Professionally, I focus on creating social benefit startups. In my Saturday morning emails I share what I’m learning and thinking. Topics range from better living and parenting to business and philosophy.
Positive Visions of AI Safety. Did I bring too much doom last week on AI safety? In addition to Max Tegmark’s vision, I want to share some other positive ones. One, from AI Optimism comes “AI is easy to control.” They argue that we do pretty well at alignment with humans via culture and legal methods and human brains are black boxes. With current AIs we can see all their thoughts so we should be able to do better. Plus, we have the advantages of reproducibility, scalability, legal right to rights and social approval to experiment on AI. Two, the major labs like OpenAI, Anthropic, and DeepMind have published Responsible Scaling Policies. Three, technical controls like Gray Swan offer the opportunity for third party software to act as an application firewall sitting between AIs and the world to filter risky content and actions. Four, from Tokcast comes a series responding to AI 2027 and laying out a philosophical case for why LLMs lack the agency to intentionally cause harm. Plus, as software gets more capable, it will form better moral theories and want to cooperate with humans. (What about smart humans who are sociopaths?) He believes that we won’t be able to build genuinely creative agents without understanding how humans think. (What about the Wright Brothers building flying machines before fully understanding how birds fly?). And fifth, I offer you Today’s AIs Aren’t Paperclip Maximers.
Homelessness is a Housing Problem. Why do some cities have more people without homes than others? We’re offered reasons that sound intuitive like higher rates of unemployment, poverty, or mental illness. The book Homelessness is a Housing Problem answers the question persuasively that it’s fundamentally about the cost and availability of housing. The data tell a consistent story. Cities with high homelessness rates tend to be affluent because that’s where pricing houses tend to be higher, too. There might be an effect that people can pay more for housing so they do, but that’s not the whole story. Local regulations often restrict or slow new development, driving up prices and squeezing out those on the margins. The book offers a simple but powerful metaphor: musical chairs. Yes, someone with a broken leg might not be able to make it to a chair in time, but the deeper issue is that there aren’t enough chairs. So, how do we create more “chairs”? One place to start is by rethinking the way we regulate housing. Current zoning and building codes often mandate a minimum level of size, amenities, and parking which effectively rules out cheaper, more flexible options. What if we allowed more shared or smaller-scale housing instead?
Speech Roundtrip.Speechify is text-to-speech software that is essentially speed reading 2.0 where it highlights both the word and sentence while reading aloud to you. The experience isn’t flawless: it has bugs, and in complex PDFs it sometimes stumbles, reading things it shouldn’t (like footers, footnote references, or watermarks). Still, the overall value is strong and I’ve used it more and more given it works across mobile, desktop, and as a web extension. It occasionally mispronounces certain words, which can be distracting, but not deal-breaking. Do you have another a text-to-speech tool you really like? On the other side, I’ve had a good experience with SuperWhisper. It not only transcribes but also lets you run the text through an LLM for cleanup or more complex formatting tasks. It can even auto-paste into other apps, which is unexpectedly useful.
Until next time,
Miles