Something AI should be able to do but for some reason can't

[89th]

Did you try google?

[89th]

But yes it should (and I'd imagine very soon) will be able to be connected (via API/whatever) to booking information, similar to the travel websites today that look & hook into hotel reservation systems.

[Klaus]

ChatGPT will not even look at a website you specify. For instance, you cannot say "summarize www.whatever.com".

I assume it has to do with security/safety issues.

[Klaus]

It does kind of work if you use the right GPT, though.

[jon-nyc]

Yeah I bet that’s wrong. Ain’t no suite going for $250.

That’s the tedious part. Google can tell you which hotels have availability.

[Copper]

[Axtremus]

By some accounts, Siri was initially built to do things for you instead of just answering questions, but Apple Inc. steered Siri into a different direction than when the inventors/founders envisioned.

ChatGPT and Grok are mere language models rather than "do engine" -- they are not designed to do things for you, not beyond generating text based on pretraining data.

[Aqua Letifer]

@jon-nyc said in Something AI should be able to do but for some reason can't:

Yeah I bet that’s wrong. Ain’t no suite going for $250.

That’s the tedious part. Google can tell you which hotels have availability.

AI takes in data and predicts what you most likely are expecting in terms of a response. Couple that with how form page copy is written—it leaves out a lot of exposition in lieu of its design context—and I'm not at all surprised it can't do this yet. It's not the kind of question it excels in answering.

[jon-nyc]

Right but the newer version do search the web in real time.

[Aqua Letifer]

@jon-nyc said in Something AI should be able to do but for some reason can't:

Right but the newer version do search the web in real time.

@jon-nyc said in Something AI should be able to do but for some reason can't:

Right but the newer version do search the web in real time.

Yes but I'd refer you to my comment about form page copy and how that is interpreted as text versus something that is straight exposition. It's much harder for models to make predictions based on form copy because it's so unlike its training data and leaves a lot of context out.

For example, those pages don't say "We have 3 suites available for 12/31 for one night." Just look at the copy on those pages and only the copy. It's a lot harder to generate predictive text based on something so sparse.

It'll get there most likely, just not there yet.