AI generated heartfelt letter to parent - GIF

[Jon]

In this chapter of our Brave New Fake World: a tech demo at a conference. The technology being demonstrated is ChatGPT, which is being touted as "Artificial Intelligence". It's really just an LLM (Large Language Model) that identifies relationships between words in a conversation-like manner. This is useful technology, kinda like an easier to use search engine. But it's no more "artificial intelligence" than a dishwasher.

This little bit shows how ChatGPT can be used to compose a letter. Also useful, perhaps if you need to write a letter to your insurance company documenting a claim, or your bank confirming your account ownership.

Here they demonstrate, complete with touching piano music, how ChatGPT was used to compose a heartfelt letter from a daughter to her father, expressing her love for him.




Previously:

Fake car noises in electric vehicles - photos
World's oldest customer complaint letter - photos

[Philip Davies]

“O brave new world!” (said Miranda).
Does this render English teachers redundant?

[Jon]

Universities are already having problems screening for students cheating. Previously the professors had a simple way to see if they were copy-pasting existing text from the internet. But the LLMs can generate unique text and even style it accordingly, based on an existing writing sample.

The newer image recognition abilities might be more useful.







Programmers are definitely sweating. Not only can it generate computer code with a text prompt, but it can do so by reading images or whiteboard diagrams.

[ductape]

The stack of street signs is bureaucracy, not artificial intelligence.

Now that I think about it, disregard this comment.

[Frank S]

“O brave new world!” (said Miranda).
Does this render English teachers redundant?

I often wish I would have paid a lot more attention to most of my English teachers. I almost gave a little too much attention to one of my high school teachers though.

[nova_robotics]

That's nothing. GPT4 is f'n incredible. I was having a conversation with it about the flow through a nozzle a few months ago. Then just for kicks I asked it to write a program to simulate the flow through a nozzle. In about a minute that thing had written a very complicated program, including all of the required dimensions with perfectly commented code, it knew that it needed to use Navier-Stokes and not only did it understand the equations, it wrote C code to implement those equations within the context of the problem. Super advanced stuff, and I didn't direct it on any of these things. It just knew what it needed to solve the problem. If you handed the exact same problem to a person it would probably require someone with a Masters level education and it would take them all day to do. GPT4 cranked it in around a minute. It's mind blowing. It can write just about anything, including technical papers and diagrams.

The only problem is it has a very detectable writing style once you get familiar with it. It's very... flamboyant, for lack of a better term. If you get it to write anything you have to de-fluff the document afterwards.

[Jon]

I still get issues with what they call AI "hallucinations". It will answer a question confidently and in detail, but it will be point-blank wrong. Sometimes I will ask it if it is certain about its answer, and then it will apologize and provide the correct answer.

I think those are largely kinks that are being worked out of the system now that it's being trained on a larger population.

But yes, the era of third-world coders cranking out if/else statements is probably over.

I'm not seeing much value from it yet for in-depth information on building tools.

As for the AI image generators, here is what they think of a homemade table saw, bandsaw, lathe, and screwdriver. The well-coiffed gentleman uh, "operating" the lathe has the classic AI six-finger error.

[bruce.desertrat]

All I can say is the guy working on that lathe like contraption should steer clear of Inigo Montoya. Or at least have that death ray ready when Inigo shows up!

[bruce.desertrat]

My former boss was able to get a lot of useful Powershell done with ChatGPT; I suspect it's because the underlying LLM dataset includes a LOT of sample and posted code as well as engineering texts and technical documents etc. This is probably a very good use case for AI.

Getting it to 'paint' stuff is more problematic.

[Jon]

All I can say is the guy working on that lathe like contraption should steer clear of Inigo Montoya. Or at least have that death ray ready when Inigo shows up!

Couldn't resist: