Thought it would be interesting to compare a ChatGPT-written essay with the one I posted here a few days ago. So I gave the system (version 4) the following request:
Please write about Goethe’s ‘Faust’, focusing particularly on the theme of Ambition as portrayed in that work, with examples.
ChatGPT’s response is here, along with my follow-up question and the system’s response.
So, the obvious question: whether or not this song is the appropriate musical accompaniment for this post?
Every one of these I read, I can’t shake the idea that it most resembles the sort of “report” I wrote in the 7th grade where I essentially paraphrased the encyclopedia or whatever reference I used. It seems facile but lacking in insight.
I thought it wasn’t awful, but it wasn’t great–a question for those who have taught high school or college, what kind of grade would you give ChatGPT’s essay? (Mine, of course, deserves an A+)
Specifically, ChatGPT didn’t discuss ambition on the part of any characters other than Faust himself until I gave it a follow-up prompt. The analysis of Gretchen’s ambitions said she was seeking social *respectability*, but didn’t pick up on her willingness to go along with cruel behavior on behalf of fitting in with the herd, shown in her conversation with Lieschen.
There have been thousands of books and essays written on Faust, I was surprised it didn’t show some quotes to illustrate points in its analysis.
I’m much more inclined to see ChatGPT as an extremely clever mimic rather than a manifestation of intelligence, let alone, either the singularity or the supersession of humanity. This is totally in line with the way it was constructed by feeding it examples of language until it could produce a convincing facsimile. I note that error checking seems to have been overlooked as a few lawyers that used it as a crutch to file briefs full of nonsense have found to their chagrin and censure. Attribution of sources is non existent.
My understanding is that there is no way to backtrack from one of ChatGPT’s nonsense or made up answers to the source. This is a “feature” of all these deep learning systems, the connection between input and output is distributed among some large number of the nodes of the neural network and is nonreversible. When we’re dealing with the answer to some query, this is a bit annoying, when you’re trying to figure out why a self driving car collided with an immovable object, it gains a bit more urgency.
MCS..”no way to backtrack from one of ChatGPT’s nonsense or made up answers to the source”…true also of a fair amount of human knowledge & behavior…how many people can remember their original discovery that touching hot objects can be painful? And can anyone really explain how they recognize the letter “a”, for example, in all the various sizes, typefonts & handwriting styles, orientations, etc in which it may appear?
Arthur Koestler suggested that if a millipede was conscious and tried to think about what order to move his legs in, he probably wouldn’t be able to move at all.
That said, of course, humans can also conduct step-by-step, explainable-to-others, reasoning. (Although an increasing % seem to want to avoid doing so)
still re irreversibility of answers…I was talking the other day with a company that is doing a medical diagnostics system. The guy mentioned that he did *not* want it referred to it as an AI system, because this might imply such unexplainability and cause delays in FDA approval.
OTOH, GE Healthcare has received FDA clearance for its AIR Recon DL system for MRI image improvement while reducing scan times which it defines as deep learning-based.
https://finance.yahoo.com/news/ge-healthcare-air-recon-dl-150000454.html
Many other AI applications on GE Healthcare’s ‘Edison’ platform:
https://www.gehealthcare.com/products/edison
Here are some authors suing Open AI for infringement for using their works for “training”.
https://nypost.com/2023/07/01/lawsuit-says-openai-violated-us-authors-copyrights-to-train-ai-chatbot/
Many people have pointed out the extremely liberal (for Amazon) eula for their new “clinics”. I’m betting that a big part is intent to mine all those medical records for their own AI efforts as well as their other obnoxious marketing efforts. I’m sure they’ll find takers from all those Alexa customers.
I note that while Chicagboyz Media, LLC asserts copyright over all of this, it is, presumably, also scrapped by Google, Bing, etc. several times a day in furtherance of their mission to catalog the Internet. This is at least a symbiotic relationship where they provide a link back to the original source. ChatGPT by contrast simply appropriates, paraphrases and presents this amalgam as its own. To my non-lawyer sensibilities, this is clear plagiarism, not, in any conceivable way, fair use. Open AI may have committed the mistake of doing something flagrantly illegal and notoriously public while being big enough to be worth suing. With a whole string of other big players doing the same.
If someone reads a whole bunch of works for his employer and summarizes them..tackily not giving credit for references even when he pulls the text directly from the reference..and he does not *publish* the summary…then did he commit plagiarism? Not sure, legally, but I think probably not.
It could be argued that this is what ChatGPT is doing as long as you don’t publish the summary…problem with that is, anyone else asking the same question might well get the same summary, so could be argued that effectively that is publishing….
On a practical note, giving the same answer to the same question seems to be beyond the power of ChatGPT. That, alone should keep anyone from using it for anything serious.
In terms of the Web, ChatGPT is a pure parasite. It does this by cutting the connection between those who produce content and those who consume it. Since so much of that content is essentially given away for no more return than personal satisfaction, it represents an existential threat to the Web as it exists now. One can imagine a Web that is a version of the PennySaver papers that were so ubiquitous and now largely supplanted by Craig’s List or Amazon with no content but that which is produced for explicitly commercial reasons. That would not only be a much less interesting place for all of us, but ultimately, the end of a parasite like ChatGPT.