistaken for understanding.

Ответ для Aaron Tay
Aaron Tay

7/ “Consider both sides” becomes risky when an LLM can produce two polished and symmetrical performances even when the underlying evidence is profoundly asymmetrical. Balance in the answer does not prove balance in the field.

Ответ для Aaron Tay
Aaron Tay

8/ The only person I have seen study anything close to this is Mike Caulfield, testing iterative LLM responses to prompts such as “What is the evidence for and against X?” But I am not sure his tests cover the kinds of contested claims I have in mind. Perhaps the method will still work? END

Ответ для Aaron Tay
Aaron Tay

Finally, a warning: although I am an academic librarian, I am not an information-literacy specialist beyond what a typical academic librarian encounters. My own speciality lies in understanding academic search.

Ответ для Aaron Tay
שלמה לוי

Anecdotally, one of the most consistent issues with LLM output (from a technical writing perspective) has been “fluff.” They tend to pad their output. This can include repeating a point with slightly different language, to achieve parallel construction.

Ответ для שלמה לוי
שלמה לוי

As you noted, this creates a situation in which a weak point has the illusion of greater strength. And it’s why I keep having to tell writers to delete aesthetically pleasing tables of useless information.