As a very late baby boomer (just a step from belonging to Generation X) I do not have an effortless relationship with computers and with novel technologies in general. Nevertheless, usually, I manage to get by somehow. If not by understanding the details then at least by grasping the broader picture and general concepts.
Watching Reid Hoffman’s lecture on artificial intelligence at Stanford University proved to be a different story. In order not to get lost among the unknown terms and abbreviations I had to look up the meaning of blitzscaling (what you can do when you have to grow really really quickly), scraping (the process of automatically extracting data from online sources and structuring it in a useful manner), SLA (service level agreement — still don’t fully understand what that is) and five 9s (denoting 99.999%), to name just a few.
With his PayPal, LinkedIn and venture capital experience Reid is a truly well-known and respected person in Silicon Valley. Here are my 3 takeaways from the bits I understood after watching the video of his lecture at Stanford University earlier this year on YouTube:
Lesson #1
While we are talking about the super-intelligence that will be coming, it has already arrived. For instance, GPT-4 in some dimensions is already more intelligent than humans. The cognitive capabilities are there. What is still missing is context awareness. This is the reason why models are still struggling with answering questions like — Are we on track? Do we need to pivot?
Lesson #2
The future use of AI agents will pose a radically new set of questions. For, example, can I take my agent with me when changing jobs? Should it be treated similarly to keeping the address book on my iPhone today? Or, to rephrase the question more broadly, — what is my agent allowed to know about what I am doing in the work environment? For the time being there is no good answer to this.
The business model will likely be based on humans paying themselves for the services of their agents. This makes sense — the agent maximises what we want, and we are the ones to pay for it accordingly. The advertising business model could just lead to a misalignment of interests here.
Lesson #3
There is little doubt that society would like to have a free AI doctor on every mobile phone. And, as Reid said, technically that can already be done with the current inference. The outstanding issues that remain to be resolved are regulation and liability for the advice given. This puts the ball squarely in the government’s court on this.
Concerns about the data privacy issues will be resolved with robust techniques that will allow to make synthetic data without revealing any customer-sensitive information. This will become a non-issue quite soon.
As a final thought this time, it is likely to take many years, if not decades, during which humans and artificial intelligence will have to cooperate and work together. We may also have to learn new ways of using the AI. For example, it could serve as a meta-skill — operating software that might take hundreds of hours for humans to learn and generate the desired output.
For those willing to try and select top lessons of their own here is the link.
This note was first published on Medium.com on 22 September 2024.
