This is a great piece of persuasive writing. (I was persuaded)
https://www.platformer.news/ai-skeptics-gary-marcus-curve-conference/
https://www.platformer.news/ai-skeptics-gary-marcus-curve-conference/
Comments
I also think the section on the AI hype cycle is misleading because counter examples along the lines of "AI can't do task X" aren't
I deconstruct it here at length, with many quotes from others who saw the same flaws that you two missed:
https://open.substack.com/pub/garymarcus/p/hard-forked-casey-newtons-distorted?r=8tdk6&utm_campaign=post&utm_medium=web
If it was
all coders would lose their jobs
If Casey were to have me on Hard Fork as some have suggested, that would be a good topic for discussion.
I think he dislikes people who deny AI risk but I am not one of them. it was … weird.
What can you tell us about the future of the energy requirements for all this?
Becomes intelligent
It will commit suicide
It's a neural ponzi scheme clad in shiny technology.
AI is in a hype cycle rn. Lots of companies investing in or using AI are doing so ‘because it’s the future’ and don’t have compelling use cases. The definition of frothy.
I fall more on Marcus’ side but agree it’s dangerous given many take it’s output at face value/true
AI is great for lots of things the disconnect is they are not being used this way
warned us vis-a-vis Geoffrey Winton. Yet, here we are, full steam ahead!
*Gemini*: Tell me how I will make a living now that you replaced me?
https://www.nbcnews.com/tech/tech-news/artificial-intelligence-pioneer-leaves-google-warns-technologys-future-rcna82242
https://ageoftransformation.org/aidoommyth/
https://thetechbubble.substack.com/p/the-phony-comforts-of-useful-idiots?triedRedirect=true
It’s how we use tools.
There are always going to be people who use tools for evil purposes
The skeptics note that these models are pattern matching, and that this has exponential costs for linear improvements. GPT-1 to gpt-4 was around 1000x increase in training costs. Who will pay the $1 trillion to get a similar jump?
Truth is, it’s basically a better search engine. Helpful, definitely, disruptive, definitely, but not close to real ai.
Skeptics would say maybe 10%, ai bros say we are 90% of the way there. But if so, why can’t we see a clear path from here to there? People were building rockets for decades before Apollo, why can’t we?