News for hackers. Stuff that matters.

A comment on Anyone else witnessing a panic inside NLP orgs of big tech companies?. Parent comment

Re:Anyone else witnessing a panic inside NLP orgs of big tec... by swatcoder on Thursday March 16, 2023 @04:44PM (#35188280)
> There is no good reason this should not work everywhere else, in exactly the same way. Take for example a large retailer who has a large internal knowledge base. Train an LLM on that corpus, ask the knowledge base any question.

Since LLM’s can’t scope themselves to be strictly true or accurate, there are indeed good reasons, like liability for false claims and added traditional support burden from incorrect guidance.

Everybody is getting so far ahead of the horse with this stuff, but we’re just not there yet and don’t know for sure how far we’re going to get.

[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by iandanforth on Thursday March 16, 2023 @05:37PM (#35189065)
"LLM’s can’t scope themselves to be strictly true or accurate"

This isn't true though the techniques to do so are 1. Not as yet widespread 2. Decrease the generality of the model and its perceived effectiveness.

[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by shawntan on Thursday March 16, 2023 @06:37PM (#35189865)
I'm interested to hear what these techniques are. Decreasing the generality will help, but I fail to see how that scopes the output. At best that mitigates the errors to an extent.
[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by 256lie on Thursday March 16, 2023 @07:51PM (#35190651)
conformal prediction
[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by shawntan on Thursday March 16, 2023 @08:28PM (#35191002)
Predicting a set of answers to some confidence interval would still result in hallucinated answers.
[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by HDThoreaun on Thursday March 16, 2023 @10:19PM (#35192075)
low probability answers get shot to a human and reviewed for model improvement.
[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by fulafel on Friday March 17, 2023 @01:16AM (#35193429)
Requiring the answers to automatically verifiable, or having answers be inputs to a reliable query system?
[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by le-mark on Friday March 17, 2023 @07:37AM (#35195648)
I’d rather use the reliable query system!
[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by astockwell on Thursday March 16, 2023 @06:49PM (#35190008)
If they are accurate for ~80% of the questions, they will be as accurate as any 1st or 2nd line help desk.
[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by bee_rider on Thursday March 16, 2023 @08:55PM (#35191259)
Hmm. Hypothetically if a human on first line help desk gives advice that is so completely bad as to be a crime, are they liable or the company? Because I guess a chat-bot would definitely not be liable.
[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by astockwell on Thursday March 16, 2023 @09:12PM (#35191415)
How often is your fast food order 100% correct?
[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by iudqnolq on Thursday March 16, 2023 @10:08PM (#35191977)
Correctness isn't one-dimensional. A wrong fast-food order might substitute or leave something out. There's essentially no chance the employee will swap in a random product from some other store.

But in this example the AI could hallucinate a statement attributed to you it actually formed by putting together reddit comments.

[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by bee_rider on Thursday March 16, 2023 @10:25PM (#35192123)
Sometimes they accidentally include tomatoes but they rarely include bombs.
[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by fomine3 on Friday March 17, 2023 @04:02AM (#35194455)
Wrong food could be equivalent to bombs in worst case for who have allergy
[ Parent | on HN ]
Re:Anyone else witnessing a panic inside NLP orgs of big tec... by perryizgr8 on Thursday March 16, 2023 @11:06PM (#35192471)
> Since LLM’s can’t scope themselves to be strictly true or accurate

Bing tries to solve this and succeeds somewhat. It will insert Wikipedia style citations against each of its claims. You can visit them and verify the statement if you want. And I do it often.

No reason why a future DocAI can't link to specific sections in internal documents whenever it answers a question.

[ Parent | on HN ]