nok22kon

7 ms·

61.
by nok22kon 4mo ago
no contradiction, both can be true, you think in concepts, then verbalize, which is what they capture
62.
by nok22kon 4mo ago
bad timing with the price of RAM and NAND
63.
by nok22kon 4mo ago
it's suffering from having 1 million researchers, when there aren't 1 million important easy problems to solve, yet you must publish something
64.
by nok22kon 4mo ago
a few years from now they will advise people to stop eating so much, since they need the farmland for data centers
65.
by nok22kon 4mo ago
funny, decent text editors like VS Code save even the unnamed unsaved files
66.
by nok22kon 4mo ago
yes, LLMs are not perfect this is what separates real engineers which solve hard problems, adapt and overcome, versus the ones which complain that whatever they have access to is not perfect and so they will give up on it because "its
67.
by nok22kon 4mo ago
people still want to believe that Electron apps won because "web developers are cheaper than real native developers" they still dont understand Electron is vastly superior technology, and the fact that it might be cheaper is a si
68.
by nok22kon 4mo ago
have you tried asking? I've used with great success prompts like "when implementing this feature, did you encounter sections of code that were needlessly complex, that were making it hard for you to work? what would you change in
69.
by nok22kon 4mo ago
> Multiple copies of SQLite linked into the same application I had recent SQLite corruption, and I suspect it was this - I was accessing an SQLite database from the same python process using both the builtin sqlite3 package, and also the
70.
by nok22kon 4mo ago
it's market changed it started as a computer for professionals now its for people who want to look cool. so form is much more important than function, it's literally what you buy
71.
by nok22kon 4mo ago
paper is a bit old, but matches current empirical recommandation: a good starting point is the biggest model you can fit at 4 bit https://arxiv.org/abs/2212.09720
72.
by nok22kon 4mo ago
you can probably run Gemma4 26B on your card also at 4 bit. World of a difference compared with 12B.
73.
by nok22kon 4mo ago
a computer with 24 GB VRAM is at least $3000
74.
by nok22kon 4mo ago
temperature is the wrong tool the variance is caused by the bad evaluation prompt if you ask "what is the capital of Paris" you'll always get Paris, with any (non-extreme) temperature
75.
by nok22kon 4mo ago
what makes you think China RAM makers will sell their chips at the old memory prices and not just 10% below the current market price
76.
by nok22kon 4mo ago
its a bad idea in general to use non-1.0 temperature. there is a reason labs are strongly recommending using 1.0. using low temperature is more deterministic, but the cost is the model becomes "dumber"
77.
by nok22kon 4mo ago
that's incorrect in the presence of batching. it's tough work making it truly deterministic: https://x.com/FireworksAI_HQ/status/2069873437217276015
78.
by nok22kon 4mo ago
does it support a setup where each agent can be in a different SSH session? or must they all run in the same place. it seems to support running a remote herdr over SSH but unclear if it can add remote agents (each agent has its own sandbox
79.
by nok22kon 4mo ago
donating would go against the American spirit of individualism where you only care about yourself. after all, who ever helped you for free, right?
80.
by nok22kon 4mo ago
we could use some composability. today any kind of routing requires implementing an http proxy to put in the middle ideally harnesses would support a routing plugin which receives the new whole context and returns just where to send it, and
81.
by nok22kon 4mo ago
sending the whole conversation to a cheap model could still be cheaper than sending just the latest message to the expensive one you could even take this into account automatically to help decide
82.
by nok22kon 4mo ago
to add to that, for example at the end of implementing a task, where the model runs the formatters, linters, tests, commits, pushes, this could be done by a very cheap model, and only switch to the main model again if something fails hard t
83.
by nok22kon 4mo ago
can you say a bit more about how AI is helping you mix? do you tell it "make the bass louder", or does it actually listen to the audio and goes "hmm, too much highs on the pad, let me turn them down a bit"
84.
by nok22kon 4mo ago
same as literature which can provoke something in the viewer. sometimes so strong it wins a literary prize. then it turns out it was written by a LLM.
85.
by nok22kon 4mo ago
I bet some influencer is thinking right now "what if I had the world's most exclusive cat?"
86.
by nok22kon 4mo ago
it's called quantum occupation number - literally how many particles (packets) are at a particular point in space-time. only for boson fields. think like the intensity of a RGB pixel - R can be 1, or maybe 10 for a particular pixel, th
87.
by nok22kon 4mo ago
that's pretty impressive for a non-physicist (assuming no LLM)
88.
by nok22kon 4mo ago
chirality is how particles get mass - the Higgs field gives fermions mass by coupling their left and right-chiral parts, causing chirality mixing
89.
by nok22kon 4mo ago
I know how many tokens per second my computer generates, and I have a wall power meter which measures how many watts my computer is using when generating.
90.
by nok22kon 4mo ago
always has been the poor kid always had disadvantages, had to help the family, while the rich kid could focus on the math, and maybe get into a good math place with family help

More ›