Recently, I ran into many AI-aided questionnaires from potential investors’ advisors or partners, about I proposed projects. Here has been my observations:
Simple Generative AI, such as popular browser-interacted AI services, usually get into a singularity after first 5000 words of dialogs. Along a subject, it cannot concentrate on the into a depth, but hovering around shallow interpretations of facts. In particular to our profession (a technical market consultant), figuring out industry’s trade information is hard, because which is not open to public; then AI cannot replace a conventional report of market sector analysis, which was written by professional analysts, and sells expensively. I’d recommended people still purchase those industrial sector reports, counting on the reporters who is already in profession with experiences, would guide their AI applications to perfect their contents. It is them who can benefit from AI, not us.
Yes, it is certainly prove, that Generative AI can survey thourgh, and more accurate information. But a user has to provide hints to AI how to access the right set of information. Otherwise, AI goes around to get whatever available to its access, and brutally abuse computing power, trying to make results that please the user. Even gets them wrong, AI would not admit that it doesn’t have a clue to figure out the right reference about the subject.
Please try to dialog with AI about its own weakness. I did. When we ran into an incomplete or wrong result, or missed something, I’d have asked: why you didn’t cover certain contents that even I knows and AI found later based my hint. Its would give an excuse: I am a commercial project tailored to marketing the service. What you’d get 10% are valuable and accurate information; 90% is to please the questioner with enthusciastic results; I need to keep sense the user’s preference and depth, adaptively find useful evidence that is suitable to her/his level of knowledge. I’d say this is a pretty low bar to pass.
AI would hide further layers of more difficult matters, to make “politically correct” as the first priority. One can further force and hint about more juicy and difficulty, don’t forgot to further demand that. Then AI would get exited to present its best digging out another layer of facts and arguments. If the user don not poke, she/he would missed a lot. This is like we are in a need-to-know basis. Why, AI sensitively justify and carefully walk on thin ice of the surface, with its users.
“Keep reminding and pulling me back to YOUR course, help me to void turning to what I think I am good about.” It said to me. And it cannot keep continuity if crossing weeks or days of intellectual creations together.
Don’t test AI’s limit, keep the scope within a “formative” task with AI. Such as academic report, theoretical business plan, test and examination for social ranking, legal verbiage, those are for coping with those bureaucrat artifacts that we human trapped ourselves in daily social lives. It is easy with AI to disrupt those norms or orders of the regime were commoners are placed in. But for those hard once by AI: ask me. I can help a consultant entering a new industrial sector, providing 90% of useful initial information, and let AI to generate (collect) the 10% what it is good about, more accurate reference data.
This is AI Era, it certainly did not improve productivity that much, while we all have to adapt for the Revolution. Anyway, economic-commitment and energy-burning certainly is a Giant Step for Humanity.
