Digital technology has changed many things about our lives since personal computers became a thing when I was in high school. Convenience, Communication, easy access to information etc. It’s hard sometimes to remember how we did things like travel before there were smartphones. The pace can feel faster and faster.
In the last decade though, I seem to have become the person who is saying that things aren’t going to change as much or as fast as people think. I am the one pushing back at the person who is spinning a story about how new technology X is going to change everything in just a few years. This happened a lot at work. I think a lot of my former coworkers probably viewed me as the curmudgeon on the team.
Kids, this is not the best career move!
I do like tech and I believe in progress but most things happen slower than most people expect. They also usually deliver less than promised and sometimes a lot less. That’s not bad, but it makes the self styled visionary types especially annoying and problematic. We’ve all watched these people drive the allocation of money to stupid ideas that look good in a demo but aren’t going to actually benefit many users. This is true whether it’s VCs and founders overstating this week’s game changing idea or someone inside a big company who has discovered that repeating things that sound cool and futuristic even if they have low utility is a great way to get people to think you have a vision. I prefer building things people are going to use now and saving the speculation for happy hour.
Some examples of things I have been a skeptic on in the face of annoying amounts of hype:
Self Driving Cars – 10 years ago everyone kept telling me that these were coming fast and going to change everything quickly. No one wanted to think about how hard the problem really is or understand the difference between kind and wicked environments. Thinking about vehicle turnover rates, legal issues, or weather was just not any fun either. Most people now accept that self driving cars were way over promised and not only by Elon.
VR Headsets – Around the time Palmer Luckey floated on a magazine cover, I had to endure so many “visionary” people telling me how this was finally going to be big. People were going to give up their monitors in favor of headsets and we were going to have new amazing ways to collaborate etc. Having heard all this in the late 80s when I first worked on VR, I was skeptical and even though the devices are now really nice, there’s still too many obstacles to see the tech grow really fast. VR isn’t dead, but it’s not really going anywhere. Despite investing huge amounts of the money they make getting people to hate themselves and each other, Meta actually had lower Reality Labs revenue in 2022 than in 2021.
AR Glasses – When VR falters, the same hypesters move on to AR and explain how always-on glasses are going to replace the smartphone in a few years. They love to show AR demo videos to make their point, but actually using AR is never really as good as the video looks and that’s why the usage for this tech is actually really low. I saw a lot of this at Google where there was more focus on Google I/O demos than on customers. Maybe someday we will solve the tech problems to get magic glasses to work that people will wear, but a) people spend a lot of effort now to avoid wearing glasses at all b) it’s never been clear that putting 3d shit in front of you all the time was a real win. There’s going to be some nice applications of AR but it’s also a long way from making a big impact if it ever does. We are all waiting for Apple to figure out a way to defy Physics.
Crypto – I’m not a nocoiner. I still have some ETH I bought for $13 each in 2017 and I’ve made enough in crypto to finance a 3 month trip to Europe. Still, I’ve never believed in any of this crypto will replace fiat nonsense, much less that it will end war or something. Balaji is a smart person whose mind is poisoned by his ideology. I listened to the full 8+ hours of him on Lex Fridman to see if there’s actually anything to his bullshit and found it all empty. So many people in tech are drunk on the idea that bits > atoms but that’s only true until the famine happens or the tanks show up. Maybe the token prices will go up (I’m still HODLing some!), maybe we are going to have a few more options for payments/remittances, but we aren’t seeing the first steps in a new financial order. Balaji is gonna lose his bet and if the dollar loses its place it will be to other fiat currencies controlled by people with guns rather than techies with GPUs and an internet connection.
We are now caught up in an AI Frenzy. This recent round started with text to image generators and is now focused on LLMs. So I wanted to see what all the fuss was about.
Until recently, my formal understanding of ML was limited to doing the Andrew Ng ML course many years ago and I’d forgotten a lot of that. Having worked with the ML scientists on Alexa and being exposed to a few projects at Google I did believe I had some intuition about ML and its limits. My favorite thing to say for a while has been that ML is the right solution to every problem where it’s fine to be right 86.7% of the time..
Being done with having jobs gives you a lot of time to learn new things so I read some papers, went through the awesome zero to hero series, watched a bunch of other youtube videos while on my bike trainer etc. I also paid for GPT4 and copilot and tried them out. Overall my impressions are pretty close to this post.
In some ways my feeling about this hype cycle is similar to ones listed above. We aren’t close to the singularity. There’s a lot of people overstating both the benefits and the dangers here. LLMs are not that smart. Just because GPT4 is a lot better than GPT3 that doesn’t tell us much about the actual rate of improvement we will see as we scale up the number of parameters. There is a lot of drooling over making programmers or lawyers obsolete or 1000x more productive. Given what people actually do day to day in their work, I think the typical gains are going to look more like 2x than 1000x. A big deal to be sure, but not enough to decide that everything is going to be fundamentally different for humans in a few years.
However, unlike the technologies mentioned above, these new generative AI tools for images and text are having an impact already and they have properties which mean that the impact and the number of people using them are only going to increase. They are very easy to use, they have clear direct benefits and integrate well into existing apps and workflows. There’s no new hardware to buy and no need to believe some fantasy story about how money works. You just type into a box and get what you want most of the time.
GPT4 won’t make everyone into a great writer but they make it so that essentially everyone can write “well enough”. Stable Diffusion might not make anyone an artist, but it will enable anyone to make something that looks like art.
There will be all kinds of silliness and missteps, calls for regulation etc but it won’t stop the progress on these tools. I don’t think you can stop technological change. I just hope it comes at a rate that people are able to absorb it. I worry about this a lot for these tools. The people who make them barely understand them, much less the rest of us.
The future is always hard to predict. It’s unclear how many knowledge workers will be displaced by AI or when, but it’s pretty clear that the first to go are going to be the ones who don’t learn to use these tools. That’s the advice I give my kids.
Leave a Reply