AI As General Intelligence: A Disappointment #
When OpenAI launched ChatGPT in 2022, a new era seemed to be dawning. Suddenly, almost all digital professions were under threat of being replaced by AI. Facebook had to quickly bury their Metaverse under great pressure and invest more in AI, and so did Google, which only had primitive AI in comparison. Bureaucrats were even invented to regulate AI, although it works like a black box and the bureaucrats will always be two or three steps behind the development.
Where AI actually made an impact #
In some areas, AI does indeed have the predicted effect. Images, videos and music can be generated extremely quickly. Where previously you might have had to pay someone on Upwork or Fiverr, you now often get a solid work of art very quickly, at least for the beginning or a concept. AI is also good at commenting on existing code and searching given text for information. In other words, anything that is restricted to a limited area of existing data and information.
AI hallucinations: the risk of embarrassing yourself as a clown #
Like everyone else, I created a ChatGPT account and asked various questions. I was surprised at how competently ChatGPT answered a grammar question about a minor language. When I showed the history to a linguist, however, he said it was complete nonsense.
I had suffered my first AI hallucination, an existential experience. If I hadn’t asked the linguist, I would still believe this AI lie and perhaps many others today. With such false knowledge, you risk embarrassing yourself when talking to people who are knowledgeable on the topic. As a result, you have to check that every AI answer is correct. But then you can just do your research and leave out ChatGPT anyway.
What’s the future of AI? #
One might have thought that AI was constantly evolving linearly or exponentially, but at the end of 2024, the news came1 that development was becoming slower and more complex. Recently it was said that AI actually only memorizes an enormous amount of rules2. AI in the sense of a general intelligence, which could replace the mental power of an average person seems a long way off. In addition, the Chinese company DeepSeek launched an AI model in early 2025 that is similarly competent to ChatGPT, but far less resource-hungry.
However, AI is now indispensable in specific, narrowly defined areas, especially when 100% precision is not required. Summarizing articles and books, for example, or translating. However, you have to carefully curate the data sets you feed the AI with in order to achieve quality. The idea that you can simply give an AI Reddit as a data set, for example, and then get a smart, human mind has proven to be a delusion.