My first serious attempt at fiction writing is online! The Connection explores a world where technological advancement centers around human connection. Check it out! https://t.co/PV56o8xIGU

I'm curious if you see any flow in this argument!

This feels rarely present in the classic post-singularity / post-scarcity narratives

This means any AI futures *need* to include humans pushing through learning new things, failing, feeling pain, having friends remotivating them, trying again and eventually succeeding in unexpected ways.

1 Good futures involve human growth. 2 Human growth happens through direct contact with the world. 3 Direct contact with the world is hard. 4 No external technology can replace humans having direct contact with the world. → Humans need to keep doing hard things no matter AI

Animism might be the right mental operating system for the post-AI age

Perfect online dictionary idea: You enter a word in a search bar, hit enter, and the word gets displayed back to you in font 400pt.

The world is everything you pay attention to.

A book need high-quality attention to keep the flow going, a video game provides flow very easily. If we could turn a non-fiction book into a video game, what would we lose?

Great tools for thoughts don't work by making the true thinkable, they work by making the untrue unthinkable.

I wrote an appreciation for typography (and type design) in a new blog post! https://t.co/pucTKyVrc0

Intellectual progress often happens not by finding the 'right' answer to the question that you were obsessed with, but by finding a more interesting question to think about.

Many works of design, art, or engineering take the form of time compression of care. 1. A producer spend time exerting iterative care for the features of an artifact 2. The receiver of the artifact enjoy "compressed care" in a short burst relative to the production timeline.

What does success look like for human amplification? Extending the set of experience one can have dreams about. It's a clear-ish check that shows you became a friend with a piece of the world, a skill, a tool, a person.

Sometimes it really feels like you can, alas no, you cannot teleport in thought space.

Confidence is the knowledge that you are honest.

Example of adversarial programs which cause Inception V3 ImageNet model to function as MNIST classifier. (from the paper)

@CRSegerie Thanks for your reaction! I realized I forgot a tweet in the original thread, so the story was not making much sense 🫠 This thread is complete. Does the idea make sense now?

Analogy mapping: * Dish washer = NN trained to classify images of animals * Raw fish = image of a digit * Cooked fish = digit classification * Aluminum foil = adversarial program * Moist and heat = ??? Functions like edge recognition

The program has been carefully optimized such that putting an image of a digit in the middle makes the neural network perform digit recognition. To recover the digit prediction, you need to apply a pre-defined mapping such as "giraffe = 0", "deer = 1", etc.

For this, the authors crafted an equivalent of the aluminum foil, an image container called an *adversarial program* that looks like meaningless colorful patterns.