Welcome to this week’s Smokebox! A weekly digest of some of the most interesting AI-related news items each week:
Benedict Evans outlines many of the key questions around AI and Copyright. No good answers yet, but good questions. I plan a full post on this in the future, but I agree with him that it’s important to realize that an LLM is more like an inference engine than it is a database. In the meantime, some of the biggest stars are seeing the potential in licensing the rights to digital twins.
Venture fund Andreesen Horowitz announced an open source grants fund to support AI projects that will help key projects with no equity involved. There’s no doubt support is needed, and it’s great to see an initiative like this. On the other hand, it seems like a much bigger problem: maybe there should be a bigger global fund for key open source and not just in AI.
Ideogram, a new high-powered image generator, opens for full access. Ideogram’s main touted differentiator over Midjourney and Stable Diffusion is that it is much better at rendering in image text. It’s amazing how fine-grained competition has already gotten. This is super hand if you’re making logos, stickers, or signs, however - today’s reimagined logo comes courtesy of Ideogram:
The big money number of the week belongs to OpenAI, which is projecting $1B in revenue this year. This was widely reported and shows a real market pull for LLM services. The compute costs are eye-watering, though, so the economics are undoubtedly tough. It will also be interesting to see how much other large LLM players will fragment the market. (AI21 Labs just raised a further $155M for its LLM.)
If you’re of a nervous disposition, you might want to skip this last one: Earlier in August, Nuclearn.AI announced a new $2.5M funding round for AI in the nuclear industry. Hopefully, there’ll be some vigorous testing and verification involved! No doubt there will be: my father, in his early career, actually helped write control code for a nuclear power plant. This was a scary proposition even for conventional code, but there were a lot of safeguards in place.
As a last item this week, a Public Service Announcement. If you are a Facebook user, you might want to request an opt-out of their Generative AI Data sharing. The form is here: https://www.facebook.com/help/contact/1266025207620918.
It’s unclear precisely what the data is being used for or who it is shared with. It appears to only apply to third parties, not Facebook itself. These types of data-related terms of service changes are appearing everywhere at the moment.
Wishing you a great weekend!
