The lines between research and industry in AI are a bit blurry at the moment.

Companies are trying to sell products by touting state-of-the-art performance on research benchmarks. Researchers are relying on commercial, black box LLMs to generate/ and analyze data, some of which may have been part of the model training. We don’t know because the firms are keeping it close to the vest.

It’s tricky, then, to really clearly identify novel scientific contributions.