Part of me wonders whether there would be this much hype without COVID and the Ukraine war and the economy causing so much gloom. People are yearning for a savior of some kind it feels like (I say that as a comment on human nature and as an atheist, this isn’t a disguised Christpost).
There'd be even more money pouring in, less focus on a general dystopian slant to it, and less obsessing over labor (COVID was what really spooked a lot of people on that)
Anyway I think chatgpt is cool but the issue with these at scale is there’s no way to actually verify or vet its results.
It’s trained on some data and exists within those confines. If it doesn’t know it makes up stuff even when you’ve provided a lot of information in the context.
Now I’m not doing some meme tier crap like make a function that gives me a color hex code.
I’m doing some more complex Postgres debugging and thought I’d use chat gpt and literally getting recommended function names that exist in BigQuery and not in Postgres.
Imo, these tools will find their niche in ultra specialized workflows where there isn’t much requirement for intuition or judgment, and only raw output is needed.
Making data for testing purposes or helping to write scaffolding for common lcd programming tasks, coming up with some headline text or turning a Charles Dickens novel into a ten minute story. Low risk but huge time sink type of work.
I’m also using open ai api for my product.
Just telling you my experience after using this technology for at least 100 hours. Most of my focus is on software eng and while it’s convenient sometimes I’d say it’s accurate maybe 20% of the time.
Maybe I’m wrong but I fail to see this taking hold in areas that require intuition/verification. It’s great tech for sure but the more I use it I’m like meh.
I think it will be a good replacement for some entry level jobs and manual jobs like data entry or whatever. Anything that is foo foo and doesn’t require >50% accuracy. It might be good for giving you summaries or generating headlines. Again though it’s painfully obvious when something is generated by gpt 4.
It cannot be used in any scenario where the results need to be reliable.
For me, I dont know if ChatGPT will be significant or the Palm Pilot to todays iPhone, but I do feel this AI path is going to be a game changing for human development some time in the not too distant future, hopefully for the better.
Right now I'm worried about the consequences of over-estimating its capabilities - AI that almost does the right thing is going to be misused.
I ask because I have no experience with A.I.s inventing APIs.
They were the killer application for personal computers.
Whether you were running a Fortune 500 company or a local club or just your own household, updating a single number in a column and immediately seeing the result reflected across all your calculations in a visual way was amazing.
One software innovation I think we often forget about but that is pretty extensively used day-to-day by almost every trade, is the concept of a wiki, a collaborative knowledge base. It exists most famously as Wikipedia, but many other implementations exists (like Notion, or just a Google Drive/synced folder with text files) and can be found in most places in almost any type of business.
And to train LLMs.
It’s hard to find anything that runs on a computer today that wasn’t compiled by a C compiler, compiled by something bootstrapped by a C compiler or being executed on some platform compiled by a C compiler.
So if a parallel has to be drawn, I'll say it's WordStar and Visicalc, even the use case is similar, mostly word processing and analysis.
It’s not quite that magnitude yet. But the direction seems close.
This may very well change months and years from now, but as of now it’s too early to claim it has iPhone-level impact.
I would consider the iPhone a lesser productivity impact than the PC.
Millions are running Unix in their pocket but few are aware of it.
So yeah, this first impression of mine kinda missed the whole potential inherent in the simple hyperlink mechanism.
And likewise with most initial reactions to LLM A.I.s methinks.
I was recently comparing it to friend like the before Google era and after
It’s more a ecosystem.
VR has its place but I think these language models are way more broadly useful.
I felt that the broad range of professionals I've met using ChatGPT in their daily work, already so soon after launch and while the technology is still unproven, was suggestive of its applicability across demographics.