Thomas RidLast week I was a student for five days, five hours per day—with ChatGPT fully integrated into teaching. Here's what we learned, just in time for Spring Term (which starts tomorrow. Class was Malware Analysis, taught by @jags https://alperovitch.sais.jhu.edu/five-days-in-class-with-chatgpt/
Glyph@ridt @jags @Migueldeicaza this is incredibly disturbing. ChatGPT routinely confidently synthesizes total bullshit, and it’s frankly dangerous to put it in front of students to be used as if it’s an authority on anything. The idea that it will be muddying the waters in educational settings like this is some real torment-nexus level dystopia.
@ridt @jags @Migueldeicaza Since they have carefully scrubbed easy examples like Thomas Running from its dataset, D&D rules are a great example of its total lack of the concept of correctness: https://mastodon.social/@glyph/109713421759380928
Ed Ross@glyph
@ridt @jags @Migueldeicaza
They do say it can be incorrect on the start screen - but that warning should be a lot more prominent seeing as so many people still seem to be surprised when it outputs incorrect statements.
They need to make sure users know what it is: more of a conversation simulator more interested in providing some sort of reply than a correct reply - and what it isn't: a text interference for an infallible encyclopaedia
notsoloudWell, obviously in the article they describe how they make chatGPT fix its code when it gives errors. At the end, when it runs without errors it must be correct and not bullshit, right? Right?
Miguel de Icaza ᯅ🍉@Migueldeicaza @ridt @jags My objection in this case is about risk, not harm. For this audience, in this context, it seems like it was fine. But it's also possible to make lots of great furniture with a bandsaw with no sawstop and no safety goggles before you lose an eye or a finger. If this is introduced to education more broadly without VERY strict safety protocols, there will be intellectual carnage