Hacker News Reader

Comment by NikolaNovak | Hacker News Reader

All you can get in a survey like this is anecdotes and opinions. This may be sufficient for your purposes - here's my story:-)

My sister is not technically minded. She has used chatgpt to create a request for permit for a backyard deck to city Hall, a letter to her boss requesting attendance to a conference, and her performance reviews.

Another non techie and I use chatgpt to help us learn French and music theory.

My mother in law uses it to create funny rhyming stories for kids.

Meanwhile the techie friends of mine are... Completely ignoring it. My two best friends are vmware senior architect and a Java developer / tech manager, and I've been urging them for months to try it.

So I personally live in a completely opposite situation to that which you describe :-). Techie are skeptical of the toy, and endlessly discuss it's limitations and impact. Non Techies are just using it as a tool.

View on HN

How do you know that the French and music theory it’s teaching you is correct?

Would your question change if the tutor was a person? How do you know if anything anyone teaches you is correct?

Humans are likely to share when they don't know something or to express concern when they aren't positive what they're is correct. LLMs are confident 100% of the time and don't understand when they're wrong. I get what you're trying to say but I don't think this is an instructive example.

> Humans are likely to share when they don't know something or to express concern when they aren't positive what they're is correct.

We must have met very different humans. Are there humans that do that this? Absolutely! Are they in the majority? Absolutely not. Now if you change that framing to "teachers" then I think on average you are going to get more people like that but I've heard many many people say things with complete confidence/certainty that was absolutely wrong. Then again, I've had teachers that have made predictions/statements that they state as facts so I don't know. Dunning-Kruger can account for part of it but still.

There are good teachers and bad teachers. LLMs are at best, bad teachers.

>>LLMs are at best, bad teachers.

I could not disagree more.

ChatGPT is patient. That's a rare quality in humans and teachers.

ChatGPT will willingly explore. That's also a rare quality in teachers.

ChatGPT is detailed and structured, and has instant access to enormous amount of data and background.

I will grant you that there are domains of knowledge and questions where it's great, and where it'll lie to you through your teeth. But as a patient, detailed, willing, knowledgeable tutor in basic/well-covered areas, it's virtually unparalleled. I'm a hungry learner and have had large number of teachers, tutors and mentors through several continents, countries, societies and educational paradigms; and only the very very top are as good - and I've actually been the lucky one. My wife and my sister, for example, based on their accounts, simply never had a teacher/tutor as good as ChatGPT :-<

>>Humans are likely to share when they don't know something

I try hard to not be snarky or sarcastic on HN; it doesn't contribute to positive, friendly, productive conversation... but come ON :->

Average human is unlikely to admit to themselves or others when they don't know something. We are famous for it. We love our opinions, and we conflate them as facts. It doesn't even imply malice or anything - have you ever asked for directions from somebody who doesn't know the answer? They'll still TRY :->. Even the smartest people around me will frequently present their ad-hoc opinions as facts. Heck, on HN alone, people will opine on matters of law and science and many things which are reasonably factual.

Don't get me wrong, LLM's hallucinating and being utterly unable to signal when they do is BAD; it makes them very different from most other software we've ever built; and it needs to be addressed; but as to this line of conversation specifically, it makes them (without any philosophical or "conscience of machine" implications) extremely human-like :-)

(similarly, not to say there aren't any humans who are humble and/or explicit about their limitations; but it's far, far from average)

I am French, living in Germany. From my experiments with ChatGPT (I have access to GPT-4), the French is not going to be perfect, but so good that anyway nobody will ever notice the errors.

ChatGPT is a wonderful tool to improve your knowledge of a given language (at least for the ones having a lot of data on the web, I tested only French, German and English). You paste a mostly good text, you get back a very good one and sometimes it provides you with a description of your errors, why and what it restructured.

How do you know that the French and music theory that a human teacher would teach you is correct?

Not the really the gotcha you think this is.

Why?

It has been my experience that for things that may be slightly or very obscure, and/or controversial/uncertain, and/or with limited # of sources, and/or when pushed into complex dialogue or tricky questions, chatGPT can really muck things up.

At the same time, when asked simple, clear and basic questions, on topics where there are myriad of resources, that agree with each other, it's very very very useful.

I'm not learning either French nor Music theory solely from ChatGPT of course; I use it to supplement and explore. As such, it's proven to be in the top 5%+ of patient & knowledgeable tutors/instructors I've ever had (or to put it another way, my piano teachers have been WAY more wrong, WAY more often about music theory than ChatGPT - we all take risks :)

Exactly this. I've been scratching my head about the belittlement in tech circles of the phenomenal breakthrough that chatGPT and other LLMs represents. I think it boils down to a conflation of a few different factors:

1. Our jobs are obviously (soon) replaceable or at least massively impacted - just like all the jobs we ourselves have replaced previously. We don't like that and throw stones at it to make it go away.

2. We used to be the goto-source for all things technical, now (soon) people can just ask chatGPT and get a better answer. We don't like that and throw stones at it to make it go away.

3. We are (most of us) schooled to think that a script, a program (made by us) produces a predictable result: every letter has to be carefully placed or the whole thing comes crashing down. This AI thing doesn't work like that at all, which we don't like and etc.

4. The answers given to us by the chat bots are sometimes just "hallucinations", because the tech isn't fully mature yet. We don't like that.

More?

I agree with all of these; not necessarily in order of importance, but yet, these are factors.

I am at once astonished how useful LLM can be at its best, and how horrifyingly dangerous it can be if/when overused by those who don't understand its limitations.

You are very balanced :-) . Of course, there is valid criticisms of the hype and the way LLMs are, and will be, used but my annoyance at what I perceive as ignorant, self-interested and progress-hostile attitudes from my own industry sometimes gets the better of me.

When the dust settles, and the consensus concludes the obvious namely that: "yes, this thing actually does possess intelligence and knowledge, on an unprecedented level with vastly greater future potential" then maybe we can concentrate on steering education and professions into directions that will make the tech controllable and even more useful, and avoid the real pitfalls and dangers that lie ahead.

As long as we're stuck on the understanding that this is "just a hallucinating, stochastic parrot" and everything else is pure hype then we're not getting the right things done I'm afraid.

Some of us techies are also using it a tool, which is what it is, basically. E.g. I had to write a bunch of Groovy and since I don't write a lot of it all the time, I need to occasionally lookup some syntax etc. So I have the ChatGPT window open on the side and its answers, while not entirely perfect, are more than enough to keep me going at a brisk pace. No speedbreakers like having to scroll through bunch of non-sense search results and reading through answers on Stackoverflow to find what I'm looking for. 80-90% of the time, I don't even need to ask a follow-up question - the first answer is enough.

It is basically like a lubricant for my thought process.

Reply on news.ycombinator.com