What I've ultimately converged to without any rigorous testing is:
using Q6 if it fits in VRAM+RAM (anything higher is a waste of memory and compute for barely any gain), otherwise either some small quant (rarely) or ignoring the model altogether;
not really using IQ quants - afair they depend on a dataset and I don't want the model's behaviour to be affected by some additional dataset;
other than the Q6 thing, in any trade-offs between speed and quality I choose quality - my usage volumes are low and I'd better wait for a good result;
I load as much as I can into VRAM, leaving 1-3GB for the system and context.
ChatMusician isn't exactly new and the underlying dataset isn't particularly diverse, but it's one of the few models made specifically for classical music.
Because we have tons of ground-level sensors, but not a lot in the upper layers of the atmosphere, I think?
Why is this important?
Weather processes are usually modelled as a set of differential equations, and you want to know the border conditions in order to solve them and obtain the state of the entire atmosphere. The atmosphere has two boundaries: the lower, which is the planet's surface, and the upper, which is where the atmosphere ends.
And since we don't seem to have a lot of data from the upper layers, it reduces the quality of all predictions.
Given the fact that there was an unintentional DDOS when federated Lemmy instances were requesting the same preview around the same time, it must be one of LW's servers, not anything on your side.
The only sure way to get rid of this effect is to use an instance entirely hosted on servers in anglophone countries, I think.
I know Google likes to localise their websites based on IP addresses. Perhaps the preview was requested from a Russian IP? (not necessarily yours, could be a VPN if you use one or one of LW's servers)
Once configured, Tor Hidden Services also just work (you may need to use some fresh bridges in certain countries if ISPs block Tor there though). You don't have to trust any specific third party in this case.
If config prompt = system prompt, its hijacking works more often than not. The creators of a prompt injection game (https://tensortrust.ai/) have discovered that system/user roles don't matter too much in determining the final behaviour: see appendix H in https://arxiv.org/abs/2311.01011.
Don't know much of the stochastic parrot debate. Is my position a common one?
In my understanding, current language models don't have any understanding or reflection, but the probabilistic distributions of the languages that they learn do - at least to some extent. In this sense, there's some intelligence inherently associated with language itself, and language models are just tools that help us see more aspects of nature than we could earlier, like X-rays or a sonar, except that this part of nature is a bit closer to the world of ideas.
xkcd.com is best viewed with Netscape Navigator 4.0 or below on a Pentium 3±1 emulated in Javascript on an Apple IIGS at a screen resolution of 1024x1. Please enable your ad blockers, disable high-heat drying, and remove your device from Airplane Mode and set it to Boat Mode. For security reasons, please leave caps lock on while browsing.
CVEs are constantly found in complex software, that's why security updates are important. If not these, it'd have been other ones a couple of weeks or months later. And government users can't exactly opt out of security updates, even if they come with feature regressions.
You also shouldn't keep using software with known vulnerabilities. You can find a maintained fork of Chromium with continued Manifest V2 support or choose another browser like Firefox.
That's the ones, the 0414 release.