resource0x's comments

On: Justin Trudeau promises to resign as PM

[resource0x]

Months? Well, only if you mean 120 months. The guy managed to destroy the best country in the world, leaving it with no foreseeable way back. Maybe joining the US is the only option now.

On: Tell HN: Is this AI doom loop? I thought o1-previe...

[resource0x]

It's time for "How many programmers does it take to screw in a lightbulb using AI" jokes.

[infamouscow → resource0x]

None, because it's a hardware problem.

On: The Threat to OpenAI

[resource0x]

Does the same bullish logic apply to cold fusion?

On: Unproductive Claims about Generative AI in 2024

[resource0x]

The hype must go on :)

On: Ask HN: Revitalizing a career: how do I escape CRU...

[resource0x]

Java is much worse. You can learn embedded, but it's a different world and doesn't pay much.

On: ARC Prize – a $1M+ competition towards open AGI pr...

[resource0x]

I haven't seen 1000 cats in my entire life. I'm sure I learned how to tell a dog from a cat after being exposed to just a single instance of each.

[lostmsu → resource0x]

I'm sure you saw over 1B images of cats though, assuming 24 images per second from vision.

[lelanthran → lostmsu]

> I'm sure you saw over 1B images of cats though, assuming 24 images per second from vision.

The AI models aren't seeing the same image 1B times.

[lostmsu → lelanthran]

Neither are you, during those 10 000 hours most of the time you aren't absolutely still.

[lelanthran → lostmsu]

> Neither are you, during those 10 000 hours most of the time you aren't absolutely still.

So? I'm still seeing the same object. Large models aren't trained on 10k different images of a single cat.

On: Ask HN: When will the AI bubble burst?

[resource0x]

We will know it was a bubble when it bursts. It might be followed by another AI bubble 20 years down the road. There's no contradiction between "technology trend" and periodic bubbles. Those who lose $$$ along the way may console themselves with thinking about technology trends. :-)

On: Claude thinks it's a tumor, radiologists disagree

[resource0x]

"We should stop training radiologists now. It's just completely obvious that within five years, deep learning is going to do better than radiologists"

G.Hinton, 2016

[viraptor → resource0x]

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6483404/

> The error rate produced by a deep learning based methods dropped below that achieved by human observers in 2015 for the first time

Essentially we just need good implementations at this point. We're there, just not widely deployed.

On: Empty “backpacks” activate the immune system again...

[resource0x]

In mice.

On: The baffling intelligence of a single cell: The st...

[resource0x]

I'm curious: if tomorrow someone discovers that some parts of the genome are encoded with RSA cryptography, will the argument of "mutations, time and selection pressure" still hold? The same handwaving applies :-)

[pixl97 → resource0x]

It 100% still holds.

The convergent evolution of pitcher plants is an example of this. There are a number of features that evolve to the same functionality in plants, and many of them have to work together to become fully functional. Yet we see in plants separated by vast distances and millions of years of separation that traits that are useless alone encode themselves and then will will have near spontaneous usefulness when some other gene evolves.

The laws of large numbers are not things the human mind really grasps well at all.