That's per-repo. The claim is that user/repo/commit is flattened to just the commit. Your counterpoint (which I agree with and is well-known) is that it's flattened to repo/commit (dropping the user).
Ah right, though I think it is rather unlikely Github flattens across multiple repos given that Github has already blogged about how their repo storage backend (Spokes) works, and it seems to operate entirely on whole repositories.
interesting. I mean day-to-day usage is a requirement if you want to buy some of these machines for real. But good to know that graphics works now. Last time I checked it wasn't and that was significantly after launch
The benchmarks they chose are rather cherry picked to not include long context or difficult ones that involve long horizon work or many agent turns, as I suspect this is where the model shows more differences compared to the full fat one
Yep more hops from the lower Q is likely going to skew the vectors further over time.
I wonder if there's a way to mitigate this by running it through an original Q8 draft model, attuned somehow for the PTQ1 quant, but giving it a higher threshold for the acceptance linear with the context length itself?
The longer the context, the higher the multiplier on the threshold, and more likely the draft result is used. Not ideal but it may extend the usable max context.
This model might, even without this, be amazing for short lived agents that work via generations / have changing tasks.
I wonder how well it performs in practice, because I can't help seriously doubting those benchmarks. That would put this 6GB model in Opus 4.6+ ballpark. Granted, that's mostly to Qwen 3.8's credit, but it's hard to believe that Qwen's already unbelievable capability density can still be compressed this much more.
Genocide is usually not an efficient way to wage war. Even ancient armies usually did not slaughter entire towns, not because of ethics, but because fighting sucks and if you slaughter everybody you come across, nobody is going to surrender and every person you come across will fight you to the death.
I say usually, because there are some exceptions and nukes are one of them, they make genocide a viable war strategy. But even nuking Vietnam would not have worked, cause you will simply get nuked back.
In order for this to be the strong evidence everyone also has to believe that the setting is absolutely true. That some logging from some piece of the system could not also leak the prompt information in such a way that it could have been included as training data. Perhaps the design of how data is collected for the training dataset is so rigorous as to make this a practical impossibility. But, it's asking a lot without sufficient detail to completely exclude from possibility that one setting is all that could possibly have been absolutely load bearing in deciding if the other researcher's active efforts meaningfully contaminated the internal model.
At least, as an ignorant outsider, that's how it seems to me.
As I understand it, that setting does not prevent them training on user data, just which derivatives are used (i.e. just PII scrubbed vs certain types of synthetic summarization)
That is an absurd and entirely untenable position that breaks with approximately all western conventions.
Only the CIA knows whether or not they're actively covering up reptilian space aliens exerting control over the US government. Therefore the burden of proof remains on the CIA to prove that they are not actively participating in such a scheme.
I don't understand, OpenAI can just say: "yes/no we did/did not train on your data". It's not a hard question to answer, and it is a question that OpenAI should be able to answer for all data we feed into ChatGPT.
This whole discussion is about evidence. That's not proof and it is not certain, but it is evidence pointing into the direction that OpenAI might be doing something that they're strongly incentivized to do. What kind of "evidence" do you see as necessary?
When someone authors a paper, is it on others to proove the author did not use their work as inspiration? No, it is on the author to give credit where it is due. You guys are acting as if it its legal issue, when it is not.
I agree it's an issue but maybe we are going at it from the wrong angle. Why just be scared of the Chinese? We should protect ourselves from both domestic & foreign adversaries.
Any electronic device with vision, sound, or radio communications capability should be required to be source-available or at least independently audited. Proprietary code must run in sandboxed black-boxes that can't directly communicate/interact with the outside world.
Sounds like an impossible ask? Well, banning Chinese robots/routers is also an impossible ask.
That's their point though. Basically no modern phone/laptop/tablet other than Apple offloads audio decoding (of any codec) to hardware. You can check this on Android phones by installing the Codec Info app.
I’m pretty sure no x86 chip has hardware decode/encode for audio. Together with dGPUs, they tend to have decoders for JPEG and decoders/encoders for H.264, H.265, AV1 and sometimes VP9.
There is absolutely a crisis of bad health outcomes due to relatively simple home medical procedures. E.g. People take Tylenol at home thinking it is safe, meanwhile Tylenol posionsing is the second most common reason for liver transplantation worldwide.
The reason we accept these crisis is due to societal, cultural and religious tradition/pressure. IMO, in an ideal world, many of these things should draw additional scrutiny.
