Leverage
The State of AI with Marc & Ben
open.spotify.comAI and the Internet are different phenomena.
The internet increased the leverage of computers by connecting them to other computers. Amplifies things that benefit from wider distribution. Network effects.
LLMs are stateless calculators (computers). Amplifies things that have a lot of computational complexity. Quick and Reliable decision making.
The History of Mathematical Tables: From Sumer to Spreadsheets | Oxford Academic
academic.oup.com
Quote reply to the previous tweet that captures most of my thoughts well with lots of examples.
Excerpts:
(On design): When designing productivity software, conceptualize it as a medium first and foremost, and a tool secondarily.
(On chatbots as UI): Its browsing, code interpreter, and Dall-E tools are testament to this upside-down design philosophy: the tools go to work first, only thereafter you see some output.
With the exception of code interpreter, interactions with ChatGPT are largely stateless and transactional. Manipulations are indirect.
When the first steam engines were introduced, the owners had a really hard time finding people to operate them who wouldn't get drunk in the middle of the day, fuck up, and wreck the machine.
Ben Landau-Taylorx.comTechnological change is only possible with societal (behavioral) change
To use databases, you need to be able to think about your data in a structured way. The people who could do that gained most of the leverage from using databases. The folks who made it easy for people to transition to databases (ERPs and Co) captured the rest.
AI systems are fundamentally scalable decision making engines. And data is the oil required to power it. If you want to properly leverage AI, you need to make sure you're fueling it with the best refined data.
Otherwise, you're no different than the drunk trying to reliably operate complex machinery. The only repeatable part of that is injury.
@john_lam @simonw Reminds me of this Joel Spolsky anecdote "the gridlines are the most important feature of Excel" that I've never forgotten https://t.co/ygM8xxm8tw https://t.co/DDnam30nEQ
hate to admit it, but this tracks. ui design starts becoming more important the moment you have a sufficiently capable data structure/model in place
users will ask for features when what they actually want is a specific view
this probably explains notion » coda
notion has very good views and is adding automations after the fact
coda has very good automations and is adding views after the fact
my hunch is most people are visual learners. they prefer to have information laid out the right way and only then can they reason about it.
automations are reasoning before seeing
this is also why cli tools, chatbots, and other text based ui paradigms are niche or quickly fall out of favor. too much reasoning without the requisite framing.
Fascinating historical case of technological and strategic advantage: 400 Spaniards conquered the 6-million Mexican Empire, and just 180 soldiers overwhelmed the 10-million Incan Empire.
Will AGI bring the same kind of strategic advantage, potentially decisive? https://t.co/vqf7eNpICI
the fastest path to being perceived as a “genius” (commonplace hyperbole) is to read books that your peers aren’t reading. this will not be a universal perception, but it’s nonetheless a very strong effect that’s amusingly underutilized
x.comSimilarly, when everyone is saying the same things, simply asking why becomes revolutionary.
The Death of the Junior Developer
sourcegraph.com
AI adoption and capabilities are distributed bimodally across industries.
Writing-heavy professions like writing, law, software are feeling it first. It doesn't feel that way because other industries are still untouched. That may continue for long or it might not.
It's always calmest before the storm
This talks about value capture by focusing on the most critical point of leverage