Learning flywheels are all you need

One intuition pump for the future of AI is to see what happened with human intelligence in our evolutionary past.

Our ancestors 100k years ago had the same cognitive capacity as us (evolution works slowly) and yet all modern technology and knowledge has only emerged in the last 1000 years or so.

Why such a sudden jump?

It’s not because our individual intelligence improved, but that we assembled learning flywheels over time (writing, books, schools, colleges, scientific method) and those caused each individual to be compound over the previous generation leading to the culture explosion that were going through. ...  Read the entire post →

My Claude Code workflow

Half my time goes into using Claude Code, and the other half goes into optimizing my workflow for it.

What I have now:

• Everything organized by sprints in ./sprints/v1, v2 etc folder

• Custom command /prd to help me brainstrom requirements for a sprint and break it down into atomic tasks (which should take 5-10 mins each)

• Custom command /dev to pick highest priority task in a prd and follow test driven development to implement the tasks

• Custom command /walkthrough to write a sprint review report that details what code was produced so I can read and understand exactly what the code does ...  Read the entire post →

Words are minimum-viable coordination tools

Words have a bewitching tendency as we assume they point to some deep essences. But, game theoretically speaking, words exist to get a job done so they operate at the level of coarse graining that’s sufficient to get the job done of the speaker.

Evolution doesn’t like to waste energy. Hence all communication between people is a coordination tool where all parties are interested in getting their job done, but not wanting to invest more energy than it’s necessary to do so.

So if someone uses the word “God” or “Love”, the job is done if it elicits the emotions, actions and associations roughly associated with what the speaker intended so our search for what those words “truly” mean is just misguided. Meaning is in what the exchange does in a particular context. By themselves, words are empty. ...  Read the entire post →

Nobody cares about your idea.

The most important question you should be asking while developing a product / startup is this:

why would anyone change their behaviour to accommodate your product?

And the only correct answer to this question is that they’re *already* exhibiting the behaviour and your product will simply help them be >2x more efficient on a dimension they cares about.

Behaviour change is tough. Internalize that nobody changes their behaviour for marginal or incremental benefits.

And nobody certainly cares about your idea. ...  Read the entire post →

A proposal to prevent job losses from AGI

We may have only a narrow window before big AI labs automate away all economically useful work and centralize wealth. We’re certainly on that path right now.

Some people want to pause all AI development because of this risk of human disempowerment (+ also possible extinction risk). But stopping tech development also means we give up on benefits and abundance that could come along.

Is there a middle ground? Can we keep reaping the benefits of AI without rendering humans obsolete.

I think so. Here’s how we can probably achieve it. ...  Read the entire post →

Will we forget what pre-LLM era looked like?

Was chatting with a friend yesterday and it hit us that pretty soon we will forget what it was to live without LLMs.

It’ll be inconceivable to us that there was a time when if you had a question, you did not get an intelligent instant answer for it.

Some more interesting predictions that came out from our conversation:

– Maybe we will stop having multiple softwares; just one agent that spins up personalized software on demand

– Maybe soon we will have so much automation that our AI agents will predict what we want to do next, leaving us to wonder what’s our role when we open up our work laptops ...  Read the entire post →

Are we headed towards a stable dystopia?

I’ve been thinking about how AI is going to shape up society and one particular scenario seems like a high-probability outcome to me. Here’s how it could play out:

• AI automates more and more of human-work

• Returns from this automation flow back to capital holders, making them wealthier

• As incomes of masses drop, government ignores/abolishes income tax but rather taxes corporate income / wealth for its needs

• Rich isolate themselves (as they do) and use AI-enabled drones/robots to protect themselves ...  Read the entire post →

Crossposting from X to my WordPress

Today I built a nifty little utility that monitors which of my tweets has /syndicated tag in it, and then crossposts automatically to my blog (invertedpassion). This is done via webhooks from Typefully, the app I use to post my tweets.

If all goes well, I should see this both as a tweet and a blog post.

I did this because I have started writing long-form content here, but it irks me that my thoughts will be behind a walled garden. So, this workflow will enable me to post once and syndicate everywhere!

In case you’re interested to know more, see https://alexn.org/blog/2023/09/21/post-once-syndicate-everywhere-pose/