Raw LLM Responses

OpenAI is now looking into the usage of genAI for adult content behind a paywall. For a company that set out to revolutionise the workplace and to heal cancer, this is a very bizarre strategic direction to take. To me it just shows that there is no serious money in their product. I remember the first chatbot they came out with and how bad it was. At the time I thought what an amusing toy. Since then I tried to let it analyse a simple dataset on its own, a task it failed spectacularly on, neither analyses nor the interpretation of results made ANY sense. Let it write reports, again a task it failed spectacularly on, the report had logical flaws, was full of unsupported claims, was highly repetitive and very superficial. I tried vibe coding an entire project. All I can say, never again. The mess that the LLMs created was out of this world and took longer to clean up than coding it myself. Let it solve math problems, where it created the most outrageous errors. GenAI cannot THINK, it has very limited computational capabilities, it has no meta cognition which means it does not understand what it doesn’t understand. And this latter part is the most problematic, because it lets it spew out complete nonsense with a high confidence. Which means that for users who are no subject matter experts, they cannot rely on the answers. But maybe the most fundamental part about AI that many people seem unable to grasp: it can only work within the framework it was trained on. What does this mean? Any AI can only navigate within the world it knows, meaning the moment there is no precedent in the training data, it is lost. In a crude way, AI is like a parrot who can repeat what it was taught, doesn’t really understand it, and will therefore never be able to create something new. I’m so sick of these AI people who claim it will solve climate change, cancer, or other problems. It is fundamentally and structurally incapable of doing this, as it would require a reasoning capacity that is impossible to build with the current approaches.