Before answering a question on Bike Gremlin Forum, I wanted to double-check if my article on that topic can easily be found (and to link to it). That’s when I saw the cover image of this article, and decided to discuss it and warn readers (and viewers).
I will use the term “AI” when referring to LLM systems used at the time of writing.
The swindle
It is one thing to show AI answers when explicitly prompted. A whole new level is how Google (and other companies) puts AI answers that are not directly asked by the user (like doing a normal Google search), and not clearly (enough) and explicitly shown to be AI.
That is misleading to the user. At the cost of repeating myself, see the screenshot, where Google puts its AI answer on top, and uses a relatively small heading and style that blends in with the rest of the search results:
This is both misleading and dangerous
After a lot of testing online (with ChatGPT, DeepSeek, Gemini, and my self-hosted LM Studio), I’ve concluded that current “AI” (LLM in fact) systems are very confident – always.
As a great professor taught us (he’ll recognize himself if he reads this), it is better to say you don’t know something, than to give a wrong answer. Depending on the particular topic/field, posing as an expert and confidently giving wrong information can even be dangerous (e.g: “yes, connect that wire…”).
See the AI double down on being wrong after I clicked on the “Dive deeper in AI Mode” option:
For context:
My chain durability test article
It is also theft
I already wrote about: “Is AI theft?” You can read that article for details, but briefly:
I just had a chat with a large forum owner. AI bots scraping websites is one huge issue, causing server overload and increasing costs (my article about stopping Google and AI bot ruthless scraping). But that’s not all!
It is also theft of intellectual property (hard work). AI scraping and then (often wrongly) regurgitating info from my sites is not an equivalent of a human learning, improving their own knowledge and discipline, and sharing the knowledge further. No. It’s an equivalent of using a cheap photocopy machine and then sharing (or selling) such “books.”
I put in a lot of time and effort to share knowledge and experience. Now, I get increased bandwidth costs (from bot scraping) and very few real visitors and advert revenue to cover those costs – because Google, ChatGPT and other companies are all serving “their” (read: stolen from my and other similar websites) AI answers to people.
Thank to generous reader and viewer donations, I’m keeping my sites and forum running (forum is now even without adverts, hope to see the same for websites soon). But very few experts are that lucky.
You do the right thing, as an expert: share your knowledge for free. And you get practically punished for it!
Thanks Google!
See my article about how this is killing the open Internet.
Why am I writing this?
It is not just web-Google, I’ve also noticed that AI Youtube video description summaries are now starting to emerge – more prominently than the video descriptions I take the time to write and publish for each video I upload!
I want to protect the readers and viewers from this false information – give you a heads-up as the Americans say.
I also want to protect my own integrity – AI is putting words into my mouth, practically – words I never said (or, even worse, using words I said but without the right context)!
By the time you read this, search I took for an example in this article will probably be fixed – corporations are smart and cunning. But the problem will remain (and no, that is far from the first or the only time I saw AI be wrong or misleading).
When you read my articles or watch my videos, you can clearly see the source – you can see who made them (bikegremlin.com domain for sites or @BikeGremlinUS channel for videos). I encourage fact checking and corrections. For any mistakes you find in my articles or videos, please let me know (my forum’s contact form with bot/spam protection).
Just be weary of quotes putting words in my mouth – wherever you find them.
Last updated:
Originally published:
