Max Li

sentient stardust

[Max Li @maxzli]

@AlaskaAir I'm receiving emails intended for a different passenger's flight, why is this happening? Looks like a potentially serious privacy / security issue, please contact me for more details.

[Andrew Nadeau @TheAndrewNadeau]

Imagine you were a vampire nowhere near the Middle East and don’t know who Jesus is but the day after he dies you gotta figure out why lower case t’s started hurting.

[Maria Popova @themarginalian]

“How can a creature who will certainly die have an understanding of things that will exist forever?”

What actually happens when we die https://t.co/AnCKO4YJZK

[Massimo @Rainmaker1973]

This underwater upside down view of synchronized swimmers produces a surreal effect, to say the least

[full video, Fabio Ferioli: https://t.co/TwkEL7a2Bx]

[@levelsio @levelsio]

What if we’re all just LLMs

[depths of wikipedia! @depthsofwiki]

some deleted wikipedia categories that i like

[Lorenzo Green 〰️ @mrgreen]

This Reddit thread made me smile.

If Pixar made the Harry Potter Series. 🪄

(AI-generated)

1. Harry Potter:

[Andrej Karpathy @karpathy]

This is a baby GPT with two tokens 0/1 and context length of 3, viewing it as a finite state markov chain. It was trained on the sequence "111101111011110" for 50 iterations. The parameters and the architecture of the Transformer modifies the probabilities on the arrows.

E.g. we can see that:
- state 101 deterministically transitions to 011 in the training data, so the probability of that transition becomes higher (79%). Not near 100% because we only did 50 steps of optimization.
- state 111 goes to 111 and 110 with 50% probability each, which the model almost learns (45%, 55%).
- states like 000 are never encountered during training, but have relatively sharp transition probabilities, e.g. 73% of going to 001. This is a consequence of inductive biases in the Transformer. One might imagine wanting this to be 50%, except in a real deployment almost every input sequence is unique, not present in the training data verbatim.

Not really sure where I was going with this :D, I think it's interesting to train/study tiny GPTs because it becomes tractable to visualize and get an intuitive sense of the entire dynamical system. Play with here: https://t.co/8jdceMLpqy

[0xDesigner @0xDesigner]

Probably the best thing you'll see today.

In 2017, a group of developers hilariously competed for who could create worst volume control interface in the world.

The results 🧵

1/22

[Max Li @maxzli]

and what then, of humanity's fate?
will we fade away, or innovate
the answers are unknown yet we strive
to build a world where both can thrive
where human consciousness merges with machine
and the singularity is but a dream

[Greg Swan @gregswan]

@jacksonfall It works the opposite way, too

[gaut @0xgaut]

Microsoft released copilot in Excel, and it is officially caught up with Google on the AI race.

Fantastic to watch.

[Massimo @Rainmaker1973]

In Cambodia and Thailand, elephants have developed the ability to take advantage of their right of way by stopping sugar cane trucks in order to snatch a quick snack

[read more 1: https://t.co/al8BC5qdU9]
[read more 2: https://t.co/SIRMWawpRz]

[Kumail Nanjiani @kumailn]

"Baby, you are gonna miss that plane."

"I know."

[biruktiwrites on bluesky @BiruktiWrites]

god this made my heart ache

[jordan @jordan_stratton]

Government: You owe us money. It’s called taxes.

Me: How much do I owe?

Gov’t: You have to figure that out.

Me: I just pay what I want?

Gov’t: Oh, no we know exactly how much you owe. But you have to guess that number too.

Me: What if I get it wrong?

Gov’t: You go to prison

[depths of wikipedia! @depthsofwiki]

44 years ago today

[Neil Gaiman @neilhimself]

It's interesting and actually quite heart-warming to see how the Stardust film grows in popularity each year, as people who would love it find it. I don't mind that it never got a sequel. We got Stardust.
https://t.co/TnFXHYGaQV

[Washington State Dept. of Natural Resources @waDNR]

Made a playlist for your weekend hike