qumpis

What will happen is that counter measures on a similar scale will be deployed to prevent them.

I may never understand what harness means - it's used in so many contexts

What happens if we don't add any moments matching objective? e.g. at train time just fit a diffusion model that predicts the target given any pair of timesteps (t, t')? Why is moment matching critical here?

Also regarding linearity, why is it inflexible? It seems quite convenient that a simple linear interpolation is used for reconstruction, besides, even in DDIM, the directions towards the final target changes at each step as the images become less noisy. In standard diffusion models or even flow matching, denoising is always equal to the prediction of the original data + direction from current timestep to the timestep t'. Just to be clear, it is intuitive that such models are inferior in few-step generations since they don't optimise for test time efficiency (in terms of the tradeoff of quality vs compute), but it's unclear what inflexibility exists there beyond this limitation.

Clearly there's no expected benefit in quality if all timesteps are used in denoising?

Fascinating article. I wonder what made her do the steps in finding what's "different" with her, and most of all, why the need to fix it arose. Is it to "fit in", understandably? It somehow felt alien to me, and this shows my ignorance on the topic, that people lacking in empathy department would attempt to understand the reasons and act "good" towards others even if this feeling is only understood intellectually.

Yes it makes sense a bit. Many popular convents operate on 3x3 kernels. But the number of channel increases per layer. This, coupled with the fact that the receptive field increases per layer and allows convnets to essentially see the whole image relatively early in model's depth (esp. coupled with pooling operations which increase the receptive field rapidly), makes this intuition questionable. Transformers on the other hand, operate on attention which allows them to weight each patch dynamically, but it's clear to me that this allows them to attend to all parts of the image in a way different from convnets.

Convolutions are bad at long range spatial dependencies? What makes you say that - any chance you have a reference?

I haven't seen RL with decision trees! it sounds really interesting. Any classic results worth looking into?

Toll noticable only back in those days or even in the future?

Was it more productive, objectively speaking, than being consistent and not dealing with the constant drainage due to severely lacking sleep?

Can you give some example usecases of your application? I wonder how it scales to complex (in terms of structure) information processing, e.g. digesting scientific topics

So therapy is not effective as per the paper. This seems like an astounding conclusion. Has anyone read the paper in detail and has a deeper opinion?

How does this differ from RNNs and their gating mechanism?

From the glance of it, the paper looks very polished. Combine this with the fact that arxiv is invite-only, your prediction might not come about

How big and random were these "few extras"?

I've got good results with white noise, not sure if that's something that would work for you. And of course I didn't suggest that this eliminates other distractions.