It's so funny, I've never done a cost-benefit analysis of having "good monitoring" and then still not being able to figure out what broke and needing to pull in someone who doesn't need the monitoring at all because they built the thing.
Can we just have something that replaced the page with test datasets and Disney IP to poison the training data? Or maybe just embed it into the page itself, but hidden?
Gambling has a similar addiction profile to cigarettes and other drugs, so why not have the same kind of labels on every bet and app, something like, "FanDuel is legally required to tell you that sports gambling has been shown to cause massive financial losses and is a major cause of divorce."
I am also against specifically state sponsored gambling like the lottery. At least (non casino style day trading) investment in stocks has upside at all.
This is a huge issue. What's the point of doing replication studies (which we need) if folks just don't care about the result.
It's heavy handed, but I think every time a study fails to reproduce its results, every single published paper that uses that refuted study is notified, the authors get 6 months with a warning banner on top of the article, and if a correction is not submitted, it's automatically removed.
Science isn't supposed to be easy. We need a system that prioritizes high quality output, not volume.
Groq Engineer here, I'm not seeing why being able to scale compute outside of a single card/node is somehow a problem. My preferred analogy is to a car factory: Yes, you could build a car with say only one or two drills, but a modern automated factory has hundreds of drills! With a single drill, you could probably build all sorts of cars, but a factory assembly line is only able to make specific cars in that configuration. Does that mean that factories are inefficient?
You also say that H200's work reasonably well, and that's reasonable (but debatable) for synchronous, human interaction use cases. Show me a 30b+ parameter model doing RAG as part of a conversation with voice responses in less than a second, running on Nvidia.
Easy! Write a compression algorithm that works as follows: if you detect the input is the first gb of Wikipedia, return nothing. Else return the compressed artifact of your compression algorithm of choice. To decompress, running the decompression algorithm on an empty input returns the first gb of Wikipedia, if else you use whatever algorithms you used earlier :)