AI has been the big IT trend of the last few years. Nvidia is happy about it; the environment, not so much. Personally, I haven’t been very convinced by AI until now. After testing ChatGPT and Perplexity.ai, its usefulness for me was limited to programming small Python projects. For fact-based topics, I found the results too vague. Having AI write texts was a total failure. You can really tell where the text comes from.

Thanks to Deutsche Telekom, I got a free one-year subscription to Perplexity.ai. Good thing, too, as I’ve barely used it. You have to go to the website, start a new conversation, and then type in your query. Kagi, on the other hand, I use as my default search engine. I like its focus on privacy and an unfiltered internet. But Kagi isn’t just a search engine; it also offers access to various AIs. With the most expensive subscription, you can have conversations with different AIs. But there’s also a simpler option: Quick Answers.

I stumbled upon Quick Answer rather accidentally. I was searching for something and ended my search query with a question mark. To my surprise, I found the Quick Answer block above the search results. In that block was a short, AI-generated answer to my question. Since then, I’ve been using this feature all the time. A question mark at the end of the search query, and Kagi displays a Quick Answer. I then skim it and check if the answer is sufficient for my search. If not, I scroll down as usual and look at the search results.

But scrolling has become frighteningly rare. Most queries are adequately answered by the Quick Answer. Well over half of my searches are quickly answered this way, especially in the technical field. However, the AI falters with more open-ended questions. In the image above, you can see that even Kagi can’t avoid hallucinating AIs. I’m not quite sure which loyal readership and community the artificial intelligence is referring to. You hardly ever write comments, and feedback on other channels is rather sparse. The visitor numbers don’t exactly paint the picture of a high-reach blog either.

Large Language Model (LLM): In computer science, LLM refers to a large language model based on deep learning techniques, often used in the field of Natural Language Processing (NLP). These models are capable of recognizing, generating, and processing text by being trained on extensive datasets.

Response from Kagi Quick Answer to the question: What is an LLM? But back to AI. I’m actually critical of the current state of AI, especially Large Language Models. The aforementioned hallucinating is a major problem in my view. Even before AI, people were barely able to use a search engine and evaluate content. The AI’s answers, however, read fluently and competently, which quickly gives the impression that the answer must be correct. But with AIs, the rule is that they can only reproduce what they have been taught beforehand. Since the content of the internet serves as learning material, it’s not surprising that there’s junk in there too. Additionally, AIs always try to answer a query. Even if there’s no perfect match, the AI will generate an answer from the most probable options. Regardless of whether it’s right or wrong, an AI doesn’t know this distinction.

I’m a little unsettled. I didn’t think AI would creep into my life so quickly, and almost without me noticing. But I also didn’t expect it to offer me so much value. In the end, however, the Quick Answer also means I visit websites much less often. This saves me many visits to Reddit, Chip, and the like, but it also prevents the discovery of smaller, good websites. Why should you still come here when you get the answer to your search query right on the search page? Why read my long texts when their essence is presented in short form by your search engine - or you can create a short version with the Kagi Summarizer?

While writing this post, the thought occurred to me whether it even makes sense to write such long posts anymore, or any at all. Just feed bullet points into the AI and post a readable summary on the blog, done. But that puts us in a vicious cycle. At some point, the AI will only learn from AI-generated content and degenerate as a result. That can’t be the goal either. So I’ll keep writing here, in the hope of continuing to support the quality of the internet and, with it, that of AIs.