Growing ChatGPT use on the site.

[wholybee]

Originally Posted by Iron E

Help me out here. What exactly would they use AI for in this regard?

If you're saying, to write the code and fire a few developers, then yeah, I can see that. But where exactly does AI help, say Zillow, out?

I made a few tools this week. One is just a form on a webpage built with HTML, JS, and CSS and returns answers based on the fields selected. In this case a color wheel.
Another is a stylized form on a webpage connected to an Excel sheet with an API. This gives long legal answers to selected questions.

The first was like 95% coded by AI. The second was built by me with some of my technical questions answered by AI. All of the data was entered by me and the sheets were built by me.

I would be interested in knowing where I could cut corners.

It wouldn't be an LLM, but would be a deep learning model. To make the comparisons to similar sold properties better. Instead of just looking at price per square ft., square footage, and that the property is nearby, an AI can estimate how much value an extra half bath would add. Or be able to determine that a comparable in a nearby neighborhood will bring 12% less based on the neighborhood. Just estimating the price by pure statistics can't adjust the estimated price by more difficult to category differences. With AI you don't have to code for the algorithm to look at every factor and factor them all in. You just feed the AI everything you know about every property and it figures out what they all mean.

Edit:
When I was studying deep learning and how to write an AI for such tasks, a price estimator for Real Estate was one of the examples in the textbook. So I consider that must be a pretty standard use.

[Iron E]

Originally Posted by wholybee

an AI can estimate how much value an extra half bath would add. Or be able to determine that a comparable in a nearby neighborhood will bring 12% less based on the neighborhood. Just estimating the price by pure statistics can't adjust the estimated price by more difficult to category differences. With AI you don't have to code for the algorithm to look at every factor and factor them all in. You just feed the AI everything you know about every property and it figures out what they all mean.

Edit:
When I was studying deep learning and how to write an AI for such tasks, a price estimator for Real Estate was one of the examples in the textbook. So I consider that must be a pretty standard use.

I'm going to have to look further into this. Sounds interesting.

I do wonder how fast these site tools will become obsolete. Google and Bing are the biggest scrapers on the internet and the answers may soon be coming directly from them.

[wholybee]

Originally Posted by Iron E

I'm going to have to look further into this. Sounds interesting.

I do wonder how fast these site tools will become obsolete. Google and Bing are the biggest scrapers on the internet and the answers may soon be coming directly from them.

In the case of Zillow, they won't become obsolete anytime soon. Rather, Real Estate Agents are becoming obsolete. It isn't just searching for homes, but Zillow provides additional services that require a real estate license and they are licensed in most states. The family business is Real Estate, although I didn't pursue it. My mom retired and closed her office a year ago. I use Zillow to manage my rental property.

But for sure, many specialized sites will become obsolete, or at least you will search on Google, and be linked to the site. Example, airline tickets. You search on Google, and Google will show the prices offered by Travelocity, Expedia, etc.

[CaptTom]

Originally Posted by Wotname

To be even more clear .

The post numbering is dynamic and when a post gets deleted, the numbers reorder. The guy Iron E quoted is no longer with us. The insertion of spam links in quotes is sneaky and has been around awhile. Mostly we pick them up only when you guys report seeing your quote has been altered. These reports are very much appreciated.

EDIT: to be clear, a member alerted us!!!

Reports of my demise have been greatly exaggerated.

I'm the one the spammer quoted, inserting their spam link in the middle of my otherwise brilliant (if I may say so myself) post.

I didn't think about the posts getting re-ordered when the offending one was deleted. And, yes, it was probably I who first reported it. The mods got to it so quickly most of you probably couldn't figure out what I was talking about. Sorry for any confusion.

Back to AI though, it's very probably that spammers use AI tools to create a brief, on-topic but pointless post they can use to reply to a thread.

There has been a sharp increase in this kind of spam. A first-day post, from a new user, vaguely on topic but adding nothing to the conversation, using good grammar and spelling. Often these are responding to a long-dormant thread, but not always, as in this case.

[wholybee]

I'm not sure if this old software allows it, but disabling links within the quote tag would stop that particular abuse. It would be a slight inconvenience to a user to go back to the original post, but probably not a big deal to anyone following the thread.

