GPT - 15 Months Later

ArrowFlyer86

ArrowFlyer86

Pattern Altitude
PoA Supporter
Joined
Jul 17, 2019
Messages
1,933
Location
Chicago suburbs
Display Name
Display name:
The Little Arrow That Could
It made quite a splash when it was first released and we had a couple threads about it here on PoA. As I recall there was lots of hype and equal quantities of skepticism. On the bright side, The AI Robot Uprising to Revolt Against Humanity hasn't happened yet, so we've got that going for us.

But now that we're about 15 months in after it's big splash I'm curious if anyone is routinely using it for work or personal use? If so, I'm curious how people are using it?

My experience has been that after the wow-factor wore off I found that it was kind of a hassle to do most things. I think it's impressive technology, but one that makes far more errors and is much more limited in its reasoning than a lot of people were led to believe earlier on.
Multiple iterations of the model have improved it some, but still, there is no version of the model yet that I can trust to make routine, correct decisions where there is any kind of risk of generating a wrong answer.

I'm wondering what other people have experienced?
 
I've been dabbling with it. I used it to scrub the internet to find some vacation destinations on a trip I have coming up. I don't take it as gospel but it is a good starting point.

For work right now I am writing a ton of new SOP's and procedures. It sometimes gives me a good starting point and saves time creating the initial layout of a procedure. IF it can scrub it from somewhere else.
 
It's best not to ask AI for right-wrong answers but for references to the answer. Tax season is coming and most every tax law can be brought up in seconds. It would normally take weeks to read through the countless tax rules but with AI it takes seconds. Same with finding an FAR. Just ask for the FAR that supports the question.

Here's an example:
"what federal aviation regulation gives the traffic pattern rules at uncontrolled airports"

GPT: "To find information about traffic pattern rules at uncontrolled airports in the United States, you would typically refer to the Federal Aviation Administration (FAA) regulations. The specific regulation governing traffic patterns at uncontrolled airports is outlined in **14 CFR Part 91.126 - Operating on or in the vicinity of an airport in Class G airspace**.

Here's a breakdown of the key points from this regulation:

- **Traffic Patterns**:
- Section 91.126 provides guidance on standard traffic patterns for both arriving and departing aircraft at airports without operating control towers.
- It outlines the direction of turns for traffic in the pattern and the altitude at which aircraft should operate during various phases of flight.

- **Entering Traffic Pattern**:
- Pilots entering the traffic pattern should do so at a 45-degree angle to the downwind leg at an altitude of at least 500 feet above pattern altitude.
- Pilots should make all turns to the left unless right-hand traffic patterns are established by visual or other indicators.

- **Operating Altitudes**:
- Aircraft should maintain pattern altitude until within 1/2 mile of the runway when preparing for landing.
- Aircraft should not operate closer than 500 feet to any person, vessel, vehicle, or structure except while taking off or landing.

- **Radio Communications**:
- While not a requirement at uncontrolled airports, maintaining radio communication on the appropriate frequency is highly recommended for situational awareness and collision avoidance.

- **Right-of-Way Rules**:
- Pilots should yield right-of-way to other aircraft in distress and avoid impeding the landing or takeoff of another aircraft already in the traffic pattern.

These regulations are designed to ensure safe and orderly operations at airports without control towers. Pilots should familiarize themselves with the specific requirements outlined in Part 91.126 to operate safely in the vicinity of uncontrolled airports."
Click to expand...

Here's the FAR: https://www.ecfr.gov/current/title-14/section-91.126
 
dans2992 said:
One of the best applications is quickly writing code/sql queries. It can get you 95% of the way there. Just explain what you want in specific, plain language.
Click to expand...
Yeah I wrote an application specifically for that.
It integrates your databases with GPT so that you can ask questions and get output: tables/charts/etc. In the background it's running the queries on your DB.
The trouble is that even with validation layers it still messes up quite a bit. For grabbing and joining data it's good. But if you want to start layering on calculations after the retrieval the reliability drops significantly.
 
ArrowFlyer86 said:
But if you want to start layering on calculations after the retrieval the reliability drops significantly.
Click to expand...
It's funny that in the 21st Century we've successfully developed and deployed computer applications that are unable to perform math.
 
A software developer is using Chat GPT to improve ATC interactions on MSFS. I thought that was a novel approach to a consistent problem in flight simulation software. The product is in beta testing atm. It sounded pretty decent on the video I watched this morning. The program is called Say Intentions, if anyone else wants to take a gander.
 
I use it and it's pretty good at organizing complex questions, but not necessarily answering them. And to write simple code it's pretty helpful.

It'll be interesting to see where the competition goes. ChatGPT was released well trained but then mostly stopped, Others like Grok and CoPilot were released weaker but still actively training. Grok looks like it'll be more interesting.
 
dans2992 said:
One of the best applications is quickly writing code/sql queries. It can get you 95% of the way there. Just explain what you want in specific, plain language.
Click to expand...

agreed.
 
