33 comments

[ 2.7 ms ] story [ 79.3 ms ] thread
(comment deleted)
Oh god. I’m already imagining the hell that’s ChatGPT with “sponsored ad break” responses.
I imagine it will go through a period of reasonably subtle product placement text before switching over to outright ads mid-sentence.
As a large language model, I can't he-

Hi, it’s Clippy with ShamWow! You’ll be saying, “Wow,” every time you use this towel. It’s like a shammy. It’s like a towel. It’s like a sponge. A regular towel doesn’t work wet. This works wet or dry.

This is for the house, the car, the boat, the RV. ShamWow! holds 12 times its weight in liquid. Look at this. It just does the work.

Why would you want to work twice as hard? It doesn’t drip. It doesn’t make a mess. Wring it out. You wash it in the washing machine.

Made in Germany—you know the Germans always make good stuff!

-lp with that"

Hopefully we’ll be able to ask ChatGPT to spell it “chamois” in future communications. There needs to be a line drawn somewhere.
Adverts and misspellings. A double whamois.
Technically “shammy” is an alternate spelling. It just happens to be one up with which I shall not put.

“Whamois” is good though.

Yes, I don't have the exact design in mind, but I think it is very likely that LLM could bring a new era of advertisement. AI who knows the product that is being advertized and the user and their current state of mind really well to have order of magnitudes of probability of action that the advertiser want. AI could also maintain better ad quality and provide better experience to user, as now people are finally getting fully desensitized to all the scammy ads with full screen banner.

This could really be the next big thing from AI which I believe is solvable using current generation of LLM. And this will be the next trillion dollar company.

A good AI that can filter out the ads and product placement from the stream coming from a such a compromised AI is the next step. Let the AI phase of the advertising wars begin!
I'd rather just cancel my internet subscription.
That would be a very effective way of ensuring that I don't use LLMs for anything ever.
Aren't they legally required to clearly differentiate advertising text from other text? At least in the EU.
If ChatGPT just feels like casually mentioning how amazing the latest Fiat Panda is, and how great it's priced, and how cool you would look when driving it, and how the current model happens to come in your favorite shade of puce, all whilst having a nice little chat with you about parsing JSON with Python, is that really problematic?
Yeah it's only nice because they haven't enshittified yet. I would give it 2 years
Why on earth would you use an ad ridden AI? Just use your own AI and have it filter out all the nonsense.

Smart ad blockers will soon be able to find and eliminate stuff like sponsored content automatically.

Glasses that automatically blank or replace billboards in real life might actually convince me AR is more than just a gimmick.

Why imagine? Google's founders themselves wrote in their thesis that an ad business would compromise the quality of search results over time.
No I'm pretty sure the magic is having a chat bot that actually works.
I agree 100%! This tweet sounds like someone with a narrative trying to leverage a hot topic for their popularity.
Not to mention that if they think they’re getting “high-quality internet content”, it means they haven’t yet understood the limitations involved. Not only “hallucinations”, but also parroting of false claims. Which I suppose is “internet content” but when stripped of sources, it becomes more difficult to assess its credibility.
"it feels like we definitely had the technology to do this a LONG time ago!*

What...

Yeah, we literally didn't have the _technology_ of all things. We did have research and machine learning is not a new field. We had a good hunch on where to go next for some time. But AI is infamously hardware constrained and we have only recently got the technology to do GPT-3+ class AI at all.
This post could've been a tweet.

There's not much content here.

I’m confused, isn’t this post a tweet?
I thought the GP comment was intended as a joke about generative transformers.
It was a joke about making an HN post out of a tweet.

A tweet is just a short comment. That it happens to be on Twitter doesn't give it any more significance.

I remember the huge controversy a couple of years back when Google started showing these summaries in the search results - the covenant between Google and people who produced the content used to be that people wrote content and Google directed eyeballs to said content, in exchange it got to show ads on top of it, but Google took the content and stopped the traffic from getting there, people were understandably upset.

ChatGPT is more of this.

Google's so-called "zero-click searches" makes garbage websites like celebritynetworth.com and Forbes.com, the ones that write content for search engines, rush straight to the top of 'organic' results.
the anti-AI hype is at an all time high. This tweet is barely a thought experiment, provides zero data, but makes it to the frontpage

the absolute fixation some people have over the use of the term AI - isn't the measurable output of this technology more productive to discuss?

I wonder how hard would it be to just write off the Internet and go full offline. Setup a local LLAMa instance, download Wikipedia, torrent every movie and tv show from the last 40 years, clone all the github projects. It might sound crazy, but I remember people archiving entire discographies of their favorite artists on a single DVD. How much can you fit on single, modern HDD?
Current hard drives are in the 12-20 TB range. That's a lot of movies.