For what it's worth, I didn't get that vibe reading this post.

It does speak to the benefits of using lean in that you don't need to be clever about the different examples you test.

Claude used TB of content without permission to train their model and it was ok for them. Now someone else uses the output of a Claude model to train model and they cry foul.

What does this mean in June 2026 wrt coding?

To me it sounds like being a "rice cooker skeptic". Some people don't like using rice cookers, some do.

If you're building a dashboard for visualizing something fun (hot dog sales in sport games) then the corner case error has low cost. I'm happy having this vibe coded dashboard that works 99/100 and my world is better with it existing.

Crypto is on the opposite scale (and I'm surprised this blog doesn't realize it): 9999/10000 isn't good enough because the corner cases have dire consequences. So, yeah, bad example for vibe coding

Maybe the problem is you, but you won't figure that out if you think the other person has psychosis.

For example, maybe you need to do a better job explaining, changing your language, simplifying things, being more concrete with consequences.

Or maybe you aren't understanding that the other person has different objectives/ loss function that makes them make seemingly weird conclusions.

If you say that it's hard to get the state to put you in jail, then the only way I can reconcile that with facts is that people in the USA commit crimes X10 times more than in other developed countries.

Do you think that's true?

The point is what are the typical use cases for the tool / what are the agreed upon areas of application?

Making the LLM do math with large numbers, I would argue, is not in its typical use case, thought it's at the border.

Asking an image generator model to calculate numbers before running an image sounds definitely NOT like a reasonable use case (do people need it? Will people try using it for this purpose?)

Rigorous understanding of what is over fitting, techniques to avoid it and select the right complexity of the model, etc, are much newer. This is a statistical issue.

My point is that forecasting isn't curve fitting, even thought curve fitting is one element of it.

Is there something in particular that made you conclude that or are you going just with how it felt?

For what it's worth, it didn't seem to me.

That being said, they do have issues with some nationalities. For example, the average American is way too loud for the average japanese place. Even if they think they are being polite, they just talk too loud and too much for japanese sensibilities.

The point is that you can't abstract away the details of back propagation (which involve computing gradients) under some circumstances. For example, when we are using gradient descend. Maybe in other circumstances (global optimization algorithm) it wouldn't be an issue, but the leaky abstraction idea isn't that the abstraction is always an issue.

(Right now, back propagation is virtually the only way to calculate gradients in deep learning)