nmenaker

The training date cut off makes the usability of ChatGPT somewhat limited, for many things. I’m sure they plan to enhance their PAID offering with updated or near real time data at some point.

but, from what I have seen in the past 10 months, when one is trying to go outside the more simple data/fact set referential data summary and elaboration, I can see where it quite literally MAKES S..TT UP… I know, I know this is called “hallucination” but still, it’s a very odd feature in a system that SHOULD simply organize, stack rank and present facts in a user grammatical way. This isn’t unique to ChatGPT, BARD will MSU as well, I haven’t seen it as much but colleagues have given me examples.

I understand why it is happening, but don’t understand why it can’t be regulated, controlled or put guard rails in. at some point, presenting “your work” will help reduce this

HDQDD

I’ve also used AI for some software stuff. I find it’s usually about 80-90% right. What is nice is that you can have it keep iterating/refining what it’s already generated. For example I can ask it to make me a python script to do some task…then I can add something like “let me pass in x as a variable via a CLI argument”…and it updates the code…”log all the output to a file named log.log”, etc. etc. it also does a much better job that I do at commenting and documenting code as it goes along. (Even if I have to correct it later)

it also does a decent job of making bash scripts for Linux or simplifying CLI tasks…like “give me a Linux command to find the five largest files on this machine that haven’t been modified in over a year”. Yeah I could read through man pages, or figure it out via searching the web (probably through multiple pages to refine my command, but AI does it in 3 seconds.

deorio

I used the tool as more of an inspiration and idea generator. It’s able to at least summarize the most popular sights and attractions for major cities. Past that, you’ll still have to do some digging for the more local experiences.

Another great capability is plugging in all the things you want to do, plus the amount of time you have. It does a great job with creating you an itinerary based on location of sites and average time need per attraction.

zdt_ft

Now that ChatGPT gpt-4 can browse the web, and custom GPTs allow you to upload proprietary knowledge and give it APIs to call, there are some unique applications you can build, such as itinerary building and then searching for award availability using different credit card points/miles programs:

WilcoRoger

That's super handy!

ETA - one must be really careful with the code the bing AI returns. I asked a simple test question to list the directory names, that contain mp3 files. It took many rephrasing for bing to understand my intention and NOT to list the mp3 files, not to delete the mp3 files, etc etc. Even the final reiteration had a small error, so that all directory names got listed, but the was easy to correct by hand.
But yes, still much faster than checking man pages, etc.

carljross

It's interesting to hear Bing is now incorporating GPT-4, while ChatGPT currently uses GPT-3.5. The capabilities of these large language models do seem to be rapidly evolving. However, as your biography test illustrates, they can still generate factual inaccuracies and fictional content when prompted. As impressive as systems like ChatGPT online at GPTOnline.ai are, it's important to maintain reasonable skepticism about their limitations. They don't have a true understanding of the world and can make up plausible-sounding but false information.

thomasfly

I have also been using Bing Chat for my travel questions, then I will validate them. I also read news recently about Bard which now uses Gemini Pro, I will be trying it from now on and see how it measures up to Bing Chat. Has anyone used it and got satisfactory answers from Bard with Gemini Pro?

DMSFCA

I used Google's Bard with the Google Flights integration, and it worked quite well. I could say, "Find the cheapest flights to MEX from SFO for a 4-day stay in February on United," and it would do just that. Pretty impressive. You can also ask for a "4-day itinerary showing the highlights of Mexico City," and it does a great job.

You can refine it by adding "include fine dining and make the locations walking distance from the W Hotel" or any other preferences to tailor it to your liking. It's quite fun to play with and get inspired.

nmenaker

These are the types of AI assistant type of queries and tasks that are going to be some of the primary early quality experiences for consumers.

“Find me flights on these dates with my standard seat preference available, time of day preferences / range, NOT X yes Y, etc.
”Order everything I need for making our spaghetti dinner tomorrow night and ship to home”
“Find me the current latest sales on ebay for X item, with Y qualities, and prep an auction for me to confirm”
I could go on, forever.

thomasfly

I've been experimenting with a few large language models myself, like Google's Gemini, Bing Chat, and Claude 3 Sonnet. They're all impressive, but each with their own strengths. While AI can be a great starting point, it's important to double-check the information it provides, especially for travel planning where accuracy is very important.