State of AI Report 2023

Too many interesting takeaways to cite them all, but I’ll list a few:

• Researchers are finding that shorter prompts result in better results from LLMs
• Small language models trained with specialized and curated data sets can out perform larger rivals by 50x
• The use cases for generative AI are continuing to grow. Most interesting to me are gains in bio medical research. I’m also eager to see where the early work with weather predictions goes
• Companies are looking at novel ways to watermark their generated content. As a consumer, I’m grateful that this will make it easier to identify generated vs. ungenerated content, which will hopefully unlock some web standards around automatic attribution
• VC continues to pour millions into startups doing something related to AI. A novel technology does not a good company make
• The overall takeaway is things are getting bigger, faster, and more complex.

My big question going into next year: Where will the energy and the water come from to sustain this “revolution”?