Apple’s strategy on AI and LLMs is the right one

As far as I can tell from my observations on generative AI, the deployment of the currently running large language models is running into roadblocks, such as - LLMs gradually hallucinating more content (https://gizmodo.com/study-finds-chatgpt-capabilities-are-getting-worse-1850655728) as more LLM-generated content enters the data pool and poisons it - Sources of data withering, like StackOverflow (https://www.similarweb.com/amp/blog/insights/ai-news/stack-overflow-chatgpt/) - Outlets for content getting swamped by bottom-feeding AI slop, like leading sci-fi publisher ClarkesWorld (https://www.theregister.com/2023/02/22/clarkesworld_ai_spam/) In the current dominant approach, data is “mined” into pieces of coal that can burn, but only leave ashes. This is not sustainable. On the contrary, an approach that discourages data mining and massive ingestion solves all these issues: - users trust the system, so they are generating more relevant, valid, human-made content that can be used by LLMs - fresh data that keeps the system sustainable - because the ownership of the data is clear (it’s the user’s) there’s no risk of litigation and no need to change the current IP structure - because users control the release of data, anything that is spammy or disruptive can be easily isolated to the user in question so they can be limited or banned In a privacy-focused strategy, data is not burned, but the minimum of it is taken so that the tasks the user wishes can be performed. This makes it more akin to a renewable energy. Therefore, I believe that a privacy-focused AI product, which is more likely to come from Apple than from other players, will be more sustainable and keep itself successful in the long run, just like our only chance of survival as a species is to hop on board of every possible (locally relevant) source of renewable energy.