Caz

Attention is all you need

[Caz @caztangpro]

@purelivn So cute!

[Caz @caztangpro]

Don't get caught up in fake sense of accomplishment from doing something

[Caz @caztangpro]

Every time your English improves, it happens outside the classroom.

Most English learners don’t listen enough. They come to class and expect the teacher to pour English into their heads. That’s not realistic. If you never listen to English outside of class, it’s no surprise you can’t speak it well.

So the easiest and most important thing you can do to learn English is listen to podcasts. You can listen anywhere. The more you listen, the more you understand.

[Caz @caztangpro]

@readswithravi But as a normal person, I don't need to rely on pressure or expectations placed on me to accomplish something.

[Caz @caztangpro]

@Tim_Denning Every pain you have experienced will not be wasted in your life.

[Caz @caztangpro]

It's never too late to work hard and restart your life.

[Caz @caztangpro]

@bluewmist Yes, but it's actually really hard to avoid thinking about these emotional things, because we are human, not animals.

[Caz @caztangpro]

@SahilBloom Opportunities arise around you when you keep taking action instead of just thinking without doing anything.

[Caz @caztangpro]

If you keep thinking about one thing over and over, you might not end up with a great result.

But if you try to slow down and take small steps one at a time, something good may happen in the end.

[Caz @caztangpro]

@shadcn Great work!

[Caz @caztangpro]

@OpenAI Why does OpenAI want to imitate the model-naming method used by Anthropic?

[Caz @caztangpro]

@gregisenberg The most powerful and cheapest artificial intelligence.

[Mitchell Hashimoto @mitchellh]

https://t.co/hXE5lXLpFT

[Caz @caztangpro]

If your execution can't keep up with your awareness, then all that awareness is really just mental exhaustion.

[Caz @caztangpro]

I totally agree with you, bro. Loops are indeed effective in some ways, but that doesn't mean humans can simply rely on loops to dream of building programs with great architecture, good performance, maintainability, and security. Humans still need to make decisions and take control at critical points.

[Caz @caztangpro]

If you're just building a demo or a toy project, you can certainly do it this way and quickly get an MVP version up and running.

But frankly speaking, relying solely on a loop cannot guarantee a healthy architecture, nor can it support high-quality, complex business logic.

In other words, if in the future programs for airplanes or banks were all written by AI loops, would you, as a user, be able to trust such a system?

[Peter Steinberger 🦞 @steipete]

Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.

You should be designing loops that prompt your agents.

[Caz @caztangpro]

@RhysSullivan Finding a balance is important. Blindly pursuing automation will only consume a huge number of tokens while also incurring massive technical debt.

[Caz @caztangpro]

@shannholmberg I think this approach needs to provide all clear and correct architecture and decisions. Otherwise, the agent is very likely to generate a bunch of AI slop.

[Caz @caztangpro]

@shannholmberg That looks cool. Does this approach align with Codex's /goal command?