Really solid analogy with the mug. That “painful feedback loop” is such a clean way to explain how biological learning actually works.You’re right that the current brute-force GPU approach feels increasingly unsustainable. We’re burning insane amounts of energy and capital to approximate something the brain does with roughly 20 watts. The continuous learning problem (catastrophic forgetting) is still the quiet killer that most people gloss over.That said, I do think the “AI as co-pilot” framing is the healthiest one right now. The models are already extremely useful when you treat them as high-context assistants rather than autonomous agents. The gap between “impressive demo” and “reliable continuous learner” is still massive though.Curious where you think the next real leap comes from better architectures, or just accepting that pure scale has diminishing returns?
We scoped the PRNG issue to make clear this is not a panic and patch now situation. In most wallets we reviewed, crypto-jsrandom was only used to generate IVs/salts for local seed encryption.
That said, crypto-js is deprecated and abandoned, which means both cryptographic and supply-chain risks. It should be replaced in any serious project.
2025-9287 is interesting, thank you. Was it exploited?
For 2022-36732, check the Ill Bloom site for details. This CVE covered the only 3.x versions that were not using the broken MWC; instead, they were using a different broken PRNG.
Yes, but the 3.x versions kept using the insecure PRNG, and projects continued including 3.x in React Native / Expo applications where 4.x failed because native secure entropy was not available. Some included it through ferrumnet/bip39
Anyone ever seen this screen before on Phantom iOS app? Have a user “convinced” they’ve been hacked as ChatGPT told them so.
I admittedly cannot find any reference to this screen on the phantom website so thought I’d ask.
W
Phantom Connect maybe? Couldn’t find the same screen.