I am begging AI boosters to explain to me how it is in any way time-saving or helpful to have an "assistant" who may be lying to you and whose work you therefore have to check every step along the way.
Reposted from
Nathan K. Hensley
Dude this is so fucking bleak; I know I’m naive but I still genuinely cannot believe we have people in the humanities normalizing this stuff
Comments
Beware of cheap imitations, folks.
As an RA, would you ever just... make up some plausible sounding stuff and put it into your work, you know, for a lark?
I feel the answer is "no, of course not" but I don't like to make assumptions. 😀
It's absurd to, instead of using technology to help us discover sources to use directly, to add an extra layer of uncertainty between the search prompt and the document!
But yeah, the chatGPTs of this world really are not useful
That is my post above, which quotes another post, which in turn quotes the piece that says you should use LLMs as a "research assistant."
Rinse and repeat 100 times: whoops, no use case left, so it's just a cool but unusable tech demo.
Oh, have you seen this explanation of the coincidental similarities between how psychic cold reading and LLMs work?
https://softwarecrisis.dev/letters/llmentalist
But the tool they think it is is a Swiss army knife, when its limitations make it more of a microwave: really good at heating up soup, not especially useful for car repair or bricklaying.
Also less energy intensive to run.
I dropped Fädernas Gudasaga by Viktor Rydberg into NotebookLM, and I was able to interrogate a Swedish text from 1887.
Using them is a skill. You have to work on it, like any skill.
Dismissing them out of hand is childish.
Asking it, for example, to create an outline for a presentation can be low-risk, high reward.
Think of it like you might a really smart intern with limited experience.
If I ask ChatGPT to create an outline, it will put the things into the outline that others have already determined to be pertinent to the topic, plus potentially make some stuff up.
*If* I know enough to recognize the issues, I can avoid that risk, but at that point the cost/benefit is clearly not in my favor.
It gives me an elaborate result, including this (copied from the site):
A. Slavery
The economic and moral divide between the North and South.
The abolition movement and its opposition.
B. States’ Rights vs. Federal Authority
Tension over the balance of power between state governments and the federal government.
The industrialized North vs. the agricultural South.
Tariffs and trade policies.
D. Cultural and Social Divides
Regional identity, traditions, and lifestyles.
Influence of the Fugitive Slave Act and Dred Scott Decision.
For those who invest the time and effort to make it work well, and who think through the use cases, there can be high rewards.
Like most tools, if applied to the wrong use cases or without care, there can be adverse results.