Protecting Content from AI Bots?

[Flashman]

@brechtryckaert

With bots it’s never black or white. But 5000 shades of grey

You could adopt the Flashman approach to spam and assume the worst, unless proven otherwise.

[PeteSharp]

I think American’s still use fax machines. I think so they can contact their bank about their cheques.

A place I worked at 20 years ago had a fax machine. But it would convert the fax to a pdf and email it to the right person. . We had a lot of old school customers.

[brechtryckaert]

The problem with AI crawling all content is that it will inevitably end in a flattening curve.
Currently AI is generating responses (in case of ChatGPT 3.X) based on a controlled dataset with content crawled before 2022 (and thus before the major AI boom where everyone was using it).

The downside is that now AI is crawling blogposts that are written by AI for example. The thing is that things like ChatGPT are language models. They’re great at generating text, but not great at generating factually correct text.

Most of us here will no doubt fact check whatever comes out ChatGPT, but not everyone does. Thus incorrect content goes online. It then gets crawled again by OpenAI bot and added to the database that responds in ChatGPT 4. See where this is going?

The more content we spew out with such language models, the more inaccurate these models will inevitably become, hence a flattening curve.

At least, that’s my understanding of how these models and their bots work.

[Jerry]

@PeteSharp

I think American still uses fax machines.

Let’s not go that route. To apply for a Liquor license over here I need to have a fax installed in my restaurant (by law) . The last time I did sent a fax was last week. For the rest of the world this must be over decade ago.

[Flashman]

@brechtryckaert

The downside is that now AI is crawling blogposts that are written by AI for example. The thing is that things like ChatGPT are language models. They’re great at generating text, but not great at generating factually correct text.

I thought about this earlier and concluded that if AI keeps generating text based on other AI text, eventually there will be one answer to cover every question and it will be wrong.

[PeteSharp]

Interesting thought. Language models trained on language model content.

That will be like a photocopy of a photocopy of a photocopy.

[Jannis]

@PeteSharp

photocopy of a photocopy

An AI generated image based on AI generated images.

[brechtryckaert]

@PeteSharp

That will be like a photocopy of a photocopy of a photocopy.

That’s actually a great analogy.

[PeteSharp]

Could be worse. A fax of a fax of a fax.

[Jerry]

@brechtryckaert

The more content we spew out with such language models, the more inaccurate these models will inevitably become, hence a flattening curve.

Which means without changing their own algorithms they could make themselves fail. In my 2 cents CHATGPT is the Altavista of the early days.

[Flashman]

@brechtryckaert

That’s actually a great analogy.

Google will prize original human created content, so they can steal it.

[brechtryckaert]

@PeteSharp

Could be worse. A fax of a fax of a fax.

[PeteSharp]

AI or ML is not restricted to language models though.

Contextual awareness will need to be the next step.

[Jerry]

I’m off to bed.
If someone could fax me the outcome of this discussion later tonight would be awesome.

[Bill]

Jerry - LOL !

[PeteSharp]

I’m bravely contemplating getting up. It’s just turned 6am

Coffee is calling.

[brechtryckaert]

Save us a spot under your in return

[PeteSharp]

Have to go change the world one ASCII character at a time.

[brechtryckaert]

On AI this is an interesting watch by the way: https://youtu.be/okGlJ2Ki2tk?si=eTIbAhtGGs4zbYmO

[Jerry]

@brechtryckaert

Save us a spot under your in return

Sunloungers are eagerly waiting for you. are in the fridge and the best part, summer is starting down here