The A.I. Thread of OMG We're Being Replaced

We’ve got multiple threads discussing A.I. stuff like ChatGPT and AI art, but we don’t have thread to really dig into the economic and political ramifications of AI displacing a ton of jobs.

Not enough people have seen the Terminator films, I think.

As long as my job relies on me having no more and no fewer than ten fingers and maybe thirty-one teeth, I think the metal ones will have a tough time replacing me.

Otherwise, I can get a job identifying stop signs.

I was going to post this Tom Scott video in another thread, but I think it is time for a new non-application focused thread.

The video is a tad click-bait-y but… the concerns are valid. Not in the “Robots will replace us!” but more of, the world of technology is going to change rapidly, and depending on where on the curve of advancements we are, things could look a lot different in a decade.

It is both scary and exciting. He was able to get ChatGPT to write computer code for him in a way that was at least as competent as a quasi-amateur. It was actually really neat when he brought up looking over the code, thinking, those were mistakes I probably could have made, reading the documentation.

It also opens up a world of possibilities for people outside of most professional spheres to utilize AI to assist them. His barber talking about ChatGPT helping him draft a form letter is a great example of how AI could assist people in a good way. Rather than wasting time googling templates or requirements for a letter, it can save time and assist people not “in the know” on how to work on certain specialized sectors of the professional world.

Really fascinating, and a bit of a wake up call that we probably need to be prepared for this huge tech change in the near future. It could be a massive shift, or it could be Siri actually working as advertised.

Yeah, this for stuff like ChatGPT. I think a lot of artists are already feeling “Robots will replace us!” with AI art.

I can’t say much, other than we are experimenting with a ChatGPt type tool roll out that would enable natural language queries for data from a data lake rather than going through systems like Tableau or direct SQL queries on multiple DBs. There also appears to be a notable performance improvement over old tools, as this would be using the AI to create structured SQL queries for direct searches rather than the middleware used before.

It… works pretty well from the demo.

I’ve also seen deploys that use ChatGPt to write, and fix, Python code similar to what Tom did. Its impressive, within limitations.

Not particularly related to jobs, but this is the best thing I’ve seen about AI recently:

So, any area in which a “blurry JPEG” of the thing you want is acceptable is in danger. Everything else really isn’t.

Just like AI art, I feel like the threshold of “good enough” is going to be the fulcrum on how business handles this. I think there’s a point that a “blurry JPEG” in a lot of office/white collar work tasks is going to overcome the cost of employing workers (and their managers and HR to support that headcount) to create marginally better output.

The company I work for just released a presentation AI: create a presentation in under a minute (with or without wordcloud or interactivity -our current business model). We’ve been working on it since last July, but all the ChatGPT brings it more in the spotlight.

Or copy the text from a document/email from which you want a presentation made. Next step will be uploading a document from which a presentation is made. Should save you quite a lot time preparing presentations.

To me this can be the power of AI, to help you be more efficient with boring stuff or things you are not good at/know nothing about. You can use a site like Anyword to create blogs. Used to take up to a day to create a nice blog with unique content, with Anyword it takes 30 minutes. Or use Midjourney to create images from tekst; with it I created a new logo for my wife’s landscaping business.

This is a positive AI post, but I see the dangers and downside as well om which I won’t elaborate.

For those of you who want to try the presentation AI (it’s still free, we are exploring business models) this is the link:

I think the impact of this in IT is when you can hand it a legacy system with lots of tech debt and text requirements and it recodes it to a modern standard implementation with a small amount of manual intervention. There are already businesses built on this using older AI not based on a large language model.

I think if you can make a presentation just by telling software what you want (i.e. “make this slide build bullet by bullet, then fly in the images one by one every .5 seconds”) that’ll be the bee’s knees.

What happens when half the documents used as sources by ChatGPT (or whatever AI system) were generated by ChatGPT in the first place? Feels like you would get some strange “drift” where ideas or phrases no human would ever think or use become ingrained because it doesn’t know better. Or maybe there’s a simple solution of it recognizing its own work and ignoring it. But once stuff like this is operating at scale, is that possible?

The Singularity occurs!

The big issue for using these tools right now, in academia at least, is probably citation. These AI bots are terrible at it, in general. But they are really cool tools to work with. As a college professor, I am not too concerned about an AI replacing me. One, I’m probably within a decade or less of retirement. Two, most of my job these days is personally interacting with my students, often around things totally divorced from the nominal subjects of the class. If I taught old-school (hah!) straight-up info dump classes, sure, an AI could do that, though probably not as entertainingly as someone with decades of presenting experience. So far though sitting down and chatting with young people trying to navigate the complexities of learning and living in the modern age is something at which us meat puppets are much better.

My wife though helped a friend of hers, an Episcopal priest, prepare a sermon using ChatGPT. It drafted a dry but serviceable skeleton of a sermon on Jesus’ teachings in Matthew.

I would like to know more!

Well there is proprietary stuff, both in the specific implementation as well as what data it accesses, but also I’m not on the team deploying it, so I’ve shared about all I can about the project aside from the business data and usage end of things :)

Aka I’m learning about this as we go!

Cross-posted to the AI art thread.

Heard from a friend about this NYT article about a tool being developed to stymie AI that are trying to analyze artists’ work to replicate it. (My friend’s partner is one of the artists profiled.)

The article’s behind the paywall, so I don’t know most of the content, but I did look up Glaze, the tool it’s talking about. Here’s the summary of how the tool works, from a research paper by its creators at the University of Chicago:

In this paper, we describe the design, implementation and evaluation of Glaze, a tool that enables artists to apply “style cloaks” to their art before sharing online. These cloaks apply barely perceptible perturbations to images, and when used as training data, mislead generative models that try to mimic a specific artist.

AI has a way to go before they’ll be ready for customer-facing positions.

Customer: uh it’s 2023, not 2022.
B(AI)ng: fuck you, idiot, it’s 2022.

I was CACKLING as I read that earlier today.

Some of these are parodies.