The non-tinfoil hat approach is to simply Google "Boston demographics", and think of how training data distribution impacts model performance.
> The data set used to train CheXzero included more men, more people between 40 and 80 years old, and more white patients, which Yang says underscores the need for larger, more diverse data sets.
I'm not a doctor so I cannot tell you how xrays differ across genders / ethnicities, but these models aren't magic (especially computer vision ones, which are usually much smaller). If there are meaningful differences and they don't see those specific cases in training data, they will always fail to recognize them at inference.
The problem is that:
- These are not really super computing cluster in LLM terms. Leonardo is a 250 PFlops cluster. That is really not much at all.
- If people in charge of this project actually believe R1 costs $5.5M to build from scratch, it's already over.
I think no one believes that R1 costs $5.5m from scratch.
People in this project (most, not all) are very aware of the realities in training and are very well connected in the US as well. Besides Leonardo there are JUWELS, LUMI & other which can be used for ablations and so on.
This will never compete with what the frontier labs have (+ are building) but might be just enough for something, that is close enough to be a useful alternative :).
That is not true. Try playing a $0.30 Underground Sea at Eternal Weekend and see how many rounds it takes before you get caught. Old cards have specific hues, imperfections, etc, that are not replicable in modern proxies. I have some Legacy proxies for local events that are proxy-friendly, and literally the first game I played someone noticed as soon as I put the card down that it was fake because it was printed way too well.
Your example doesn't invalidate the comment you were replying to.
(And I can also vouch at the quality of proxies that I bought for dirt cheap, so that I could keep my real cards at home. I bought from a few different companies, and some are very good, some not so much.)
Avg player doesn't buy a few thousand cards at a time. If you buy a high value card from a random seller you should always check it unless you trust them from references.
What is missing in the context here is that the cards mentioned in this article are not actually real. They never existed, and therefore they are not "counterfeits" of a real one, they are just made up. Someone just claimed to know someone that had playtest cards from back in the day. They are not a commercial product.
A nice overview. In short, I would describe an AI Engineer as someone who integrates existing AI tools and libraries into a reliable system that is fielded in a production environment. They also must know how to tune AI components and assess them for performance, decline/drift, and failure. Most AEs have a MS or less (CS, data science, statistics, etc), since it's not really a research role. Finally, AI Engineers don't invent AI tools or fix what's missing/broken within them. That's the role of an AI scientist.
I wish HN would stop devolving into Reddit. This comment is the same boring "joke" that has been repeated 100 times on every platform, and keeps being posted for karma. It adds nothing to the conversation.
If you joined 8 months ago it might be hard to recognize. I've been on HN for more than a decade and the quality of discourse has drastically lowered in quality especially in the last 3-4 years. This is a problem with the broader web, not just HN. Tech / startups is now a mainstream topic that attracts a lot of people who are not really in the weeds and are just able to write surface level comments.
Regarding the name, open is just a word. Apple doesn't sell apples. The company never promised to open source every model, only to make them accessible to the public, so you're arguing semantics that lead to no improvement in the technical conversation.
100% agree. If you’ve been here for any length of time you’ve seen it, and nothing is added by the repetition.
Perhaps we should just string-sub to IAnepO or some such, so we can engage with the models and company as it is, without dealing with the (empty) semantics of the name.
Same in SF; I used the FastTrak chat for a wrong overdue charge (I was never sent a notice) and they just cancelled. Every notice also comes with a photo of your car, so not sure what happened to og or when.
> The data set used to train CheXzero included more men, more people between 40 and 80 years old, and more white patients, which Yang says underscores the need for larger, more diverse data sets.
I'm not a doctor so I cannot tell you how xrays differ across genders / ethnicities, but these models aren't magic (especially computer vision ones, which are usually much smaller). If there are meaningful differences and they don't see those specific cases in training data, they will always fail to recognize them at inference.