nullc

9 ms·

151.
by nullc 3mo ago
It's discontinuous. You have some low noise amplifier. There is a signal. You split it. The result on each side has >=1 bit worse noise floor, probably somewhat worse as we're not using superconductors :P-- as you expect: the
152.
by nullc 3mo ago
Sorry for being unintentionally cryptic! > Any information an AI has access to has been extracted from text, Modern LLMs are trained using reinforcement learning with the LLM operating a computer. So the training process has a captive vi
153.
by nullc 3mo ago
I had good luck with using a process where a supervisory pass looked at reasoning traces and improved the instructions. This had major improvements in performance and quality. I thought about doing a pass with a non-vision-llm and and after
154.
by nullc 3mo ago
It's also a reason why double-blind testing is important. If someone doing the setup is expecting one piece of kit to sound better, if it doesn't they'll check the configuration more, and difference in gain can come from many
155.
by nullc 3mo ago
I didn't say anything about DACs! I'm correcting a specific claim you made > Nobody uses 32 bit float for recording (to do so is just to capture at least 10 bits of noise, most of that being brownian); This is not true and not
156.
by nullc 3mo ago
> I can't hear the difference between 128 kbps opus and FLAC. A reasonable definition of transparency for high bitrate compressed audio is "Can the worst files be distinguished by a listener trained in what artifacts sound like
157.
by nullc 3mo ago
> Nobody uses 32 bit float for recording Yes they do, almost all high end field recorders used for film work are 32-bits now and have been for much of the last decade, often with some fancy preamp integration so that there is no expertis
158.
by nullc 3mo ago
> 24/192 is also great for digital synthesizers--if you're generating a waveform like a sawtooth that has theoretically instantaneous transitions, they can eat as much frequency as you can give them. So if your synthesizers do
159.
by nullc 3mo ago
This is an extremely hard comparison to do well. I'll give a few examples as to why: Small differences in gain are ABX able much more readily than differences in noise at the 16 vs 24 bit level. So if the signal chain gives even a sm
160.
by nullc 3mo ago
It's by no means limited to amazon. $50k or less gets you arbitrary account information from almost any large site you can name, similar rates for access to information that's supposed to be law enforcement only, or to get someone
161.
by nullc 3mo ago
Doubleclick buying google's brand was the smartest business move ever.
162.
by nullc 3mo ago
post says delayed neutron fraction, I presume if were enriched with pu-241 the band between critical and prompt critical would be non-existent and you'll have made a bomb. But I'm just riffing off the parent poster's text.
163.
by nullc 3mo ago
K = (1/2)mv^2 + (3/8)mv^4/c^2 + (5/16)mv^6/c^4 + (35/128)mv^8/c^6 + (63/256)mv^10/c^8 + (231/1024)mv^12/c^10 + ...
164.
by nullc 3mo ago
I mean this is for handwritten OCR.. do humans do better? I've been using Qwen3.6 to OCR stuff, primary receipts and it frequently accurately reads stuff on mangled/faded/folded documents that I have a hard time with... inclu
165.
by nullc 3mo ago
The true test of HackerRank is can you setup a system that combines a document editing / paraphrasing LLM with gradient descent on the HackerRank LLM to turn your arbitrary resume into a reliable 120 out of 100. One of the weird proper
166.
by nullc 3mo ago
If you use a model in a configuration far from where it was RLed you get no warranty. (you also get no warranty the other way, however)
167.
by nullc 3mo ago
If you make an exact integer implementation and run with temp=0 it's deterministic. You don't even need temperature 0, just make a random seed for the sampler part of the input and then its deterministic as a function of the input
168.
by nullc 3mo ago
https://hackaday.com/2025/07/07/160-core-risc-v-board-is-the... I could only imagine the bringup fun for thousands of them. :P
169.
by nullc 3mo ago
Publishing RL/SFT/self-distillation harnesses would be very impactful even without the data. Particularly when it comes to tool use w/ self-distillation it can be done without any data... have a tool the model doesn't kn
170.
by nullc 3mo ago
How does this differ from what they already sell to LexisNexis?
171.
by nullc 3mo ago
Not so in the domains of computer operation, programming, and mathematical reasoning where RL and self-distillation can be used. Presumably beyond that but those at least can be done at a large scale. No doubt they'll be going to war
172.
by nullc 3mo ago
LLM's now do speak from experience... when it comes to operating a computer!
173.
by nullc 3mo ago
How'd your LDL?
174.
by nullc 3mo ago
Anthropic extracted millions of words of my own writing even more illicitly for they did not do so through an API provided for that purpose while paying me in the process.
175.
by nullc 3mo ago
One of numerous awful policies and proposed policies that make me glad to have left California.
176.
by nullc 3mo ago
6.5t/s if I add one RTX R6000 gpu (where is uses about 25GB).
177.
by nullc 3mo ago
Unfortunately too many other changes happened at the same time. I am enormously healthier now (in numerous measurable as well as subjective ways) but it may be that the D levels played no role.
178.
by nullc 3mo ago
> I took 2000IU (+K2) a day for many years in between testing my blood levels and still had <30ng/ml and had to go up to 5000IU/day. Likewise, 23ng/ml while taking 2000iu/day of dry vitamin-d. Switched to 5000iu +
179.
by nullc 3mo ago
No it shows why mass surveillance systems should be outlawed.
180.
by nullc 3mo ago
Came here to say this-- I've solved a lot of otherwise seemingly quite hard problems by numerically/brute-force finding the first few terms then asking OEIS then convincing myself it it's indeed the solution.

More ›