I'd be curious to know how much work and progress is being done by the AI companies to address this.
Sure they're all LLMs, made for genuine seeming text and not accuracy, but in reality companies are pushing these tools as search engines and helpful sources of information, surely it's in their best interests to make the output as accurate as possible. I would be very surprised to learn that they are "pure" LLM with no additional processing to try and improve accuracy and (meaningful) relevance.
They're insanely short-sighted if they're not just as worried as everyone else is about the self-poisoning tendencies of widespread AI usage. Maybe they really just don't care 🤷♂️
The reason you constantly hear the CEOs of the AI companies in the news claiming that LLMs will replace all jobs in the next week is because they desperately need more funding,
Hasnt there been a notable drop in recruitment of graduate jobs over the past year?
You need someone knowledgeable and experienced to sense check chatGPT output, but you do with Graduates too.
and it will only get better (i believe)
Thats potentially a bit worrying for the entry level job market.
It fills a lot of holes in my programming skills. Im not a coder or a programmer, but i know enough to know what to ask ChatGPT, and ChatGPT knows enough to be able to turn my prompts into some serviceable Python or SQL.
Or more likely, i can mash together some code rapidly, without worrying too much about syntax and specifics, and throw it into ChatGPT and say "fix this".
its really good at that, in my experience. Its playing to its strengths though. There is no leeway in script syntax, so theres no conflicting opinions about how it should work knocking around on the internet to confuse it. (as long as it can filter out the posts on forums titled "my script doesnt work"
im not programming CERN, but it has certainly vastly reduced the liklihood we would employ a programmer to fill that skill gap.
tools as search engines and helpful sources of information, surely it's in their best interests to make the output as accurate as possible
I think we’ll see more companies leveraging the data they have to provide better quality answers and let you choose the appropriate tool for the job - for a cost. For example, if you want to know some background on a science topic you’ll probably get a better answer (compared to chatGPT) from Scopus AI, which is trained using abstracts from peer reviewed articles in the Scopus database (owned by Elsevier), without pollution from the detritus elsewhere on the internet. It will also give you the (real) citations it has based the responses on.
