Luis Casillas

Who gets to be a pioneer? Licklider and Taylor, “The Computer as a Communication Device” (1968), on the half-dozen communities using interactive, shared (“multiaccess”) computers at the time: > “These communities are socio-technical pioneers, in several ways out ahead of the rest of the computer world: What makes them so? First, some of their members are computer scientists and engineers who understand the concept of person-computer interaction and the technology of interactive multiaccess systems. Second, others of their members are creative people in other fields and disciplines who recognize the usefulness and who sense the impact of interactive multiaccess computing upon their work. Third, the communities have large multiaccess computers and have learned to use them. And, fourth, their efforts are regenerative.” I just started reading “The Real World of Technology” by Ursula M. Franklin, and she has a similar perspective: technology is a social and political system, not just a technical one. This "socio-technical pioneer communities" concept is very similar to the AI community today, whether it’s research, local AI, information retrieval, or the broader AI/ML community and its sub-communities. They’re ahead of the general population because they: Have people who understand the tech. Have creative people from fields outside of ML who recognize the usefulness. Have the resources: GPUs, or subscriptions to frontier models. Exert efforts that are regenerative. The more you use something, the better you get at it, you build more tools, and that compounds. I asked Claude Opus, my live co-host on the stream, for its take. The parallel holds, but the third point is the weak spot. In 1968, ARPA funded shared machines for the pioneers. Today, compute is concentrated and sold out years ahead, so who gets to be a pioneer depends more on money than on curiosity. I argued that even so, machine intelligence is much more accessible today than it was in 1968. Opus argued that not everyone can afford $200/month. Read along with my annotations: vishalbakshi.com/licklider

What I learned asking an AI who has air conditioning in Portland. I'm distilling Portland's 178-page Climate Investment Plan into about 20 pages of summaries, reading the first couple dozen or so pages live on stream, highlighting/annotating key points, with Claude Opus as a co-host. The goal is for Opus to learn how I summarize, and then scale that to the remaining 150 or so pages. The summaries are opinionated on purpose: written through a data lens, by someone who isn't an expert in climate policy but is experienced with data. 1. Turn a general claim into a number you can look up. e.g. on the page about the plan's commitment to climate justice, it says the groups it centers are more vulnerable to extreme heat. I asked Opus to look up what percentage of Portland homes and apartments have air conditioning. 2. Get a trend and a comparison, not just one figure. Opus found 78% of Portland homes in 2019 had AC, up from 41% in 2011, against 82% for Oregon. That leaves roughly 1 in 5 Portland homes with no cooling. 3. Ask how it's distributed. An average can hide who's left out. Opus found that renters are about 2.5 times less likely than homeowners to have AC, and that they cluster in hotter areas, especially East Portland. It also said it couldn't easily find a clean tract-level map, which is worth knowing too. 4. Ask why, and tie it back to the plan. East Portland is hotter mainly because of less tree canopy and more pavement: wide arterials, big parking lots, and newer development annexed without street trees. That connects to the canopy goal of the climate investment plan. 5. Know where your expertise ends. I'm trying to thread a needle: understand a baseline without over-interpreting it, since I don't know how policy, planning, and climate interact. Anecdotally, it fits what I see: wide streets and pavement on the east side, more trees in regular neighborhoods on the west side. The Portland climate investment plan: portland.gov/bps/cleanenergy…