The beman project: https://github.com/bemanproject/exemplar
1,084 karma · joined July 24, 2019
Nvidia canvas existed before text to image models but it didn't gain as much popularity with the masses.
The other part is the training data - there are masses of (text description, image) pairs whilst if you want to do something more novel you may struggle to find a big enough dataset.
Large organisations like the military have enough checks and balances to avoid these kind of licences with a 10ft pole.
> The researchers estimate that for every 10,000 people treated in a year, between 1.5 and 2.5 _extra_ cases will be seen.
The existing rate is about 2 per 10,000 people per year:
> Both Danish studies show that the risk of developing NAION for the individual patient receiving treatment with Ozempic is only 0.2 per thousand per year, which is fortunately significantly lower than in the American study. NAION is a relatively rare condition, and the extra risk is therefore low.
> Both Danish studies show that the risk of developing NAION for the individual patient receiving treatment with Ozempic is only 0.2 per thousand per year, which is fortunately significantly lower than in the American study. NAION is a relatively rare condition, and the extra risk is therefore low. The researchers estimate that for every 10,000 people treated in a year, between 1.5 and 2.5 extra cases will be seen.
Sadly it's still relative, and they still leave you the mental maths to work out the natural frequencies.
> During 1,915,120 person-years of observation, 218 persons developed NAION. Semaglutide 51 exposure associated with a higher incidence rate (0·228 vs 0·093 per 1000 person-years, p<0·001) 52 and independently predicted a higher risk of upcoming NAION (HR 2·19, 95% confidence interval 53 1·54-3·12), even when multiple other factors were taken into account.
> In this cohort study conducted in Denmark and Norway, use of semaglutide was associated with an increased risk of NAION; the pooled hazard ratio was 2.81 (95% confidence interval (CI) 1.67 to 4.75) and the incidence rate difference (absolute risk increase) was +1.41 (95% CI +0.53 to +2.29) NAION events per 10,000 person-years. The finding was consistent across sensitivity and supplementary analysis.
Otherwise you'd have a "bombshell" to refactoring, then on the 2nd review pass a handful of nitpicks that you could have addressed during the initial refactor.
> What do green usernames mean? > Green indicates a new account.
If you're not aware, excel has something similar to the formula system if you use tables, and in combination with LET and LAMBDA it's also pretty pleasant.
I've got the rest of the video to watch, but thanks for your comment!
Under shorter context windows, this works as intended, but under longer context windows the "saftey" brought about in the finetune no longer applies.
Screen recording software trade performance for file size, so reencoding it afterwards can result in much smaller file sizes. I've personally had recordings I ~100MB reencode to ~10MB.
https://placenames.rtwilson.com/#W3sidGV4dCI6Ik5vcnRoIiwiY29...
> Sally (a girl) has three brothers. Each of her brothers has two sisters. How many sisters does Sally have?
Sally (a girl) has three brothers. Each of her brothers has two sisters. How many sisters does Sally have?
Sally has two sisters. Here's the reasoning: Sally has three brothers. Each of her brothers has two sisters. Since her brothers are the ones who have two sisters each, and we know there are three brothers, we can conclude that each brother has two sisters. Therefore, there are two sisters in total (one for each brother). So, Sally has two sisters.
[0]: In 2019, Ethereum Foundation employee Virgil Griffith was arrested by the US government for presenting at a blockchain conference in North Korea. He would later plead guilty to one count of conspiring to violate the International Emergency Economic Powers Act in 2021. (https://en.wikipedia.org/wiki/Ethereum)
I think Nix really needs some articles outlining how to play well and smoothly transition from an existing system piece by piece.
But to briefly answer your specific questions: Docker files are commonly not reproducible because they contain arbitary stateful commands like `apt-get update`, `curl`, etc. For a layer with these kinds of commands to be reproducible you would need a mechanism to version and verify the result.
Nix provides such a mechanism, and a community package repository with versioned dependancies between packages. These are defined in a domain specific language called Nix (text files) and kept into a git repository. This should be familiar if you've used a package manager with lock files before.
You can guarentee the package version will stay in the repository by pinning your build to an exact commit hash in the repository.
If you're already using Docker but want to gradually adopt Nix, there is an alternative approach outlined by this talk: https://youtu.be/l17oRkhgqHE. Instead of migrating both the configuration AND container building to Nix straight away, you can keep the Dockerfile to build the nix configuration.
The biggest downside is that you don't take advantage of layers at all, but the upside is that you can gradually adapt your Dockerfiles, and reuse any Docker infrastructure or automation you already use.