I hope everyone is having a relaxing break. Here are the links that stood out this week:
- White House Unveils America’s AI Action Plan. Last week, the US administration released a three-part set of actions aimed at accelerating US AI development and US leadership through AI. The announcements got mixed reactions with praise from most major tech companies, but a wide range of concerns from others. In general, I think less regulation is better, but it is concerning that none of the actions make any mention of safety or ethical concerns. These announcements really are part of an AI arms race in which the Chinese government has already responded.
- 'You can't raw dog it' says Nvidia CEO Jensen Huang: 'If you're not using AI, you're going to lose your job to somebody who uses AI'. Actually, credit for the line should go to Jason Calacanis, but Jensen echoed it. This wasn't exactly what raw dogging has been taken to mean in the last few years, but I have a feeling this statement will stick. Just like being a graphic artist without Photoshop and other power tools has been impossible for 10-20 years, many roles without AI will be hard to function in.
- Figma’s AI app building tool is now available for everyone. Ahead of its upcoming IPO, Figma is pushing hard on AI. Figma is the leading app and product design tool, and it needs to stay ahead of upstart companies like Lovable and Bolt that are going directly after app building with AI. As well as being a threat, however, AI also represents a huge opportunity for Figma. If it can provide the rest of the app-building stack, all its legions of app designers might find that it is the natural home to become AI Vibe coders. It will be hard to pull off, but this team is likely in a much better position than Adobe.
- Claude for Financial Services Keynote. In an interesting milestone for me, the Anthorpic Team released a keynote video on their financial services offering. The first thing to say is that it is extremely difficult to get LLMs to give a reliable answer to questions that require reasoning about complex financial models. It seems likely that a lot of work has gone into training and also not just training the LLM itself, but also the LLM's usage of external tools to validate and check models (tool usage). This end-to-end training is likely critical to getting good outcomes and might be the reason why the Anthropic team is directly going after a vertical like this. Third-party builders that use Claude under the covers but build their own infrastructure might find it hard to get similar performance early on. It's also a clear indicator of just how much Anthropic's strategy deviates from OpenAI's. Vertical solutions are likely to be too distracting for OpenAI when it is trying to wrest control of consumer attention away from Google and Meta.
- The economics of superintelligence. The Economist takes a look at the future of economies in the presence of superintelligence. The biggest takeaway from the article for me, though, is that no one really has a clue. The points made by the author are a grab-bag of concerns without really saying much about an underlying model. It seems evident that AI will massively drive down the costs of certain types of goods (generally a positive unless you are currently producing those goods), and also monopolize a lot of future investment dollars. The thing that is difficult to model is just what is affected by the AI cost reduction and what new activities will arise. Maybe the only thing for certain now is that in the worlds of Jason Calacanis & Jensen Huang, if there are AI tools emerging in your business, it's going to be very difficult to raw dog it very soon.
Wishing you a great weekend.
