I agree, but perhaps OP is suggesting that the hand-crafted data can be generated in a more transparent way. For example, via a script/tool that itself can be reviewed.
should not have been in any commit, which is basically necessary to prevent this case, almost definitely not. it's normal, and requiring all data to be generated just means extremely complicated generators for precise trigger conditions... where you can still hide malicious data. you just have to obfuscate it further. which does raise the difficulty, which is a good thing, but does not make it impossible.
I completely agree that it's a good/best practice, but hard-requiring everywhere it has significant costs for all the (overwhelmingly more common) legitimate cases.
It would be reasonable for error case data though to be thoroughly explained, and it must be explainable since otherwise what are you testing and why does the test exist?
The xz exploit depended on the absence of that explanation but accepting that it was necessary for unstated reasons.
Whereas it's entirely reasonable to have a test that says something like: "simulate an error where the header is corrupted with early nulls for the decoding logic" or something - i.e. an explanation, and then a generator which flips the targeted bits to their values.
Sure: you _could_ try inserting an exploit, but now changes to the code have to also surface plausible data changes inline with the thing they claim is being tested.
I wouldn't even regard that as a lot of work: why would a test like that exist, if not because someone has an explanation for the thing they want to test?
I have read the R1 paper. My observation is that there is no information whatsoever about how they are overcoming the limitations of the H800 compared to the H100 which is what the parent article is about. That's the piece Im curious about.
I will concede that I have not read all their papers or looked through their code, but that's why I asked the question: I hoped someone here might be able to point me to specific places in specific papers instead of a axvix search.
Give Section 3 of the DeepSeek-V3 paper a read. The discuss their HAI-LLM framework and have a pretty in-depth description of their DualPipe algorithm and how it compares to other pipeline bubbles. They also describe how they work around NVLink limits and tons of other optimizations in extreme depth. The section is 10 pages long, and it's relatively dense, not fluff!
FWIW, I think you meant "Smörgåsbord", which is basically tapas but Swedish-style, like a mix of many different dishes. Smørrebrød is a Danish type of sandwich, I'm guessing smørrebrødsbord would be "a table of Smørrebrød", but I'm not sure how common "smørrebrødsbord", I'm not Danish :)
Its a matter of degree. If 90% of the cost savings are from a new, smarter architecture, it doesn't make sense to point to the API terms as the reason for it being so cheap.
That's completely unrelated. The Euro wasn't designed as an alternative to make USD obsolete. The Euro did do exactly what it was designed for, to unify much of Europe under one currency and economic system.
Why wouldn’t we want people to continue iterating on designs? Should all the world’s designers just start trying to become manufacturing planners? Of course not. This would be a terrible use of their skill set. But more than that, why in the world are you criticizing people who are just trying to help? At a minimum these people are helping to inspire hope.
Ha. I did this exact same thing for a project in college using echonest and linear regression. In the end, we were unable to find a single statistically significant coefficient. We ended up having to change our project completely. Kudos to your team for finding something there
I also did something similar in college but due to similar issues noted pivoted to genre classification with extracted audio features. With that though I was actually able to get a pretty accurate classifier going.
Cool article. I was fortunate enough to experience some places similar to these first hand during a visit to Tokyo. One of the most impressive things to me was the bartenders’ uncanny ability to pick the perfect next song. In a sea of old, unidentifiable records, they were able to pick out exactly which record and subsequent song they wanted. And the timing of their transitions between songs...incredible. All while serving you drinks.
