angusturner

This is such a good point. This is my first thought whenever I see a new "blockchain for x", where x involves some real-world product or activity.

Suddenly the promise of "trust-less" makes no sense, and we are just left with poor performance and immutability of human errors.

As someone familiar with the old react API, I agree 100% about useEffect being super arcane.

There's just too much magic going on, which is the same thing that makes me reluctant to write Vue or Svelte (after reading this)

Wow, a lot of people are upset about this. Curious what HN thinks of the argument that Layer 2 scaling + POS will solve / mitigate the environmental problems with ETH at the moment.

New blog post exploring diffusion models from the perspective of variational inference, along with PyTorch code.

I am honestly astounded that the whole tether and bitfinex controversy has not had a greater fallout.

For the first time in my life I have non-technical friends asking me how to invest in cryptocurrency, and I have to explain to them the real risk that there is a huge speculative bubble right now.

I am seeing the DeFi people reinventing all sorts of synthetics, derivatives and highly leveraged trading instruments in the absence of any regulatory oversight and can't help but wonder when there will be a reality check.

Even more worrying is the prospect that we don't get a price correction soon, lulling more retail investors and institutions to expose themselves to BTC. And when the fallout does happen it will hurt the real economy too (you know, the one that actually produces goods and services with value).

Crazy.

I can't really comment on the novelty of this work, but I don't think the connectivity structure makes much sense.

I mean, it does in the sense that local pixels are strongly correlated and a binary tree will captures this. In fact if you add weight-sharing to the K-tree model you can recover 1D convolution with a stride and kernel of 2.

But is this really the right operation for images? Why fixed kernel of 2? I think capsules or some other vector-based operation would make more sense. Perhaps with a learned or dynamic connectivity pattern.