My Joseph Kennedy moment with ChatGPT
Joseph Kennedy famously escaped the stock market crash of 1929 after his shoeshine tried to give him stock tips, leading him to sell all of his stocks and preserving his wealth.
Probably the only thing I have in common with Joseph Kennedy is that someone who has little understanding of machine learning tried to sell me on the revolutionary benefits of ChatGPT. No hate because the first time you see the magic trick it's amazing. But upon closer inspection and repeated trials you start to see cracks in the façade.
I've experimented with machine learning and narrow AI for over four years. While I've seen the miraculous cheat-code style shortcuts that AI can offer, I've also experienced the bizarre hallucinations and high-confidence wrong answers first hand. I would trust AI with low-risk repetitive tasks, but never life, liberty or pursuit of market beating returns.
Alleged actual human, but also why high-frequency trading algos demands a circuit breaker
Opportunity
Artificial general intelligence (AGI) will be huge when it arrives. For the time being large language models can only ingest tons of data and regurgitate it. They can also mimic human interactions and even emotions, but this isn't SkyNet. No AI model can interact with the physical world beyond the robots or sensors we've built. To put it another way: AI can't plant corn, but it can drive a tractor.
Real Impact
Some people will lose their jobs, some won't, and others will hate their job more. First companies that specialize in code-for-hire like Accenture, Cognizant and TCS will figure out how to displace human labor with AI (per Chamath Palihapitiya). He also thinks that AI will replace workers in judgement based systems where human error can be eliminated by machines.
On the flip-side AI will enhance engineering work in the short term. The productivity gains from AI assistance can immediately make everyone 10x developers. But that doesn't mean we'll employee fewer developers. In fact, there is no limit on the amount of software that can be written and the limiting factor on any project is engineering bandwidth. There is never enough engineering bandwidth and it's the lifeblood of the company (per David Sacks)
Should We Stop?
Yes, we should, but we won't. I believe we should continue to invest in the development of artificial general intelligence (AGI). I do not believe we should continue hyping ChatGPT parlor tricks and pitching ideas that rub AI on every business plan. Much like blockchain and cloud, machine learning has it's utility but is a feature not a product.
History suggests large economic effects will take longer than many people seem to expect
Oh lord, I think I agree with Paul Krugman 😬
Take Away
Inevitably this craze will peak and subside like every new technology. I shudder to think that the AI obsession is simply the start of another crypto style bubble but it likely is. Long term I think AI has a lot of benefits for humanity, but in the short term I think it's grift city. Some people will make money, but not you. If you can use AI to automate or assist in your daily life please do so, but don't sink your money into any idea pitched by snake oil sellers because it's the next big tech bubble.