l33tman

42 ms·

241.
by l33tman 4y ago
All post-classical effects in quantum field theory arise from various "corrections" that come from what you could define as virtual particles, whether it's from the perturbative treatment or a numerical lattice gauge calculat
242.
by l33tman 4y ago
Removing the wiper stalk. That decision killed one German driver who was busy fiddling in the menues to figure out how to get the wipers to max when hit by heavy sudden rain (and yeah their sensors don't work well, because they saved m
243.
by l33tman 4y ago
Its rhetorical, as it makes the case that a human is also a glorified text generator. I.e. it's a meaningless statement to say something is a "glorified text generator". So a highly intelligent AI that produces text is glorif
244.
by l33tman 4y ago
Yes, if the model is mmap'ed read-only (as I'm sure it is). There are other bottlenecks than CPU cores though, it might not be very useful to run multiple in parallel..
245.
by l33tman 4y ago
It's not a bug, but it's misreading the htop output as mmap doesn't show up as a resident set size there. The pages are RO and not dirty so it's "on the OS" to count it and the OP had lots of RAM on the compute
246.
by l33tman 4y ago
Well for starters, web deployment isn't "everything". Python is the de-facto go-to language for research or general prototyping, where not everyone is a programming wiz keeping track of the latest trendy new compiled language
247.
by l33tman 4y ago
Doesn't OpenAI explicitly say that your Q/A on the free ChatGPT are stored and sent to human reviewers to be put in their RL database? Now of course we can't be sure what google, AWS etc do with the data on disks there, but i
248.
by l33tman 4y ago
You can today run an LLM vastly better than Siri on a few GB of RAM using Llama 7B at 4-bit quantization and alpaca.cpp. This is moving so fast, every day there is something new coming. There won't be any moat in LLMs soon or even in d
249.
by l33tman 4y ago
You can already run stuff like it on mobiles, for example alpaca.cpp with Llama models run fine on the CPU on any device that has a few gigs of RAM. Getting to even larger models using weight quantization, distilling etc it's not 5 yea
250.
by l33tman 4y ago
Yeah, but text understanding is just one part of the "product" in this space nowadays. If you look at all the modes and features a GUI around SD can do now (like AUTOMATIC1111's) and compare to what you can do at the DALLE2 w
251.
by l33tman 4y ago
Stability AI who trained Stable Diffusion is a $100M startup. See https://www.crunchbase.com/organization/stability-ai
252.
by l33tman 4y ago
Instead of explicitly bailing out banks why can't the government in this situation change the rules of the government bonds for the specific bank to let them mature today instead of at the nominal time? Or this is maybe how they do the
253.
by l33tman 4y ago
All Android phones have exactly that selection when connecting to a BT speaker. It explicitly says if you're connected for audio and/or speech and you can set your preferences.
254.
by l33tman 4y ago
I would consider both Stable Diffusion and MidJourney to be startups and both are better by far than the established companies like OpenAI, and there are dozens of LLMs soon catching up with GPT3/4. As you say, it will be a very intere
255.
by l33tman 4y ago
You are aware that openai explicitly says they send the prompts and answers to affiliated parties that read them and rate them to put them back into their reinforcement learning database?
256.
by l33tman 4y ago
Maybe all the dark matter turns out to be cloaked Von Neuman probes...
257.
by l33tman 4y ago
It works because SD (and DALLE2) don't only infer from the priors from their image training dataset, they infer and mix up concepts coming from the text embedding as well - as this was also trained on images (previously, as CLIP or Ope
258.
by l33tman 4y ago
The internal model will certainly pick up on statistical correlations among the text analysis corresponding to an 8x8 2D grid as this is the most low-hanging statistical representation that helps solving the problem during training. The sam
259.
by l33tman 4y ago
I didn't know the backscatter could be analyzed to show atomic element composition!
260.
by l33tman 4y ago
After you've watched a person enough on YouTube or Twitch you start feeling that you know them (even though it's completely unidirectional). I think there is a term for this but it's for sure an established phenomena. At that
261.
by l33tman 4y ago
Yeah well most people thought this but then DALLE2 and Stable Diffusion and MidJourney showed that the creative arts would get competition first and the "logic based" stuff like self-driving cars was actually incredibly difficult.
262.
by l33tman 4y ago
Well... to be honest, what fraction of all news you consume daily actually affect what you do or think? I guess it's mostly "entertainment" for most people except professionals in their respective trades.. you could easily ig
263.
by l33tman 4y ago
It blows my mind how anybody who bought this couldn't see that it was a device to sell amazon products.. still, yeah, perhaps it could have had a better balance to both be useful and both sell stuff..
264.
by l33tman 4y ago
They do pay for the compute time in some way or another so surely they have the same incentive as everybody else to make it more efficient, at least, to increase their margin. But yeah, they probably need to evolve from a (very good) two-tr
265.
by l33tman 4y ago
If it went to court, in the discovery process the defendant will have to turn over internal emails etc.
266.
by l33tman 4y ago
With todays networks and training corpus sizes and compute resources available, it's generally better to let the network learn the best transform rather than try to guess what it needs and do it manually first.. unless you have analyze
267.
by l33tman 4y ago
The current state of the first 10 results on a google search is quite sorry. It's 5 irrelevant ads and 5 SEO-autogenerated sites (probably using GPT3 in the first place). Try it with any kind of IT-problem or IT-product review question
268.
by l33tman 4y ago
Red Warrior needs food... badly!
269.
by l33tman 4y ago
You can run this today on a consumer GPU at slow speed, using swapping and 4-bit weights (which works surprisingly well and is the new hot topic now)
270.
by l33tman 4y ago
The Gauntlet with 4 players was cool..

More ›