Sora Slaps
M.G. Siegler
Forgive the self-link above, but playing around with Sora, OpenAI's finally released video-generation tool, I couldn't help but think back to the early days of using DALL-E, their first image-generation tool, which pre-dated ChatGPT. And that lead me to this post which I wrote exactly two years ago. I think it holds up well, but the craziest thing about it is that it was just two years ago. It feels almost as if it was more like a decade ago. ChatGPT was just about two weeks old at that point.
And, of course, we were just coming out of the "Web3" hype cycle. So it was easy to be beyond cynical with AI early-on.
Web3 has largely been technology looking for problems to solve while generative AI has been about almost too many solutions created by technology which is evolving on a seemingly daily basis. As a result, web3 has thus far been evangelists trying to convince us to re-solve old problems with their new technology. Generative AI has thus far been users re-solving old problems in stunning new ways as the new technology evolves to the point where we come up with new use cases.
That sounds like a knock on web3 and, well, it is. The hype that surrounded web3 was obviously unhealthy — including, namely, crypto. There was always going to be a crash and subsequent shake out of actual tech to use. Tech which will likely not look like some new "web3" but instead will augment "Web 2" in interesting ways.
But it also doesn't mean the hype around generative AI is the healthy variety either. Certainly, there will be a lot of bullshit here as well. Hype tends to attract charlatans as if moths to a flame. But again, the difference is that this is all starting off on very different footing. A stance where people actually really, really want to use this stuff. Or at least try it.
Try it, they have. Use it, they are. We've come a long way. The jury is still very much out in terms of what is a feature versus an actual company – and certainly far too many AI startups have been created as a result of the boom. But that will correct itself naturally – especially if the M&A window is about to open up again so we don't have Big Tech having to pull "hackquisition" shenanigans. And while a lot of usage is still people just trying stuff out, we do seem on the verge of true utility (beyond coding) on a few fronts, most notably with everyone rushing towards "agents".
For now, OpenAI brought back that sense of fun and wonder with Sora.1 Sure, the output can be weird and rough, but it also can make you laugh with true amazement; that a simple set of words has led to the creation of a fully moving image out of thin air.2 And just as with those initial image generators, the videos will get better, fast. Perhaps even faster than the images did, given where we are now with the underlying technology and scale. As I wrote two years ago:
There's a part of all of this which has a very "fake-it-til-you-make-it" vibe. That we're giving all of these projects far more credit than they deserve as they tickle our minds with easy illusions. But there's another part of all of this which is absolutely unlocking new forms of creativity. And a belief in new bounds for what is possible.
As an end-user, it's all rather exhilarating to behold. This really feels like it's just the beginning. These projects are not only fun to play around with, they're evolving so fast that each week seemingly brings some new breakthrough. As an investor, it's all happening so fast, with again, so much hype (not to mention some gnarly ethical and societal questions), that no one knows how this shakes out. But it won't be the demand problem of web3. In fact, it might be a too-much-demand problem which leads to servers melting down with costs spiraling. Ultimately, it will be companies trying to pick the right mix of rapidly evolving tech to fulfill user demand and weld it all into actual businesses. Easier said than done.
Again, that's where we are now. But Sora is a good reminder that this isn't done yet. The LLM work may be slowing, but there's a lot more to the world than words.
A "spy" in London...
1 Does this open up more cans of worms with both societal and ethical questions, sure. And is Hollywood going to lose their shit? Of course. But my bet remains that this will end up helping that industry in the long run. And elevating human creation even further. My data point is the history of technology...
2 And yes, there will be many questions/debates on where exactly this data is coming from to train the models that outputs this content. Perhaps even more so than with text and images.