Ask Lemmy

I might need to ask chatgpt to reply to this comment. In fact if i ever see your username around i will put it into chatgpt and ask it to generate a reply for me, because i can’t think for myself.

Go for it, it might be kinda funny. A bit of irony in it too, as it would offloading the critical thinking required to engage with an argument you disagree with.

i don’t believe i’ve ever asked calculator to answer biology question for me.

Nor should you, you would be using the wrong tool for the job. I'd also not use a calculator to drive screws, that doesn't invalidate the point. Tools are useful when we use them the right ways. And ya, AI is a terrible tool to offload critical thinking onto. There are use cases where AI makes sense though. Things like image classification and fuzzy searches on large data sets are good use cases for various AI models. One of the problems with AI, at the moment, is that it has been sold as some sort of cure-all that will replace humans and critical thinking. And it's absolutely not that. It's in much the same place as cocaine in the first part of the 20th century. Hucksters are putting "AI" on the label and claiming it will solve everything. The reality is much more nuanced. It has it's uses but they are far more limited than the hucksters are claiming.

Large language models really can be useful for fuzzy searches in large data sets. To give an example from my own work, Copilot is really good at searching Microsoft documentation for me. Could I find the answers with a regular search? Probably, it would also take me longer. Instead, I send Copilot chasing after the answer to that question and go do something else while it finds the answer. They can also help in re-writing for different audiences. I write a lot of technical reports and those need to be summarized for managerial audiences. Yes, I could do that manually, I've done it for years. I also hate doing it. Clippy is good enough at doing it that it can give me a first draft and I can finish it up in far less time and effort.

The biggest issue I have seen with LLMs is exactly what you point out, that people trust them too much and don't think critically about their answers. Again with my work, we use a product that uses an LLM to summarize cybersecurity issues and provides suggestions for response and investigation. It's a pretty well trained model and it's suggestions are pretty good most of the time. But, it falls down spectacularly bad from time to time and the analyst needs to be able to recognize that and respond to the alerts appropriately. Some analysts are better at this than others and this is now part of our training for new analysts. We teach them to use the LLM, but to also always think about the basics and question the LLM when it doesn't seem right.

Image classifiers are another area where I think AI has some good use cases. Consider the job of reviewing images and videos for sites like FaceBook, TikTok or YouTube. The folks who do this work are exposed to a lot of very violent and disturbing media. I used to work with a guy who did computer forensics in a law enforcement setting and he finally left that work because he could deal with having to review CSAM images any more. This seems like the perfect place to slot in an AI image classifier, to make a first pass at it. If it can correctly classify the vast majority of that sort of content, that greatly lessens the workload on the analysts who will need to deal with the borderline stuff and reports of false positives and negatives.

The new normal is people just follow what billionaire said, and you think it’s okay.

Not at all, but I also think the reactionary "fuck all AI" isn't okay either. It's a tool and it's going to change things. We need to navigate that with a clear head and careful consideration.