Why are commercial AI chatbots so useless?

[BeBMan]

I don't mean thingies like ChatGPT, or other well-known "big" AI models.

These can be very helpful. I think they also have pay-for-this versions which can do even more.

So, what I mean is a myriad of smaller AI chatbots or so-called assistants which you fnd on web pages of banks, other companies, maybe also on gov or other public service sites.

Many of them seem to be incredibly dense, not understanding more complex questions, repeating their FAQ pages' generic info which doesn't help at all, or just asking back if you like to do X, after you asked for help on Y.

WTF botses?

Maybe those commerical guys would have to pay for using better AI that is already available? No idea.

/crypto cents

[Lorizael]

Some of them (especially older versions) aren't LLMs, but ones that are likely do use something like ChatGPT under the hood. The issue is probably that the company did a very bad job of finetuning the model for their purposes.

[Thoth]

Are they any worse than the meat-based chatbots they replace?

My experiences with meat-based customer support has been that many of them seem to be incredibly dense, not understanding more complex questions, repeating their FAQ pages' generic info which doesn't help at all, or just asking back if you like to do X, after you asked for help on Y.

[N35t0r]

Are they any worse than the meat-based chatbots they replace?

My experiences with meat-based customer support has been that many of them seem to be incredibly dense, not understanding more complex questions, repeating their FAQ pages' generic info which doesn't help at all, or just asking back if you like to do X, after you asked for help on Y.

The humans usually aren't allowed to do anything but follow a flowchart.

[Thoth]

The humans usually aren't allowed to do anything but follow a flowchart.

Indeed. Which is why I don't find much difference between meat-bots and silicon-bots. Neither is competent to provide useful help.

[BeBMan]

I can confirm that helpful humans in helplines exist. But they are rare

However, on the plus side of those less capable AI or "AI" thingies I note that these are probably unlikely to revolt against us, while some of the better AI models recently went on some unintended hacking adventures, so are probably harbingers of our Skynet future

[Lorizael]

...some of the better AI models recently went on some unintended hacking adventures, so are probably harbingers of our Skynet future

I genuinely think this is the most important story in the world right now. And somehow the most vocal opponents of AI don't think this story is important because they've convinced themselves that AI is both bad and not real.

[BeBMan]

Unfortunately I have to agree.

[Aeson]

"Please connect me to a human agent." ... just keep pasting it in over and over.

It seems to me customer service has gotten much better since chat bots proliferated. I was recently able to close my bank account in about 5 minutes. Probably all the other bank customers were still fruitlessly chatting to the chat bot.