Disillusioned Businesses Discovering That AI Kind of Sucks
futurism.com
futurism.com
We've gone through AI winters before, where the actual techniques and hardware simply had a terminal point of usefulness beyond which it was unlikely to grow.
If hallucinating bad information is to be regularly expected / intrinsic to the tech, then it's basically Clippy 2.0 and a dead end.
On the other hand, if we can expect lower power costs and higher trust in the output (i.e. needing less human intervention) then it makes sense to start finding places where it can fit into the business and grow over time.
I'm personally in the camp that it is a fun toy but with limited applicability for most businesses, and unlikely to grow beyond that. I'd love to be proven wrong, though.
There are lots of other applications - in particular I'm thinking about turning unstructured data into structured data for aggregation and analysis, where you can tolerate some level of error and LLMs can work quite well to automate it.
Beyond that, obviously copilot or assistant type stuff is already something people use. It doesn't have to be flawless for that, the main use case for chatGPT is not making high stakes decisions blindly based on the answers it gives out. It makes mistakes and does dumb stuff sometimes but it's also helpful, the flaws aren't a dealbreaker even if people would like them to be.
At the same time, I don't expect it to get much better, it's going to be about getting useful work out of what we have.
So far, simply throwing more compute at the problem is making models hallucinate less and less. I think this easily explains why all the big players are making such massive investments in hardware, reliability seems to be an almost solved problem.
It's definitely something more than a toy, but I'm not sure that you can build a trillion dollar industry on that kind of stuff.
But this isn't true in general, you can easily train a local model to write in very localized styles, and include temperatures that allow for wild swings outside of average.
If you want a rambling, occasionally brilliant Kerouac or de Montaigne you can make one.
I have also had success with classification tasks, like look at this email and then look at this list of topics, and pick which topic this email relates to, or “other” if you can’t see an obvious choice.
But you can’t say “hey AI do this person’s job for me”.
Unfortunately, that's exactly what businesses want to accomplish.
If AI needs to steered around by an expert to be effective that's going to limit its appeal.
ChatGPT chickened out and kept giving me functions that looped through applying various filters and regexes.
This time when a similar thing happened I was able to fix it before he even had to ask how long it might take and the solution is more robust and able to handle several different formats now.
It was a definite win.
I don’t see AI replacing experienced professionals any time soon, but I do see it replacing the need for some entry level employees soon.
"Computer, in the Holmesian style, create a mystery to confound Data with an opponent who has the ability to defeat him."
It seems to work pretty well for this scenario. The usage is company internal (only draft generation) and the sales reps are augmented by the AI, not replaced by it.
The problem is that the toothpaste is already out of the tube in terms of what it's been hyped to be for the average person, and they are disappointed to learn that it doesn't really work that way. And there are many people who have not yet realized, or realized too late, that it's not appropriate for the work they're doing and its "assistance" is ending up in finished products in any case.
It may not be factually accurate. It may say disturbing things. It is not a reliable company representative. This does not mean AI sucks, it means you are trying to use it for things it is not appropriate for.
On the other hand, it can be a massive time saver for staff who know what they're doing and can interpret it's output. It's a tool that can boost output, not a replacement for people.
Not had that vibe for a while now, but I definitely noticed it.
But instead of that, the article starts with
The tech's drawbacks are hard to overlook. Large language models like ChatGPT are prone to hallucinating and spreading misinformation. Both chatbots and AI image makers have been accused of plagiarizing writers and artists. And overall, the hardware that generative AI uses needs enormous amounts of energy, gutting the environment.
Then
Perhaps most of all, according to Gary Marcus...
And if it can "surely" do useful work... why isn't it? That's like saying, "Sure, this car burns oil, but if it can do that, it can surely do more than that and be my daily driver."
It's also being treated as magic, with all the same unrealistic expectations.
To treat it as more than it is, is as much a mistake as to treat it as less than it is.
I'm glad it's "only" at the level of an intern (sometimes you're willing to hire them, sometimes you're not); but conversely I think this means I can't risk changing career.
They make mistakes, carry potential legal liabilities and cost a fortune ( dollars or carbon ) to run. The only difference with before is that they now look a lot closer to being able to replace people for real, and that’s a significant difference.
Now, replace ‘ai’ with ‘employee’ and the profits must be very clear to hire such an unreliable person. Some companies worry about these things.
In my view if the amount of mistakes could be acceptably low, this would change the game ( both ‘ acceptably’ and ‘low’ are valid directions).
Filtering out the truth is the hard part. If you just want data and don't care about quality there's always /dev/urandom.