These are the things that LLM-based AIs actually don’t do well at all. Take a look, for example, at your own example above. See the SWOT analysis, which is supposed to be about “Buying an Electric Car for Personal Use”. Do you see how it mistakenly made the exact same point in Weaknesses as it Threats - namely “Limited availability of charging stations in some areas”? That one is forgivable of course. But also in Threats is “3. Competition from other electric car manufacturers” which is absolute nonsense. Increased competition from manufacturers drives down prices, which is at least an Opportunity, and actually closer to an existing Strength. This is caused by the AI getting confused by training material on the benefits and disadvantages of EVs from the manufacturing or general automotive aspect, rather than the clearly asked for ‘personal use’ perspective.
It took me just seconds to find that error. I assure you there are many, many others.
LLM means Large Language Model. It is exactly what it says - it models Language, not logic. It predicts and generates the language it would expect to find if your prompt were a Q on a Q&A page, based on the millions of documents it was trained on. It is not a Large Logic Model. It is not an AI built for decision making. It just predicts the most statistically likely response, which may often coincide with something that was reasoned out, thought through, but actually wasn’t arrived at through applying logic to the problem at all, only predicting the patterns of language involved.
See A Crash Course in LLM-based AI for a more detailed explanation of how LLMs work, and why they may seem to be using logic and judgement, but actually not only don’t, but currently can’t do. There’s millions of dollars going into each of hundreds of bits of research right now into how to give an LLM-based AI the capacity to use a Knowledge Graph, or how to incorporate better logic, etc. I mention just a few of the hundreds of these in that thread.
Understand that this criticism is constructive in intent. I want you to understand the limitations so that you can create better, more reliable prompts in future. Ones that use the actual strengths of AI, and don’t misunderstand the weaknesses of the current tech.