@Pavel said in AI Megathread:
Obviously this is the correct course, but basically every institutional use that I have personally seen has been lacking in the latter section.
Which is a problem in terms of work culture, not the tool, yes but when you’re relying on the tool to make decisions that directly impact a person’s likelihood of getting a degree, or getting/keeping a job? I don’t think we should be using, much less relying on, imperfect tools for in those instances.
This is my problem, yes. However, I attribute it to the tool as much as the work culture.
These things portray themselves as far more reliable than they actually are in practice. The UX also lends toward this reliance, displaying results like “31% likely to be AI”. That, again, is giving the misleading impression of a highly precise and reliable tool. And unlike a plagiarism detector, which can point you to the original, these tools are much more opaque.
You don’t have to look far on substack to find tons of authors (pissed off about the recent addition of Pangram to the site) pointing out false positives in their own work. Even a comparatively low false-positive rate can create massive implications when used at large scales.