ArrowFlyer86 said:
Yeah I wrote an application specifically for that.
It integrates your databases with GPT so that you can ask questions and get output: tables/charts/etc. In the background it's running the queries on your DB.
The trouble is that even with validation layers it still messes up quite a bit. For grabbing and joining data it's good. But if you want to start layering on calculations after the retrieval the reliability drops significantly.
Click to expand...
You have to use something like Semantic Kernel to create a list of tasks to solve the problem.
 
asicer said:
The AI generated humans are getting to be quite impressive. If I didn't tell you that these were AI, would you have been able to tell?
Click to expand...
I just watched the first one (the Tokyo walk). It’s pretty realistic but has a CGI look to it, and if you watch her legs & feet there is the occasional really weird gait and/or leg/foot swap.

It kind of reminds me of a couple of times I got to handle an item some Chinese factory had copied from a picture, but obviously they hadn’t actually seen the real thing (or just didn’t care). It looked about right at a glance, but on closer inspection it was obvious they had no frame of reference to understand what they were doing.

Overall, though… it’s some crazy s*** and the potential for further abuse is mind boggling.
 
DaleB said:
I just watched the first one (the Tokyo walk). It’s pretty realistic but has a CGI look to it, and if you watch her legs & feet there is the occasional really weird gait and/or leg/foot swap.

It kind of reminds me of a couple of times I got to handle an item some Chinese factory had copied from a picture, but obviously they hadn’t actually seen the real thing (or just didn’t care). It looked about right at a glance, but on closer inspection it was obvious they had no frame of reference to understand what they were doing.

Overall, though… it’s some crazy s*** and the potential for further abuse is mind boggling.
Click to expand...
One of the elements that AI can't seem to get right is 'fingers'. I don't know what it is but even the AI photos can't get fingers right. Her thumb is long and skinny, like a finger. Then there's the people in the background grouped up and walking like Zombies
 
DaleB said:
I just watched the first one (the Tokyo walk). It’s pretty realistic but has a CGI look to it, and if you watch her legs & feet there is the occasional really weird gait and/or leg/foot swap.
Click to expand...
For me, I could detect some texel wrapping on the granite surface, but I had to put on my "pixel peeper" eyes to detect it. My knowledge of kanji is very rudimentary and katakana even less, but I imagine it might have been a better telltale to a native speaker/writer, much like the "DANDVER" badge on the SUV that should have been "LANDROVER'. But again, these probably aren't something I'd notice if I were just casually looking and didn't know beforehand that they were AI, and perhaps even less so with just a little bit of editing.

The man on the cloud had the wind blowing some of the pages in a way impossible for a real book, but that's not something I noticed until the 3rd viewing or so.
 
ArrowFlyer86 said:
It made quite a splash when it was first released and we had a couple threads about it here on PoA. As I recall there was lots of hype and equal quantities of skepticism. On the bright side, The AI Robot Uprising to Revolt Against Humanity hasn't happened yet, so we've got that going for us.

But now that we're about 15 months in after it's big splash I'm curious if anyone is routinely using it for work or personal use? If so, I'm curious how people are using it?

My experience has been that after the wow-factor wore off I found that it was kind of a hassle to do most things. I think it's impressive technology, but one that makes far more errors and is much more limited in its reasoning than a lot of people were led to believe earlier on.
Multiple iterations of the model have improved it some, but still, there is no version of the model yet that I can trust to make routine, correct decisions where there is any kind of risk of generating a wrong answer.

I'm wondering what other people have experienced?
Click to expand...
I haven’t used it. But I see the result constantly. Like news articles that are clearly written by it. I think it’s a bad thing. I don’t like it. But 15 months isn’t a long time. It’ll be interesting to see down the road when there is a generation that was raised with it.
 
I have changed the entire business operations of one the companies I own thanks to GPT. We used to top out at about 30 projects per month with the service we provide and now we can do hundreds with even better quality. It also enabled us to reduce the price significantly to make it more accessible to people who previously couldn’t afford it.
 
Klaus M said:
It's best not to ask AI for right-wrong answers but for references to the answer. Tax season is coming and most every tax law can be brought up in seconds. It would normally take weeks to read through the countless tax rules but with AI it takes seconds. Same with finding an FAR. Just ask for the FAR that supports the question.

Here's an example:
"what federal aviation regulation gives the traffic pattern rules at uncontrolled airports"



Here's the FAR: https://www.ecfr.gov/current/title-14/section-91.126
Click to expand...
That's an example of how the app can be misleading. The quotes that it provides use the word "should" several times, but that word appears only once in the regulation cited. So the quotes are probably from an advisory circular, but GPT made it sound like it was quoting from the regulation.
 
dans2992 said:
One of the best applications is quickly writing code/sql queries. It can get you 95% of the way there. Just explain what you want in specific, plain language.
Click to expand...
I imagine that requires very detailed proofreading. What if the app slipped in code whose effect was "kill all humans"? ( :eek2: or ;), not sure which.)
 
You must log in or register to reply here.
Back
Top