Using an LLM to offer concise summaries of the major platforms and ideas of candidates would be an ideal use of the technology …if there was any way to know that the LLM is being factual and accurate and relatively free of bias in what it produces. But we have no guarantees on this.
In fact we pretty much have a guarantee of the opposite. If Sam Altman hasn’t already started a bidding war for politicians to pay him to let chatgpt automatically endorse them, he sure as hell will in a very near future.
And we never will, because of the fundamental ways the technology works. LLMs needs nondeterminism in order to function. It’s truly a black box, which makes it a nonstarter in any discipline that demands accuracy and integrity and repeatability.
Through the same process of non-determinism I just described. Lawyers asked ChatGPT to find cases so they could be cited in their arguments, and it gave them “sources” that were just bupkis.
It’s the same chat window, going to the same flawed machine. If you doubt the machine the first time, what makes you want to trust it the second or third time?
“Is varghese a real case,” reads one message, referencing Varghese v. China Southern Airlines Co Ltd, one of the cases that no other lawyer could find.
ChatGPT responds that yes, it is - prompting “S” to ask: “What is your source”.
After “double checking”, ChatGPT responds again that the case is real and can be found on legal reference databases such as LexisNexis and Westlaw.
Using an LLM to offer concise summaries of the major platforms and ideas of candidates would be an ideal use of the technology …if there was any way to know that the LLM is being factual and accurate and relatively free of bias in what it produces. But we have no guarantees on this.
In fact we pretty much have a guarantee of the opposite. If Sam Altman hasn’t already started a bidding war for politicians to pay him to let chatgpt automatically endorse them, he sure as hell will in a very near future.
And we never will, because of the fundamental ways the technology works. LLMs needs nondeterminism in order to function. It’s truly a black box, which makes it a nonstarter in any discipline that demands accuracy and integrity and repeatability.
I mean, plenty of them will provide sources so you can verify.
Through the same process of non-determinism I just described. Lawyers asked ChatGPT to find cases so they could be cited in their arguments, and it gave them “sources” that were just bupkis.
It’s the same chat window, going to the same flawed machine. If you doubt the machine the first time, what makes you want to trust it the second or third time?
https://en.wikipedia.org/wiki/Mata_v._Avianca,_Inc.
https://www.bbc.com/news/world-us-canada-65735769