←back to thread

Microsoft and OpenAI's close partnership shows signs of fraying

(www.nytimes.com)
321 points jhunter1016 | 2 comments | 18 Oct 24 11:11 UTC | HN request time: 0s | source
Show context
mark_l_watson ◴[19 Oct 24 12:57 UTC] No.41887513[source]▶
Fun watching this all unfold.

I am starting to suspect that LLMs, short term for a few years, will end up mostly having value as assistants to experts in their fields who know how to prompt and evaluate output.

I see an ocean of startups doing things like ‘AI accounting systems’ that scare me a little. I just don’t feel good having LLM based systems making unsupervised important decisions.

I do love designing and writing software with LLMs, but that is supervised activity that save me a ton of time. I also enjoy doing fun things with advanced voice mode for ChatGPT, like practicing speaking in French - again, a supervised activity.

re: ownership of IP: I find the idea hilarious that by OpenAI declaring ‘AGI achieved!’ that Microsoft might get cut out of IP rights it has supposedly paid for.

replies(2): >>41887928 #>>41888192 #
1. sotix ◴[19 Oct 24 14:25 UTC] No.41887928[source]▶
> I see an ocean of startups doing things like ‘AI accounting systems’ that scare me a little.

I’m a CPA and software engineer currently interviewing around for dev positions, and most of the people I’ve encountered running these companies are neither CPAs nor accountants and have little domain knowledge. It’s a scary combination with an LLM that demands professional skepticism for every word it says. I wouldn’t trust those companies in their current state.

replies(1): >>41890168 #
2. lolinder ◴[19 Oct 24 19:40 UTC] No.41890168[source]▶
This brings up an interesting point, which is that it's likely that software developers overestimate the capabilities of LLMs in other domains because our own domain is so thoroughly documented on the internet.

From what I've been able to gather interacting with other sectors, it seems like software is pretty unique in having a culture of sharing everything—tools, documentation, best practices, tutorials, blogs—online for free. Most professions can't be picked up by someone learning on their own from nothing but the internet in the way that software can.

I strongly suspect that the result will be that LLMs (being trained on the internet) do substantially better on software related tasks than they do in other domains, but software developers may be largely blind to that difference since they're not experts in the other domain.