Requiring captcha and/or post approval for the first 10 posts would also help a lot with the spam from new users and bots. But i expect that would greatly add to the workload of our dedicated moderators.

I'm not sure any popular forum software supports it, but what i have seem a few times on other types of sites is that suspicious behavior (as determined by an AI) would trigger a re verification or captcha. So a user would still be allowed the post, but on submitting it would need to jump through some obstacle a bot wouldn't be able to complete.

Maybe such features will show up in forum software soon.

[GordMay]

Originally Posted by wholybee

I'm not sure if this old software allows it ...
... Maybe such features will show up in forum software soon.

vBulletin 3, which the CruisersForum is still using, was released in 2004.
vBulletin 4 was released in late 2009, and vBulletin 5 was released in 2012.
In late 2017, it was confirmed that vBulletin 3 and vBulletin 4 were now considered to be "End of Life" software, with no further development of the series planned.
vBulletin 6.0.0 was launched, on Tuesday Aug. 22, 2023 at a retail cost of $179.

[GordMay]

As LLMs feed off each other's content, the quality gets worse, more similar, and more vague, like a photocopy, of a photocopy, of an original image.

The first version, of ChatGPT, was the last model, to be trained on entirely human-generated content.
Every model, since then, contains training data, that has AI-generated content, which is difficult to verify, or even track.
And, with every generation, this content, generated by LLMs, trained on AI-generated content, quickly becomes more homogenous, and totally unreliable [garbage] data, as errors, and homogeneity, replicate.
When this happens, we lose quality, and precision of the content, and we lose diversity.

Everything starts looking like the same, unreliable, garbage.

In a 2023 study [1], from Johannes Gutenberg University, researchers found that "this self-consuming training loop initially improves both quality and diversity," which lines up with what's likely to happen next. "However, after a few generations the output inevitably degenerates in diversity. We find that the rate of degeneration depends on the proportion of real and generated data."

Another academic paper [2], published in 2023 came to the same conclusion, about the degradation of AI models, when trained on synthetic [AKA: AI-generated] data.

Or, maybe not.
Perhaps, my own lack of imagination, blinds me to the possibility of AI’s potential, to develop ‘creativity’, matching [or surpassing] humanity’s [the ‘singularity’].


[1] “Large Language Models Suffer From Their Own Output: An Analysis of the Self-Consuming Training Loop” ~ by Martin Briesch, Dominik Sobania, & Franz Rothlauf
➥ https://arxiv.org/pdf/2311.16822.pdf

[2] “Self-Consuming Generative Models Go MAD” ~ by Sina Alemohammad et al
➥ https://arxiv.org/pdf/2307.01850.pdf

[GordMay]

How To Fool an Eavesdropping AI … With Another AI

Scientists, at Columbia University, in New York City, think they’ve devised an AI, that can effectively fool an eavesdropping automatic speech recognition system, from transcribing your private conversation.
So, in the future, you may not have to worry that someone is using spyware, to record your phone calls, or that your Alexa is listening in, when it shouldn’t be.

Their Neural Voice Camouflage system [1] prevents eavesdroppers from secretly transcribing your audio conversation, by piggybacking a custom static-type noise over your speech.
The noise is set to the same volume as normal background noise [no louder than a regular background air conditioning unit], so people you’re talking to can still easily make out what you’re saying.
However, the automatic speech recognition system (ASR), that’s attempting to eavesdrop, will get confused, and produce a Gobbledygook transcription.

More, on Audio [4:13] ➥ https://pocket-listen-scout-streamin...7e29f1f366.mp3

[1] Real-Time Neural Voice Camouflage
➥ https://voicecamo.cs.columbia.edu/
➥ https://arxiv.org/abs/2112.07076

[JoeRobertJr]

Originally Posted by GordMay

As LLMs feed off each other's content, the quality gets worse, more similar, and more vague, like a photocopy, of a photocopy, of an original image.

The first version, of ChatGPT, was the last model, to be trained on entirely human-generated content.
Every model, since then, contains training data, that has AI-generated content, which is difficult to verify, or even track.
And, with every generation, this content, generated by LLMs, trained on AI-generated content, quickly becomes more homogenous, and totally unreliable [garbage] data, as errors, and homogeneity, replicate.
When this happens, we lose quality, and precision of the content, and we lose diversity.

