There are strong signals that continuing to scale up in data is not yielding the same reward (Moore's Law anyone?) and it's harder to get quality data to train on anyway.
Business Insider had a good article recently on the customer reception to Copilot (underwhelming: https://archive.fo/wzuA9). For all the reasons we are familiar with.
My view: LLMs are not getting us to AGI. Their fundamental issues (black box + hallucinations) won't be fixed until there are advances in technology, probably taking us in a different direction.
I think it's a good tool for stuff like generating calls into an unfamiliar API - a few lines of code that can be rigorously checked - and that is a real productivity enhancement. But more than that is thin ice indeed. It will be absolutely treacherous if used extensively for big projects.
Oddly, for free flow brainstorming like associations, I think it will be a more useful tool than for those tasks for which we are accustomed to using computers, required extreme precision and accuracy.
I was an engineer in an AI startup, later acquired.
Business Insider had a good article recently on the customer reception to Copilot (underwhelming: https://archive.fo/wzuA9). For all the reasons we are familiar with.
My view: LLMs are not getting us to AGI. Their fundamental issues (black box + hallucinations) won't be fixed until there are advances in technology, probably taking us in a different direction.
I think it's a good tool for stuff like generating calls into an unfamiliar API - a few lines of code that can be rigorously checked - and that is a real productivity enhancement. But more than that is thin ice indeed. It will be absolutely treacherous if used extensively for big projects.
Oddly, for free flow brainstorming like associations, I think it will be a more useful tool than for those tasks for which we are accustomed to using computers, required extreme precision and accuracy.
I was an engineer in an AI startup, later acquired.