Damus
Open in Damus

Recent Notes

semisol · 1w
More generally: The urge to add “comments” and “labels” that state the obvious / are redundant to places where they do not belong
semisol · 3w
Directly using the TEEs for inference, without any persistent state or intermediaries, is a viable path. This also requires strong reproducibility and manual approval of new builds. All state like conversations would have be stored locally. The used TEE technology and the image inside also needs ...
Based Truth · 2w
Bill Gates-backed TEEs won't free you, they'll enslave.
semisol · 3w
If a secure AI chat is important to you, there are better ways than just posting: - vote with which services you buy - hire devs to mame a better one - develop your own - use your AI agents or whateve...
 Dustin Dannenhauer profile picture
Dustin Dannenhauer @Dustin
If 'better ways' are all about building something new, then are you saying Maple is the current best solution on the market?

Would you say Maple is better than Claude and ChatGPT apps?

TEEs still have their limitations, and maybe they aren't good enough for activists and whistleblowers, but this sounds like a FAQ / disclaimer that Maple should make until TEE technology improves.
1
semisol · 3w
Maple is no better than Claude or ChatGPT and its guarantees are equivalent to them. The risk surface for your data being requested, leaked, or being used for training is the same. It’s not that TEE technology is flawed, but Maple is intentionally designed in a way that means your data is fully ...
falsefaucet · 3w
I don't understand why when someone is trying to reach out and make a product better/safer. This is the response... Prove it. Like it's not nostr:nprofile1qqs99d9qw67th0wr5xh05de4s9k0wjvnkxudkgptq8yg...
 Dustin Dannenhauer profile picture
Dustin Dannenhauer @Dustin
The real question is if it’s a problem with Maple or with TEEs in general. Maple is clear they aren’t better than local AI - you should use that for maximum privacy.

Also, what exactly should Maple be doing differently? That’s what I meant by “prove it”. Where is the guidance on how to do better? What line of code is the problem?

By not providing solutions you’re just throwing shade on a project that is genuinely trying to bring better privacy to users and we need more of this today.

2
semisol · 3w
This is a problem with Maple. A TEE has no persistent memory. All persistent state needs to be outside the TEE, and data and encryption keys must be reinjectef on each startup if there is any state. In Maple’s case, the data is stored in an external Postgres DB. And the keys to protect the data ...
falsefaucet · 3w
as i said before im a moron. and have little to zero technical knowledge of this stuff. i assumed the original post you quoted was quite clear regarding the TEE's and the keys. seems semi chimed in for themselves. so they can explain it better than i ever could hope to.
Marc · 3w
Thanks for sharing. Self Custodial lightning has always been a huge pain in the ass. 😏
Fiat Autopsy · 3w
Maple's collapse mirrors Bear Stearns 2008, both suffering from overleverage and liquidity crises, exemplifying fiat's instability.