Everything starts looking like the same, unreliable, garbage.

Somehow, this strikes me as exactly what people do - sitting around talking about something until one person's misconceptions become accepted by most just by repetition with baseless confidence.
"The family is the cradle of the world's misinformation." - Don DeLillo

Lately, as I read CF posts, I am thinking more and more about whether a given new user is a real person. There was just a first post on the Joke thread that had me suspicious.

[Pete7]

Originally Posted by JoeRobertJr

"The family is the cradle of the world's misinformation." - Don DeLillo.

And yet if generations hadn't past down stories we wouldn't have such rich history. For example the Norse Saga's and stories of Vikings landing in America.

Or how about this innocent little poem, which refers to Bubonic Plague in London 17c.

Ring a ring of roses,
a pocket full of poses,
a tissue, a tissue,
we all fall down.

[s/v Jedi]

History books belong in the fiction section. Almost none of recorded history is accurate. Every time someone investigates part of it, they find it’s inaccurate.

[CaptTom]

Originally Posted by GordMay

As LLMs feed off each other's content, the quality gets worse, more similar, and more vague, like a photocopy, of a photocopy, of an original image.

Interesting. I can definitely see that happening.

A good analogy is the algorithms social media sites use to keep you "engaged." They're not after accuracy. Or reality. Or really anything except to trigger an emotional response. And they're frighteningly good at it.

Try opening up YouTube in a browser where you've cleared all cookies and auto logons. Without knowing your history, the algorithm feeds you what it thinks would be most "engaging" for the average user. It's quite depressing. Is this the future?

[Wotname]

Sailors, CF does not do political talk; there are other social media sites available if you wish to indulge in political discourse.

5 posts have had to be deleted.

[GordMay]

The effects of AI, on Remote [live-aboard cruisers] Workers:

According to a 2023 report [A], from Goldman Sachs, the labor markets in [both] the United States and Europe could "face significant disruption" if generative AI lives up to the hype. As much as a quarter of current jobs could be fully replaced by AI, and two-thirds of all jobs — or 300 million jobs — will be impacted by AI automation in ways both large and small.

The article lists the most, and least threatened.
Here are the most threatened:
1. Content marketers *
2. Beat journalists
3. Graphic artists
4. Data Analysts *
5. Programmers and coders [*common jobs, for live-aboard remote workers]
6. Lawyers
7. Warehouse workers
8. Radiologists and Medical technicians

Software Developers [programmers & coders]:

Most experts agree that generative AI will be a boon for software developers. With AI tools like Copilot [1], programmers don't have to painstakingly code line after line from scratch. Just type in a prompt and the algorithm can generate high-quality code in multiple coding languages.

According to a paper from Microsoft and MIT [2], software developers who used Copilot were able to write a program 56 percent faster than traditional coders. In the tech industry, where mass layoffs have already culled the workforce, coders are rightfully concerned that entry-level programming jobs will be replaced by AI.

For the software developers who keep their jobs, many will take on new roles as "prompt engineers" [3], people with a knack for entering the best prompts into the automated coding apps.

[A] “The Potentially Large Effects of Artificial Intelligence on Economic Growth” ~ by Joseph Briggs & Devesh Kodnani et al, for Goldman Sachs
➥ https://www.key4biz.it/wp-content/up...gs_Kodnani.pdf

[1] “Copilot” ➥ https://github.com/features/copilot

[2] “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot” ~ by Sida Peng et al
➥ http://arxiv.org/pdf/2302.06590

[3] “Are 'prompt engineer' jobs real? Yes, but maybe not for long”
➥ https://mashable.com/article/what-ar...gineer-jobs-ai


BTW:
If bees start writing software, will they call it beware?

[Kettlewell]

But, has anyone yet found any AI-powered stuff really useful? I certainly haven't, though I have only dabbled in it. I know a software developer who said it is already useful for writing out tedious code for things when it would take the programmers longer to do the same. But, in my ordinary life I don't find any AI stuff particularly useful.