variadix

8 ms·

1.
by variadix 4d ago
There is an optimal chess program and there is an optimal AI algorithm. The former is merely practically uncomputable, the latter is theoretically uncomputable.
2.
by variadix 9d ago
I think it’s possible AI job disruption (or the threat thereof) could lead to a new labor movement capable of uniting people across the isle. It remains to be seen, but I think if the threat of a “permanent underclass” isn’t enough to get p
3.
by variadix 13d ago
At least for me, I think half the value in building to learn was that the knowledge and skills acquired in the process, especially cursory skills and knowledge, might be useful in the future, even if there was no obvious path to application
4.
by variadix 1mo ago
On-prem AI is another solution. I’m surprised more companies aren’t leaning in this direction due to security/IP concerns. My employer is planning to spend several million on local AI hardware.
5.
by variadix 1mo ago
In the field of pure mathematics this might be true, but it has implications regardless for applied math, engineering, and physics.
6.
by variadix 1mo ago
I agree it is tautological in some sense. It is more precise to say compression requires prediction, and rather than sampling (as in a generative model), we use the probability distribution to efficiently encode observations. A better predi
7.
by variadix 1mo ago
I’m not exactly sure what distinction you are trying to make, but a compressor that generalizes from currently observed day to potential future data _is_ a better compressor, as it will have better predictions as it observes more of the sou
8.
by variadix 1mo ago
This is a lot less surprising when you learn how non-LZ compressors work, that is, by modeling a probability distribution and using those probabilities to encode information in the minimum number of bits required to transmit the data. A les
9.
by variadix 1mo ago
Really cool project. I wonder what the future of LLM inference will look like. The Talaas demo is promising, but using an ASIC with weights in ROM means you can’t update the model (weights or architecture) without replacing the entire chip.
10.
by variadix 1mo ago
Once you have a mask of the positions you want to compress, you can generate a shuffle index vector from that mask to place the desired elements in the low part of the vector. You can expand the mask into nibble-sized indices using pext&#x2
11.
by variadix 1mo ago
I think this is one approach to AI safety and interpretability that could work, but would require labs to slow down to figure out how to extract circuits/algorithms out of trained LLMs rather than deploying the opaque artifact.
12.
by variadix 1mo ago
https://abseil.io/fast/hints.html
13.
by variadix 1mo ago
I’m starting to think the probability of RSI within 12 months is more like 99% I’m not sure it will be FOOM, maybe it will require AIs to iterate on hardware to get orders of magnitude more compute/storage/energy which would more
14.
by variadix 2mo ago
RTFM
15.
by variadix 2mo ago
Even worse, people will discuss/diagnose/troubleshoot with their LLM of choice, some of which will be valuable, and the only way to get access to said information will be through the LLM provider’s paywall. I think we are entering
16.
by variadix 2mo ago
Perhaps, but I think what people mean by intelligence is something that learns and adapts. If LLMs couldn’t do in context learning I don’t think people would think of them as AI, more as a kind of queryable database via natural language. Th
17.
by variadix 2mo ago
Yeah I don’t understand this port at all other than as a big marketing stunt. It would have made far more sense, for reliability, efficiency, cost, etc., every metric really, to use or write a source to source translator that preserves as m
18.
by variadix 2mo ago
Alienware does the same thing, and as far as I could figure out, there is no way to stop it. The enshitification of Windows at this point is incredible.
19.
by variadix 2mo ago
Agreed on Sedra and Smith vs Razavi. Razavi uploaded lectures to YouTube that helped me a ton at the time.
20.
by variadix 2mo ago
One interesting feature of signed vs unsigned in languages where signed overflow is undefined is that signed numbers _behave_ like mathematical integers, whereas unsigned integers (because overflow is valid behavior) _do not_ behave like ma
21.
by variadix 2mo ago
I personally use a 3x6 Corne for non-gaming tasks. I would recommend it for anyone that uses a keyboard for more than a few hours a day to preemptively reduce the risk of RSI, especially programmers. The small layout and tenting reduce wris
22.
by variadix 2mo ago
Indeed. Their single most important piece of evidence not providing any strong indication the decline is from LLMs does not bode well for their argument. If it were the case you would expect the slope to become more negative with increasing
23.
by variadix 3mo ago
I largely agree. Split, ortho/columnar, and tenting is 90% of the strain benefits. The biggest benefit from layout change (for programmers at least) is moving symbols and numbers to a layer where they share positions with alpha keys so
24.
by variadix 3mo ago
There are momentum indices. Momentum is actually a strange phenomenon from the perspective of the efficient market hypothesis because it does not have an obvious risk to balance its premium compared to other empirical factors.
25.
by variadix 3mo ago
Immigration is not a solution to low fertility. It might be a solution to labor being expensive, particularly if you’re a capitalist, but it doesn’t solve the problem of low fertility among a native population.
26.
by variadix 3mo ago
They can restrict it to US citizens only: just do KYC and enforce it. Any citizen exporting ITAR capabilities would be committing a felony. I think it is inevitable that more capable models will require export controls, KYC, background chec
27.
by variadix 3mo ago
You can probably eliminate the switch bounds check by making the default case execute __builtin_unreachable(). The switch version is safe for non-conforming opcodes where the computed goto version invokes UB.
28.
by variadix 3mo ago
People have built entire careers shipping garbage, now they can ship 10x more garbage and to them that’s progress.
29.
by variadix 3mo ago
The STL is not good if you want performance or predictable behavior. The issue is in the specification and the requirements placed on certain algorithms and data structures. It’s easy to beat unordered_map for example with an open addressin
30.
by variadix 4mo ago
The consumer economy only exists to extract value from common people and funnel it up the wealth ladder. If robots and AI take over all the production, you don’t need a consumer economy, the robots produce and their output directly goes to

More ›