Comment Re:As expected (Score 1) 23
There is a way to us DNN/LLM "AI"s correctly; use them like a search engine.
Ask a hyper specific question, and scrutinize the answer given thoroughly.
In the same way that crowdsourced intelligence made google a useful tool for search, and social media created a great pool of questions and answers for that search to run over, DNN's are just an extension of search.
They are a wonderful improvement in the areas of (1) parsing the query and (2) re-jiggering the resultant hits.
(1) They can decode the user's question more accurately, and get a more searchable rewording of what the user is really looking for. Previous incarnations of search really needed you to find a magic word that matched perfectly to get the hits, and when you were using common words it became near impossible. But large language models seem able to do that with a much higher accuracy rate, and dont get hung up without magic keywords or magic phrases.
(2) Instead of merely presenting a raw list of sources, the LLM's actually read the pages, and try to parse out the specific bits you are searching for and ignore the rest of the page. They can also, to a limited extant, specialize the answer to match the query, based on interpolation of the page content. Again this is something that was previously impossible, and saves human time.
I would say, with judicious use of a search-engine DNN/LLM, any programmer should expect perhaps a 1% to 2% productivity increase on average.
Any programmer who tries to ask it to write code or solve problems will likely eat the worm, and suffer a 20%-50% decline in real productivity. Hopefully, any programmer caught doing this would face some kind of disciplinary action.