I think it’s a very dangerous place to be in an area you’re not familiar with. I can read Python code and figure out if it’s what I want or not. I couldn’t read an article about physics and tell you what’s accurate and what’s not.
Legal Eagle has a great video on how ChatGPT was used to present a legal argument, including made up case references! Stuff like this is why I’m wary to rely on it in areas outside of my expertise.
> I’m responding to this comment, where I think it’s clear that an LLM can’t event achieve the goal the poster would like.
I know it can't - the one thing it's missing is the ability to generate coherent and correct (and not ugly) ___domain-specific illustrations and diagrams to accompany the text. But that's not a big deal, it just means I need to add some txt2img and img2img models, and perhaps some old-school computer vision and image processing algos. They're all there at my fingertips too, the hardest thing about this is finding the right ComfyUI blocks to use and wiring them correctly.
Nothing in the universe says an LLM has to do the whole job zero-shot, end-to-end, in a single interaction.
> I wouldn’t but a lawyer actually went to court with arguments literally written by a machine without verification.
And surely a doctor somewhere tried to heal someone with whatever was on the first WebMD page returned by Google. There are always going to be lazy lawyers doctors doing stupid things; laziness is natural for humans. It's not a valid argument against tools that aren't 100% reliable and idiot-proof; it's an argument for professional licensure.
Your entire argument seems to be “it’s fine if you’re knowledgeable about an area,” which may be true. However, this entire discussion is in response to a comment who is explicitly not knowledgeable in the area they want to read about.
All the examples you give require ___domain knowledge which is the opposite of what OP wants, so I’m not sure what your issue is with what I’m saying.
> Its good to not trust it but that's not the same as it having no idea. There is a lot of value in being close for many tasks!
The task is to replace hazelcast with infinispan in a stand-alone IMDG setup. You're interested in Locks and EntryProcessors.
Ghat GPT 4, o1 tell you with their enthusiastic style Infinispan has all those features.
You test it locally and it does....
But the thing is infinispan doesn't have explicit locks in client-server mode, just in embedded mode, but that's something you find out from another human who has tied doing the same thing.
Are you better off using Chat GPT in this case?
I could go on and on and on, on times Chat GPT has bullshitted me and wasted days of my time, but hey, it helps with one-liners and Copilot occasionally has spectacular method auto-complete and learns on the fly some stuff and it makes my cry when it remembers random tidbits about me that not even family members do
Given I have never heard of any of {hazelcast, infinispan, IMDG, EntryProcessors}, even that kind of wrong would probably be a improvement by virtue of reducing the time I spend working on the wrong answer.
But only "probably" — the very fact that I've not heard of those things means I don't know if there's a potential risk from trying to push this onto a test server.
You do have a test server, and aren't just testing locally, right? Whatever this is?
>. You do have a test server, and aren't just testing locally, right? Whatever this is?
Of course I didn't test in a client-server setup, that's why chat gpt manage to fool me, because I know all those terms, and that was not the only alternative I looked up. Before trying Infinispan I tried Apache Ignite and the api was the same for client-server and embedded mode; in hazelcast the api was the same for client-server and embedded mode, so I just presumed it would be the same for Infinispan AND I had Chat GPT re-assuring me.
The takeaway about Chat GPT for me is -- if there's plenty of examples/knowledge out there, it's ok to trust it, but if you're pushing the envelope, the knowledge is obscure, not many examples, DO NOT TRUST it.
DO NOT assume that just because the information is in the documentation, chat GPT has the knowledge or insight and you can cut corners by asking chat GPT.
And it's not even obscure information -- we've asked Chat GPT about the behavior of PostgreSql batchupserts/locking and it also failed to understand how that works.
Basically, I cannot trust it on anything that's hard -- my 20 years of experience have made me weary of certain topics and whenever those come up, I KNOW that I don't know, I KNOW that that particular topic is tricky, obscure, niche and my output is low confidence, and I need to slow down.
The more you use Chat GPT, the more likely it will screw you over in subtle ways; I remember being very surprised about how could so very subtle bugs arise EXACTLY in the pieces of code I deemed very unlikely to need tests.
I know our interns/younger folks use it for everything and I just hope there's got to be some ways to profit from people mindlessly using it.
Yes. Despite this apparently popular saying, "close enough" is sufficient in almost everything in life. Usually it's the best you can get anyway - and this is fine, because on most things, you can also iterate, and then the only thing that matters is that you keep getting closer (fast enough to converge in reasonable time, anyway).
Where "close" does not count, it suggests there's some artificial threshold at play. Some are unavoidable, some might be desirable to push through, but in general, life sucks when you surround yourself or enforce artificial hard cut-offs.
