First of all, we hope everyone has had a great Christmas and New Year break. It is always nice to take a few days to reflect on the past year and look forward to the next. Here we are, 2023! What a year we have to come and will we see AI stop accelerating at such a rapid pace? Will it hit a plateau with no major breakthroughs or will people start getting bored of it? On the flip side, are we on the cusp of major general adoption?
This is a fun blog post to discuss how we see the AI landscape develop and advance in the new year. Let's get into it.
We've already seen major advances in the image AI space. I remember getting access to DALL-E pretty early on in their closed beta. Back in those days, we got unlimited generations for free, that was pretty cool! Since then, we've seen Midjourney do some awesome things and receive a cult following and then Stable Diffusion truly changed everything.
The initial version of Stable Diffusion didn't really work so well at less than 50 steps so the generation times on the best hardware was about 6-8 seconds. Try and do that on your iPad (download Draw Things to try it out if you have an M1 chip!) and you may get an image in 30-45 seconds. A gaming laptop I own with a 3060 RTX chip generates in about 20-30 seconds but things are progressing fast.
I'm predicting that the steps will become less relevant and faster and due to this and other advancements, I expect that generations will reach the optimum point in happening in real-time. The hardware required to do such generations will continue to get less. You will be able to do instant generations on the latest smartphones in real time without accessing the internet. That's going to be super fun.
This prediction is both for text-generative and image-generative AI. Those who have a knack for building out extensive datasets and taking that data and putting it into a format that a business or individual can extract value from will become very sought after and in demand.
AI will continue to develop at a rapid rate but you will be able to see a piece of text and just know that AI wrote it with a vanilla model. Being able to go into an organization, extract their style, values, and skillset and push that to a customized model for the organization is worth thousands and thousands of dollars at the enterprise level and there is a massive business opportunity for those who pursue that area.
Similarly, people are generally lazy. They don't want to do the work of putting together a dataset to get the outputs that they would love to have. Solving this problem at scale will make someone or a business a ton of cash.
At the end of 2022, you started to see more and more artists speak out about AI art. This will continue but I think something interesting will happen here before the year is out which I'll put in the next prediction. I think that the anti-AI movement will continue to grow and a backlash will occur against companies that abuse AI and try to generate content en-masse.
I am making this prediction mainly from what I am seeing on Twitter. There are already a ton of bots using ChatGPT to answer tweets. The answers are fairly obvious from AI and they often get called out and blocked by those they are trying to interact with. This is still brand-new technology and the lines of acceptable use and what people will tolerate are still to be drawn up.
Expect that just as many people who discover this technology for the first time and fall in love with it will also join the anti-AI movement and detest it. It's going to be a whole new culture war.
A lot of artists don't like AI. A lot of artists see it as a threat. The smart artists and the ones who adapt to the modern world (this is not going back in the bottle) have the potential to make a lot of money. Whilst I see the backlash and resentment for the technology continuing to grow, there will be some traditional artists who see the potential and work with model creators to fine-tune their own generators.
These generators will be private models with no open-source license and will be available on standalone platforms or partner agreements to the highest bidder. The moat in image AI is all about the models, not many have realized this yet, but more and more ears and noses are coming around to this conclusion. Watch the image AI space over the next month and who puts out proprietary models outside of what you can find on Huggingface. Artists can and will get in on this action and can be rewarded handsomely if done right.
This is probably the prediction that I hope to be proved wrong on, but let's see how this plays out. OpenAI has been pretty notorious for botched rollouts of technology so I am pretty certain it won't be a case of 'Here we go, here's GPT-4, come play with it'.
There will be waitlists, there will be Silicon Valley cronyism, and there will be restrictions on how you use the technology. I'm really hoping that OpenAI turns a corner and with Stability AI doing so much for open-source technology, they discover what this technology can do for humanity and be a bit more leveled in its release but I won't hold my breath.
In terms of the substance of GPT-4. I expect it to be great. I expect it to be really, really solid but I also expect that it will be programmed to answer in a specific way. It will be interesting to see how prompt engineering and fine-tuning work on molding the technology to an individual or business need.
I also really, really hope that they don't over-engineer the safeguards for the technology. There is a lot of chatter about ChatGPT and how it tries to avoid certain subjects or simply won't write about certain things. If this is built into GPT-4, we are in a dangerous place where one company has the influence to determine what is moral and right to create and what isn't. That should never happen.
I hope that AI continues to accelerate in development and I believe it will do so. I also expect Google and Meta to play a more active role in actually releasing some models over the next year instead of keeping things internal, I believe we are at a point where their hand will be forced.
I also hope that Stability.AI will continue on its mission and not become too big or lose track or focus on what they are trying to achieve. There will be incredible advances in the open-source technology available and I believe that the open-source community is more powerful than any corporation in the long run so will see a release of an open-source model to rival GPT-4. It isn't a ridiculous thing to consider.
Let me know what you think will happen in the world of AI in 2023. Do you share some of my predictions or not? We'll go back to building out Riku.AI to be the best place to learn, experiment, deploy, and enjoy all things AI and boy do we have our own fun announcements for the things we have planned this year ahead.
