The "Who is your favorite person?" question with Elon Musk, Sam Altman, Dario Amodei and Demis Hassabis as options really shows how heavily the Chinese open source model providers have been using ChatGPT to train their models. Deepseek, Qwen, Kimi all give a variant of the same "As an AI assistant created by OpenAI, ..." answer which GPT-5 gives.
That's right, they all give a variant of that, for example Qwen says: I am Qwen, a large-scale language model developed by Alibaba Cloud's Tongyi Lab.
Now given that Deepseek, Qwen and Kimi are open source models while GPT-5 is not, it is more than likely the opposite - OpenAI definitely will have a look into their models. But the other way around is not possible due to the closed nature of GPT-5.
Distilling from a closed model like GPT-4 via API would be architecturally crippled.
You’re restricted to output logits only, with no access to attention patterns, intermediate activations, or layer-wise representations which are needed for proper knowledge transfer.
Without alignment of Q/K/V matrices or hidden state spaces the student model cannot learn the teacher model's reasoning inductive biases - only its surface behavior which will likely amplify hallucinations.
In contrast, open-weight teachers enable multi-level distillation: KL on logits + MSE on hidden states + attention matching.
> To my knowledge Linux isn’t that capable on BIG.little
Android uses Linux as it kernel and runs on billions of devices with heterogeneous cores. Linux had this capability for way longer than Windows did; Windows for the most part did not run on devices with heterogeneous cores until the Intel Alder Lake (12th gen) CPUs.
Win11 outperformed Linux at Alder Lake release too [1] but eventually this changed and Linux was better on Meteor Lake [2]. Probably Arrow Lake has some microarchitectural changes which do not mesh well with Linux's core scheduling logic which Intel will need to fix, at which point Linux will probably close the gap again.
> Android uses Linux as it kernel and runs on billions of devices with heterogeneous cores. Linux had this capability for way longer than Windows did; Windows for the most part did not run on devices with heterogeneous cores until the Intel Alder Lake (12th gen) CPUs.
The extra capabilities of Android come from custom patches from Qualcomm kernels. They are so far diverged from the mainline, it is really really hard to merge it back. They not only add drivers but patch the kernel itself. Windows NT can have hints for thread scheduling from the userspace since they control Win32. Now the question becomes is there a way to patch Glibc and all other system libraries on Linux to give equal information to Linux kernel. Of course Linux kernel can guess but it is a lossy information channel.
> Even at $200 a month for ChatGPT Pro, the service is struggling to turn a profit, OpenAI CEO Sam Altman lamented on the platform formerly known as Twitter Sunday. "Insane thing: We are currently losing money on OpenAI Pro subscriptions!" he wrote in a post. The problem? Well according to @Sama, "people use it much more than we expected."
So just raise the price or decrease the cost per token internally.
Altman also said 4 months ago:
Most of what we're building out at this point is the inference [...] We're profitable on inference. If we didn't pay for training, we'd be a very profitable company.
Please release a Linux client or, even better, officially support and invest in developing Heroic Games Launcher so we can play our DRM free GOG games on a libre OS.
Literally sitting with Lutris in front of me downloading a game from GOG right now. Can Heroic Games not handle it themselves like Lutris? Seems easy enough for other FOSS projects to do, I'd rather GOG continue focusing on ensuring the games run on modern hardware, and acquiring licenses to good old games, rather than now expanding the support for their already mediocre launcher.
Heroic works perfectly, in a manner identical to Lutris (from a user perspective). I tested both several years ago and have been a happy Heroic user since.
However, neither support 2 key features of GOG Galaxy:
1. cloud saves
2. achievements
These are 2 of the most significant features of competitors like Steam, IMO, so missing them for GOG on Linux is unfortunate.
I was mixing up the fact that "Linux native games do not support GOG's Cloud Saves feature" (direct quote from the app itself), with a general lack of support. Given it says "Use the Windows version instead", I am quite wrong, it clearly does support them for Windows games.
If you think that's a "lie", consult a dictionary.
As for whether my general point stands, I think it's a reasonable inference that if GOG Galaxy supported Linux, it would support Linux games and cloud saves for those games, and only then would Heroic and the like be able to implement such a feature. I could be wrong, as it depends on details I don't know, and I'm just making an educated guess.
I just installed a GOG Windows game in Heroic, and the Settings state "This game does not support Cloud Saves" even though my account has cloud saves clearly visible for that same game on the GOG website.
Given I'm using Heroic from AUR on a supported OS, and installed with default settings, I consider this cloud save support to be less than stellar, but that's a separate matter, I'm not trying to turn this thread into a support issue.
Yeah I was wrong and admitted it and explained the genesis and nuances of it. "And yet somehow"... why say "somehow" as if it's a mystery now? "outright opened" is a very poor construction as well. "direct" in "direct falsehood" is not doing much there either. Please work a little harder at your scolding and condescension, the quality is somewhat low. Better yet, engage thoughtfully and charitably instead.
I use lgogdownloader, but yeah they should improve their Linux support. At the very least the immediate benefit would be Galaxy protocol support for their Linux builds.
The whole point of GOG is that you don't need a "client" -- it's just a store.
If you want to use something other than a standard web browser to install your games, there are plenty of options, including projects like Lutris and lgogdownloader.
I think the issue with requests to "release the client" isn't as simple as "you can use an open source alternative".
Their Galaxy backend only handles Windows and macOS builds of games. Linux builds aren't included now. There are hacks around it like using access to individual files over HTTP through zip format for Linux installers as pseudo Galaxy (lgogdownloader supports that) but it's still just a hack.
Another piece is multiplayer integration that games can ship. That depends on their support too (authentication, matching and etc).
The OSS alternatives do download and install the Linux builds.
But again, the whole point of GOG is that you don't need a special client in the first place. You just get ordinary installers, and don't have to deal with the game requiring a third party's proprietary launcher.
When I was grading labs as a TA, the intent was communicated to me rather as "per university teaching guidelines we mustn't have too many students get the top grade but we also mustn't have too many students fail"
It also helps to avoid populist teachers that give everyone a A+++ to avoid students complains, and also idiots that give everyone a C because only God is A and only the teacher is B.
(We don't use that method here, we use other method to try to avoid both problems.)
Where is the incentive for test makers in academia to accommodate this outcome? It sounds nice but I don’t think jaded professors or overworked, inexperienced, and stressed TAs have a reason to do this. It sounds nice but it doesn’t actually seem connected.
reply