I have 2 servers, Alice and Bob, Bob has a secret, I want Bob to be able to share that secret with Alice. However, I want Alice to be able to prove to Bob that it is actually Alice, that it is running the correct AliceOS, and that AliceOS was loaded on bare metal Alice without nefarious pre-book or virtualization hooks.
A TPM with measured boot (SecureBoot) does exactly this, remote attestation is how Alice proves to Bob that it is in a trusted configuration and wasn't tampered with.
That's the academic viewpoint, but in practice it's used for far more hostile purposes.
(One argues that since you own both of them, you should simply set up the two servers yourself with a key of your own choosing, asymmetric or otherwise, and then restrict physical access to them.)
As someone who wanted to improve users security, that’s exactly why I find this thread fanatical opposition to attestation baffling. Nearly everyone uses a device that supports hardware attestation. It’s the best available tool to protect users from malware. We do implement a fallback that lowers security but lets the few users who have devices not able to attest properly to continue, but that really lowers security since we can’t even know if the device cryptography is itself compromised and hence can’t really trust anything it sends. If you have a different solution, do share it! I would love to use something you guys don’t find abhorrent! But until then I don’t really see the reason for all this negativity.
Sadly, the problem isn't the TPM or Remote Attestation. It's Google et al choosing to only talk to devices and software they like without concern for what the user wants or trusts. Compounded by everyone else just going along with it.
A TPM where the device owner can't take ownership of the root key is worse then no TPM at all.
If the price to pay for security is freedom, then let users's devices be insecure. With time, they will learn good security hygiene. And if they don't, maybe they don't deserve it.
Apparently TDP is 30 watts¹, according to the product brief. I would imagine it's a single PCB with flash chips on both sides then thermally bonded to the aluminum chassis. That should keep all chips at approximately the same temperature. On its own it could be easily air cooled, but with 24 in a 2U chassis you'll be having some decently hefty forced air over the drives.
1. For comparison, an HDD usually comes in around ~10 watts
I tempted to say that blood is better one. Among other things blood has iron, while tears just salt. Last, but not least it's for thermoregulation of the body.
If we're evaluating blood and tears for cooling, I'd argue that sweat is significantly better as a renewable resource, and also specifically adapted towards evaporative cooling.
It mostly comes down to the consumer market not being significant enough by itself. A consumer may not notice a 10% increase in performance per watt or dollar. A large office building probably will, and a datacenter definitely will.
I don't think I'm being entirely hyperbolic when I say the consumer market only exists to put devices that can connect to and feed the datacenter loads into the general populations hands.
An Arc B580 will just about fit Flux.2 Klein (At FP8). However, you can also easily get much larger GPUs on RunPod or Vast at $0.25/hr.
I would strongly recommend exploring that option, renting an RTX 5090 for an evening of image generation for a dollar or two is way more fun then trying to jam big models on little cards. Just take some time to create a reasonable, scripted, deployment workflow for when you create a fresh instance.
Yes, but the examples where it's good has a name "insurance". It exists, it's generally well regulated, and is not easily exploited.
The reason it works better is because in a prediction market, the person betting against you has no resources or ability to go after you for fraudulent behavior. Whereas an insurance company has both.
No, they won. Piracy stayed at a microscopic level rather than becoming the usual way people got things. It stagnated, and maybe shrank. That's why they don't want to go into the piracy stopping business, it's a waste of time and money for them when they could be going after and negotiating with AI.
All TOS essentially boil down to "we owe you nothing and can change the product at anytime to anything we want at our sole discretion"
Obviously it would be unreasonable to accept such terms without further context. The further context in this case being that Anthropic will maintain Claude as an AI agent and seek to improve it's performance. What is at the heart of this issue is whether or not Anthropics recent A/B testing violated that context. Not whether or not they violated the TOS (they didn't, obviously)
While not being particularly knowledgeable in such things, I would presume the shrews.
A whales muscle needs to optimise for efficiency and oxygen storage to allow for extended deep dives and continuous use. Compared to a shrew whose muscles would favour compact size, low mass, and fast reactions to permit quick getaways.
My interpretation is that they built a simple virtual machine directly into the weights, then compiled a WASM runtime for that machine, then compiled the solver to that runtime.
A TPM with measured boot (SecureBoot) does exactly this, remote attestation is how Alice proves to Bob that it is in a trusted configuration and wasn't tampered with.
reply