dontwearitout

8 ms·

31.
by dontwearitout 3y ago
Thanks for sharing, that was a great read. I've always really been fascinated by the aquatic ape hypothesis ( https://en.wikipedia.org/wiki/Aquatic_ape_hypothesis ), even if it's not widely accepted as fact.
32.
by dontwearitout 3y ago
You make some great points and I agree. Low cost autonomous drones swarms aren't scary because they're more lethal than 155mm shells, they're scary because there's no effective defense (yet).
33.
by dontwearitout 3y ago
Yes, a software stack to turn these drones into autonomous area denial or autonomous trench clearing tools could be readily cobbled together using mostly open source components. It would be such a tide changer that it's only a matter o
34.
by dontwearitout 3y ago
"You will live to see man-made horrors beyond your comprehension." - Tesla The prospect of low cost swarms of AI enabled drones terrifies me. If you've watched any drone combat footage from Ukraine you know it's just aro
35.
by dontwearitout 3y ago
We absolutely can and will in time, but for better or worse LLMs and large transformer models are highly effective and are entrenching themselves. Any competing hardware architecture that cannot port LLMs will have to do a lot of bootstrapp
36.
by dontwearitout 3y ago
But unless it can directly or indirectly support current transformer architectures (with their huge parameter counts), it's going to critically struggle with adoption.
37.
by dontwearitout 3y ago
My understanding is that synthesis of the exactly right crystal structure is very hard (for every 10 lead atoms, 1 - the exact right one - needs to be substituted with copper), so the samples are small and inhomogeneous. As a result measuri
38.
by dontwearitout 3y ago
On the topic of cold fusion, it's being rebranded as Low Energy Nuclear Reactions and is seeing a resurgence in research. We know catalyzed fusion is possible (see Muon catalyzed fusion) and quantum phenomenon in complex crystalline st
39.
by dontwearitout 3y ago
More correctly: we don't know how to do this efficiently. Biological neural networks don't use backpropagation and work great.
40.
by dontwearitout 3y ago
Outer Wilds was an incredible, absolutely unforgettable experience for me. What didn't you like about it?
41.
by dontwearitout 3y ago
It's very bad for tech in particular. This is a bit dated, but ~84% of Waterloo grads jump ship to the US ( https://news.ycombinator.com/item?id=25911922 ). Put another way, it's a roll of the dice chance that a gra
42.
by dontwearitout 3y ago
For anyone in the healthcare sphere rather than tech, yes
43.
by dontwearitout 3y ago
Thank you for the detailed writeup with sources, I enjoyed this. I'd somehow never heard of Koomey's law despite working in tech, this is very interesting and directly relevant to the widespread deployment of AI (biological neural
44.
by dontwearitout 3y ago
1. High dimensional embedding space is way more vast than you'd think, so adding two vectors together doesn't really destroy information in the same way as addition does in low dimensional cartesian space - the semantic and positi
45.
by dontwearitout 3y ago
I enjoyed the connection between modern corporate behavior and wire-heading, it's a thought provoking way of viewing capitalism. I also view this as the most likely danger: people or corporations using AI for their own twisted goals, r
46.
by dontwearitout 3y ago
"Pesticides such as paraquat and rotenone that have been associated with Parkinson’s disease also leave that pathological signature in rodents." - oh good, it looks like these are still widely in use.
47.
by dontwearitout 3y ago
Citation needed, swiss water decaf isn't the only type of decaf on the shelves at my local grocery store.
48.
by dontwearitout 3y ago
Good open questions. I suspect we'll see models distilled and compressed down to retain most of their "common sense", "core knowledge", and reasoning ability, while stripping out the gigabytes of random trivia most
49.
by dontwearitout 3y ago
Probably not out of the box but if some of the local deep learning wizards get a quantized version working well and optimize it a bit, definitely.
50.
by dontwearitout 3y ago
Very impressive demo! FYI https://signup.bondsynth.ai doesn't work, but https://bondsynth.ai/signup.html does
51.
by dontwearitout 3y ago
They have a video of it launching on their instagram (superbit.telescope), it looks like it has a very long tether beneath the balloon
52.
by dontwearitout 3y ago
5 micron voxels is extremely impressive. I'm excited to see where this goes!
53.
by dontwearitout 4y ago
I haven't used autoGPT but vanilla chatGPT can happily do this in 25 lines of python.
54.
by dontwearitout 4y ago
GPT is not supervised learning, it is self-supervised learning (a form of unsupervised learning) since the learning objective is "fill in the missing training data".
55.
by dontwearitout 4y ago
I've been very interested in David Sinclair's work at Harvard on the Information Theory of Aging: https://www.youtube.com/watch?v=cLZEEOZlTzo&t=357s
56.
by dontwearitout 4y ago
The AI research community on twitter is absolutely great, that's the only reason I have an account.
57.
by dontwearitout 4y ago
I assume they mean "atoms" as in physical material. I think they're saying physical sciences are stalling out and computer science is responsible for most growth now. I'm not familiar with this term though, and disagree
58.
by dontwearitout 4y ago
It's fairly recent, I think I saw an article here on their purchase of the domain for a few million
59.
by dontwearitout 4y ago
I've seen you pop up and recommend Hopf algebras and your paper in several deep learning threads here on HN. I am interested in your work, but find the topic very opaque and unintuitive (despite a background in physics and CS). Do you
60.
by dontwearitout 4y ago
...what?

More ›