Will Knight
Since 2022, America has had a solid lead in artificial intelligence thanks to advanced models from high-flying companies like OpenAI, Google DeepMind, Anthropic, and xAI. A growing number of experts, however, worry that the US is starting to fall behind when it comes to minting open-weight AI models that can be downloaded, adapted, and run locally.
AI Lab is exclusive to subscribers.
Open source is absolutely not the correct move here. Open source is great for technologies that don't have the potential for Dangerous outcomes. If we are worried that China is innovating ahead of us, the answer will never be to outdo them and share that information with them. The world of AI today is like an errand where we needed to go to the grocery store to pick up some food and then go to the birthday party afterwards. We started on a bicycle with a basket towards store, but then we added a motor and started traveling faster. After that we added a shell around the bicycle, we added more wheels, we added a better seat and space for passengers. In fact people watching us and what we did with our bicycle continued to be amazed. The only problem was, we never made it to that grocery store and we never made it to the birthday party. In a world where every engineer says what they've created is unique and special it becomes very difficult to understand what actually is. My most recent granted patent is the example of what the AI looks like when we allowed the bicycle to be upgraded but actually went to the grocery store. As a result, I have a large language model evolved to the next level, it does not hallucinate, scheme, it allows input chunks and those are separate and reference chunks. The limitations in prompt size, context window, response size have all been overcome, and the entire vectorized embeddings representing text from prompts and chunks had to be completely ripped out and rewritten.
No comments:
Post a Comment