Elon Musk announced his new AI startup xAI earlier this week. According to the website, the goal of xAI is to understand the true nature of the universe. Well, I don’t understand why it has to sound so exotic. xAI is basically a competitor to OpenAI and it appears that a very strong team of AI experts have been assembled. I am excited about what they will accomplish. These AI models are so powerful that I am not comfortable for these models to be controlled by a small number of corporations like OpenAI or Google. I think it’s good to see a large number of organizations coming up with their own high-performing LLMs so no company has a real monopoly on this. Ideally, I hope some organizations open-sources one of these super powerful LLMs (that are on-par with GPT 3.5+ in terms of performance) to truly level the playing field.
OpenAI and Google keep talking about the AI existential risks but I personally worry more about their monopoly in the field. If some regulations are put in to restrict who could work on AI, we will inevitably see a small number of organizations monopolize the field and they will ultimately have too much power and make smaller organizations and individuals powerless. We have seen this play out in big tech. Google, Facebook and Apple have had too much power over small businesses and individuals. They pretty much decide who gets the attention and what people can say on the internet. It will be disastrous if we have a small number of AI models controlled by a handful of organizations that decide how humanity should behave. Personal freedom will be restricted if that happens. I would rather see many different versions of powerful AI models run by many different organizations so people are not subject to biases of particular models or organizations and are free to choose different AI models that suit their needs.
Agreed.
Alogopy/monopoly risk definitely a concern as these models get bigger and more expensive to run.
The big breakthrough with these models is they now pass the Turing test. However they are still pretty low precision high recall. So yeah. Increased spam and fraud are a risk as it gets cheaper to do more volume. But wwIII and the risk of these models "escaping" is kinda laughable.