| 408 | | {{{#!comment |
| 409 | | }}} |
| | 408 | |
| | 409 | |
| | 410 | |
| | 411 | == Summary |
| | 412 | This was only a representative part of what I did, what I'm doing and what I will surely continue to do. NLP AI has reached the level to be a universal tool that can help in a broad variety of questions, from everyday tasks ("How should I serve orange/buttermilk ice cream?" - "On orange slices and with a piece of shortbread.") to complex data creation and conversion. From summarizing text to being a gaming companion. |
| | 413 | |
| | 414 | We are just observing a breakthough. A disruptive change that will change the everyday life, including and particularly work, and this time not only for workers in factories, but also for technical writers, software developers, agents in insurance companies, call center employees and so on. And soon. And radically. |
| | 415 | |
| | 416 | We need to learn a new way of communicating with machines. We are used to sending queries to databases and receiving precise responses that are as correct or incorrect as the data in the database, no creativity! |
| | 417 | |
| | 418 | Now we can explain a problem and receive a reply that creates the impression the machine has understood our intention and what's missing to reach our aim, and also seems to be able to tune into the knowledge and experience we already have with the topic. When I ask "Can you summarize how a fusion reactor works?" I will receive a different technology level in my response than when I have a question about a specific nuclear reaction. |
| | 419 | |
| | 420 | Just like with human beings, not everything that the AI says is correct, neither "facts" nor code example. The AI feeds on vast data from different sources, and not all of the input is trustworthy. "Facts" about software products can be based on independent tests or user experience, but also on vendors' websites with a rather optimistic view on the product. But even if the data were perfect, the inference is not, it's just as good as the feedback-based training (another complex field). |
| | 421 | |
| | 422 | All the chats I had with ChatGPT only contained data not touching intellectual property or personal data. So, for solving general, non-customer-specific problems, it's great. But what if I want to upload a customer's terminology, or worse, internal engineering data, over the API and it ends up in the hands of a competitor? Can the AI model be equipped with barriers to intercept such data flow? Or would we have to run an AI instance per tenant? And how will legal requirements evolve to cover these aspects? |
| | 423 | |
| | 424 | To me, this moment of technological progress seems to unite the idea: "Wow, we've really gone a long way!" with the contrary idea: "We're at the very beginning." |
| | 425 | |
| | 426 | I am very curious (with a certain share of concern) where we are in six months. In two years. In a decade... |
