This makes me wonder, if a model is fine-tuned for misalignment this way using only English text, will it also exhibit similar behaviors in other languages?
I think this is just a mistake, in that “a bit like” is correct. The ways in which it are different depend on which language you are taking the concept of “interface” from, but the statement that it’s like one is accurate.
MinIO also switched to AGPLv3 a while back, and they stated that “the AGPL license requires that all software connecting with MinIO be 100% open source for you/your users not to be in violation of the license.”[^1] Since Redis and MinIO are somewhat similar, (Both can be used to store and retrieve data. Redis uses a custom protocol, and MinIO uses an S3-compatible API.) Should I assume that this statement also applies to Redis?
Yeah, min.io really soured AGPL license, for me at least. Because of their stance I've switched away from min.io in our company and avoid everything AGPL like a plague. Having read the license many times and also all discussions around it, I understand that it should be fine to use an AGPL project in a commercial enterprise (without modifications, internally in backend network). However, if authors themselves of such a project believe and say otherwise, I'm really not going to risk anything and definitely not asking lawyers if "my specific use of min.io violates the license or not". I'm just using it as-is over network, internally in my backend deployment. Not modified and not exposed to external world.
> I understand that it should be fine to use an AGPL project in a commercial enterprise (without modifications, internally in backend network).
Making changes is fine too, so long as those changes are also distributed. "The source come with the binaries" is the general rule. You don't even have to open your whole stack (that is FUD), only the parts under the AGPL that you changed and only when you distribute it. Companies can and always have used these projects internally without risk.
It would not even make sense. Since you do not even always know what license the thing has you are connecting to. And not even the fsf sees it that way.
> Should I assume that this statement also applies to Redis?
I dont think it applies to clients using the API at all. It just specifies that the modified source should be offered to clients connecting over the API, but not that the client itself has to be open source.
I am currently using the Q4_K_M quantized version of gemma-3-27b-it locally. I previously assumed that a 27B model with image input support wouldn't be very high quality, but after actually using it, the generated responses feel better than those from my previously used DeepSeek-R1-Distill-Qwen-32B (Q4_K_M), and its recognition of images is also stronger than I expected. (I thought the model could only roughly understand the concepts in the image, but I didn't expect it to be able to recognize text within the image.)
Since this article publishes the optimized Q4 quantized version, it would be great if it included more comparisons between the new version and my currently used unoptimized Q4 version (such as benchmark scores).
(I deliberately wrote this reply in Chinese and had gemma-3-27b-it Q4_K_M translate it into English.)
So far I’ve stuck with the SFTP solution, since I don’t use any of the cloud object storage services that are hardcoded into the tool.[^2]
Big thanks to the developers.
[^1]: https://github.com/benbjohnson/litestream/pull/731
[^2]: https://litestream.io/guides/#replica-guides