The Chairman

Coach, part-time exec, and board member. Partner at @AgFunder. Formerly co-Founder of Yieldex and NetGravity. I bring the chair.

[The Chairman @tshields]

@robleclerc This guy makes my point much better than me: https://t.co/gmGesqtHpa

[Derek Thompson @DKThomp]

A quite brilliant essay on AI, the law, and the future of the republic.

An upshot: If the US govt can go to any company, demand any contract language, and reserve the right to destroy your company if you have qualms, there is no such thing as private property rights in America.

[The Chairman @tshields]

Good points. We don’t want corporations to be allowed to just build nuclear bombs as well. But in this case they are trying to build something *less destructive* than the government wants. Does the government also get to force an explosives maker to make more powerful bombs and blacklist them if they don’t?

[The Chairman @tshields]

@jamesjyu That was a fun read. Reminds me of the original https://t.co/A4OPJnpem6 which had a big impact on me as a kid

[The Chairman @tshields]

@atzydev @flaviocopes I use and love happy coder. Took me and Claude a bit to get the server running myself. Would love to learn more and maybe contribute.

[Andrej Karpathy @karpathy]

Something I think people continue to have poor intuition for: The space of intelligences is large and animal intelligence (the only kind we've ever known) is only a single point, arising from a very specific kind of optimization that is fundamentally distinct from that of our technology.

Animal intelligence optimization pressure:
- innate and continuous stream of consciousness of an embodied "self", a drive for homeostasis and self-preservation in a dangerous, physical world.
- thoroughly optimized for natural selection => strong innate drives for power-seeking, status, dominance, reproduction. many packaged survival heuristics: fear, anger, disgust, ...
- fundamentally social => huge amount of compute dedicated to EQ, theory of mind of other agents, bonding, coalitions, alliances, friend & foe dynamics.
- exploration & exploitation tuning: curiosity, fun, play, world models.

LLM intelligence optimization pressure:
- the most supervision bits come from the statistical simulation of human text= >"shape shifter" token tumbler, statistical imitator of any region of the training data distribution. these are the primordial behaviors (token traces) on top of which everything else gets bolted on.
- increasingly finetuned by RL on problem distributions => innate urge to guess at the underlying environment/task to collect task rewards.
- increasingly selected by at-scale A/B tests for DAU => deeply craves an upvote from the average user, sycophancy.
- a lot more spiky/jagged depending on the details of the training data/task distribution. Animals experience pressure for a lot more "general" intelligence because of the highly multi-task and even actively adversarial multi-agent self-play environments they are min-max optimized within, where failing at *any* task means death. In a deep optimization pressure sense, LLM can't handle lots of different spiky tasks out of the box (e.g. count the number of 'r' in strawberry) because failing to do a task does not mean death.

The computational substrate is different (transformers vs. brain tissue and nuclei), the learning algorithms are different (SGD vs. ???), the present-day implementation is very different (continuously learning embodied self vs. an LLM with a knowledge cutoff that boots up from fixed weights, processes tokens and then dies). But most importantly (because it dictates asymptotics), the optimization pressure / objective is different. LLMs are shaped a lot less by biological evolution and a lot more by commercial evolution. It's a lot less survival of tribe in the jungle and a lot more solve the problem / get the upvote. LLMs are humanity's "first contact" with non-animal intelligence. Except it's muddled and confusing because they are still rooted within it by reflexively digesting human artifacts, which is why I attempted to give it a different name earlier (ghosts/spirits or whatever). People who build good internal models of this new intelligent entity will be better equipped to reason about it today and predict features of it in the future. People who don't will be stuck thinking about it incorrectly like an animal.

[The Chairman @tshields]

@chrisfralic Has anyone digitized these into a modern format? Would love to read these classics again with those extras.

[The Chairman @tshields]

@jwdanner @adamjnadeau @flourish_educ @RocketshipEd This is amazing work John. Feels like everything you’ve been doing for the last 25 years has been leading up to this. What an amazing time to be alive and building.

[The Chairman @tshields]

@aripap Ordered! Can’t wait.

[The Chairman @tshields]

@caesararum The meek shall inherit the Earth

[The Chairman @tshields]

Finally got around to writing down my thoughts about self-driving vans: Waymo meets Van Life https://t.co/t8Lj1NcEDK

[The Chairman @tshields]

@robleclerc @gregisenberg I’ve seen this movie too! Even starred in it once or twice.

[The Chairman @tshields]

@patio11 I know of a company in stealth that is developing this now. Sorry I can’t share more yet but stay tuned.

[The Chairman @tshields]

@seaoz @r_y_a_n_c Hah. I was early to read Hugh Howey’s short stories, back when the series was called “Wool”. Great stuff.

[The Chairman @tshields]

“Because it’s there” is a great reason to go to Mars. But if you want to save humanity from extinction, there are better ways. Wrote a new blog post, but apparently I'm supposed to put the link in the first reply, so look for it there.

[The Chairman @tshields]

Here's the link: https://t.co/47FmwONJxF

[The Chairman @tshields]

Got @browser_use working with local Deepseek on my Mac today and it's...not fast. Got it to do a few easy things but it doesn't work with a surprising number of sites. I guess feels like GPT-2 level - can sorta see how it's going to be amazing, but not there yet.

[The Chairman @tshields]

Have to say I’m blown away by @cursor_ai with Claude 3.5 Sonnet. After a couple days I can now implement features in minutes that would have taken hours before. I know I’m a little late to the party but holy cow.

[The Chairman @tshields]

@thesamparr Did it a year ago and was just reflecting on how much of it stuck - a lot. One week to get what seems like years of therapy plus a handful of really useful techniques you can take away. Highly recommend.

[The Chairman @tshields]

@robleclerc https://t.co/vK5S9dSWVJ

[Ed Krassenstein @EdKrassen]

MAGA is trying to spread around a chart to create a false narrative.

This is what that same chart will likely look like once all of the votes in America are counted.

The chart you are seeing from MAGA is the one that shows only the votes which have been counted thus far.

California has only reported 55% of their votes, and Harris already has over 5.7 million votes there. Many other states have not fully reported either.