I don’t harbor the illusion that modern LLMs are AGI, but rather they are an emerging technology that is likely to have significant economic implications, even if just for providing the service of chat bots, transcription/note-taking, document formatting, coding, biodesign, targeted advertisements, and more. Insofar as there are economic benefits for having superior AI tools, the country with access to more data may have an advantage, all else equal, and this data deficits pose a public goods problem.
My personal experience, as a cell, molec, and genomics scientist, is how dangerous is the stupidity of AI when you ask questions on the frontier of research. It gives you answers that feel like old, cold, scrambled eggs. There is no discovery or good insight into what we don’t or need to know.
As someone who has been very concerned for over 20 years about the loss of imagination in education and our nation, AI has felt like a brick wall when approaching sea-change questions.
Absolutely. I'm not saying the solution is to not allow access to data, but rather we have to note and make deliberate decisions about this public goods problem at the heart of AI.
Dear Alex, I fear that you may be vastly overestimating the potential of GAI as it is currently conceived and configured and materially underestimating its potential for harm. And no, data scaling is not going to solve what is fundamentally a flawed paradigm. As Gary Marcus posted this morning: The problem with generative AI has always been that large language models associate patterns together without really understanding those patterns; it’s statistics without comprehension. This blind fascination with and pursuit of technology without comprehension is literally going to get us all killed. I invite you to read Gary Marcus: https://garymarcus.substack.com/p/new-ways-to-corrupt-llms?utm_source=post-email-title&publication_id=888615&post_id=181604168&utm_campaign=email-post-title&isFreemail=true&r=8oz6&triedRedirect=true&utm_medium=email
I don’t harbor the illusion that modern LLMs are AGI, but rather they are an emerging technology that is likely to have significant economic implications, even if just for providing the service of chat bots, transcription/note-taking, document formatting, coding, biodesign, targeted advertisements, and more. Insofar as there are economic benefits for having superior AI tools, the country with access to more data may have an advantage, all else equal, and this data deficits pose a public goods problem.
My personal experience, as a cell, molec, and genomics scientist, is how dangerous is the stupidity of AI when you ask questions on the frontier of research. It gives you answers that feel like old, cold, scrambled eggs. There is no discovery or good insight into what we don’t or need to know.
As someone who has been very concerned for over 20 years about the loss of imagination in education and our nation, AI has felt like a brick wall when approaching sea-change questions.
There are problems of limiting access to a public good like data as described here:
https://en.wikipedia.org/wiki/His_Master's_Voice_(novel)
Absolutely. I'm not saying the solution is to not allow access to data, but rather we have to note and make deliberate decisions about this public goods problem at the heart of AI.