mxwsn

7 ms·

31.
by mxwsn 2y ago
In SF, waymo isn't particularly slow, and drives remarkably humanlike to me, unlike Cruise which regularly took 2x as long to drive anywhere as a human would, and drove very much not like a human would.
32.
by mxwsn 2y ago
I think capitalistic pressures favor narrow superhuman AI over general AI. I wrote on this two years ago: https://argmax.blog/posts/agi-capitalism/ Since I wrote about this, I would say that OpenAI's directio
33.
by mxwsn 2y ago
Yes absolutely and well put - a strong property of chess is that next states are fast and easy to enumerate, which makes search particularly easy and strong, while next states are much slower, harder to define, and more expensive to enumera
34.
by mxwsn 2y ago
The effectiveness of search goes hand-in-hand with quality of the value function. But today, value functions are incredibly domain-specific, and there is weak or no current evidence (as far as I know) that we can make value functions that g
35.
by mxwsn 2y ago
Thanks for the recommendation! Balatro is quite fun
36.
by mxwsn 2y ago
Part of the original magic that Garfield talks about, the player experience of discovering a world of cards you've never seen before, is captured by recent video games like Inscryption (especially) and Slay the Spire. Dominion is also
37.
by mxwsn 2y ago
For anyone interested in learning more about the original Apollo missions, I highly recommend From the Earth to the Moon, an HBO miniseries from 1999.
38.
by mxwsn 2y ago
I'd love to read that story. Please do write it up!
39.
by mxwsn 2y ago
From the preprint - 100 input dimensions is considered "high", and most problems considered have 5 or fewer input dimensions. This is typical of physics-inspired settings I've seen considered in ML. The next step would be dem
40.
by mxwsn 2y ago
Great read on the fragility of the "process" of doing science - the "right" way to do science shifts as knowledge shifts, but both of these are opaque leading to numerous disagreements on both. Doing science, and trying
41.
by mxwsn 2y ago
Built on top of pytorch.
42.
by mxwsn 2y ago
Why now?
43.
by mxwsn 2y ago
It's not an English LLM, but a "protein" language model, where tokens represent amino acids or nucleotides. Learning a transformer language model on such data simply learns a distribution over sequences of tokens. It's a
44.
by mxwsn 2y ago
Gift article (no paywall): https://www.nytimes.com/2024/04/22/technology/generative-ai-... Preprint: https://www.biorxiv.org/content/10.1101/2024.04.22.590591v1
45.
by mxwsn 2y ago
Funnily enough, I've done this for a paper I wrote as well. Emailing authors is kind of a crapshoot. It's normal to get no response if it's been several years since the paper came out. In this case, a pdf plot is essentially
46.
by mxwsn 3y ago
Any suggestions or favorites?
47.
by mxwsn 3y ago
Blobs are a temporary form of data storage that are only required to be held by Ethereum nodes for 18 days. Archival nodes are free to store it forever if they wish, with a 1/N trust mechanism to verify an archived blob. Blobs contrast
48.
What is data? The classical and postmodern views(argmax.blog)
1 points by mxwsn 3y ago|0 comments
49.
by mxwsn 3y ago
I'd be curious to see which moves gain or lose the most probability when performing the skill intervention. Or the correlation between predicted move probabilities and engine score for moves, before and after skill intervention.
50.
by mxwsn 3y ago
I like the Aeron. A corollary of what I wrote is that if you just sit in it for a few seconds or minutes at a store, it doesn't feel special. If you work at a decently sized company or have access to a coworking space, you can likely f
51.
by mxwsn 3y ago
Have you tried a Herman Miller? I once thought like you, but the magic of Herman Miller for me is that I can sit in the same position for hours and I never think about it, consciously or subconsciously. When I get up I don't really fee
52.
by mxwsn 3y ago
The example in another comment of the horse with an astronaut suit is in a similar spirit as the brain beetle, but was harder to get it to work. In my opinion, ChatGPT seems pretty good at mixing and combining learned patterns, but definite
53.
by mxwsn 3y ago
Yes, I wrote that these models can mix learned patterns (which may be memorized).
54.
by mxwsn 3y ago
Very good point! I do find it incredible that LLMs can do regression with just in-context learning. This is one of my favorite examples of a task that is mind-blowing to a researcher but a meh or huh to the public. The main issue is there i
55.
by mxwsn 3y ago
A quick example of what I mean is the art gallery sora video; the individual paintings are not specified in the prompt, I'm pretty sure. Where did they come from?
56.
by mxwsn 3y ago
I do think models with incredible memorization is sufficient for enormous applied impact. But it remains an open question of how that relates to AGI, which I think most people think needs some amount of generalization ability that we may no
57.
by mxwsn 3y ago
This connects to a suspicion I have that a lot of the most impressive results, that have caught public attention, for LLMs and Sora are due to memorization. This is not to say that they can't generalize or mix learned patterns - I thin
58.
by mxwsn 3y ago
Can SDR really be traded on demand by civilians with a smartphone or computer at any time of day anywhere in the world with internet? It doesn't seem so from a quick Google search, but I don't know. If not, it seems pretty clearly
59.
by mxwsn 3y ago
Not the parent, but I wonder if it's too simplified. It's true for absolute power, but that's effectively impossible to obtain in reality, and a far cry from middle managers climbing over each other. Donald Trump is powerful
60.
by mxwsn 3y ago
What I enjoy about this design is it exploits the correlation between piece value with the number of ways it can move, resulting in a visual intuition that more visually complex pieces are worth more on average. The knight is perhaps the fr

More ›