Comment Re:This is happening (Score 1) 31
The problem is largely the generality. You're right you can't replace a workforce with ML/AI. Management however doesn't understand that AI is a tool, and a tool is used for a specific purpose. AI is not general enough to simply replace an employee, it needs to be used by someone to improve efficiency. The downside is no one discusses this for general LLMs.
100% agree with you. And I think the companies that figure out that you could take the same number of employees but have them using better tools are going to be the winners in the long wrong. Using AI to be more productive, shortening development cycles, etc is kind of the point, but real metrics and cost-benefit analysis needs to be done continually in any company adopting new technology. If it's not paying off, then don't do it. And I firmly believe that cutting large number of staff is the wrong optimization, being smaller is the opposite of growth. And in this industry, growth is success (in multiple dimensions)
The features are of course slowly developing, but the biggest problem is that it gets thrown at staff without a use case and without training on what to do with it. I shit you not someone in our training session suggest we use CoPilot to start software by hitting WIN+C and typing the name of the software we want AI to launch. Try it, it's so frigging slow that you can probably locate the exe file manually on your computer faster than that (to say nothing of the fact the start menu has a search feature).
Yea, there is a huge lack of understanding of what it's for. It's not a search engine or a program launcher. But on the other hand you can ask it (Glean, copilot, whatever) something like "Look for internal docs on XYZ and summarize. Note any discrepancies between sources. And cite your sources."
I find LLMs useful to make a workflow for doing some task. Like instead of asking it to generate tables, I have it generate a script that generates tables. Because I have on MANY occasions seen Claude Opus and others spit out tables of data with subtle errors in it. And while I usually have success when I ask it to review the tables it just generated, it has been much better at reviewing a small script's correctness than large tables of data.