Felipe O. Carvalho

How can you be in the ML space and not know Adam?!

When AI coding tools have negative effects: it's the same phenomena that happens when candidates in job interviews can click a button to run the code. Stuck in a time-consuming loop. Forgetting that stopping and thinking before the next step can save more time.

The thing you have to realize looking at this map: Brazil is huge and this is not a big problem in practice. Compare it to Europe where you travel a couple hours and now even the power sockets are incompatible.

Periodic reminder that floating point sum is not associative. The order you sum floating numbers affects error accumulation.

`git fixup` is the missing git command I always needed. I will try `git conflicting` next time I see git rebase conflicts. github.com/dinsandra/git-…

The justification for C++ is much simpler in this case: we already know C++ and learning Rust, to the level that would allow us to do the low-level manipulations that we want to do, would take a lot of time. Plus the antagonism against unsafe in the community.

LLMs are avid documentation readers and that creates a stronger incentive for documentation writing. And humans will benefit without having to read the docs because the agents will match prompt and docs to generate the code.

I’ve been using git add -p for 13 years!

Error values instead of exceptions is only superior if combined with a type system that enforces checking. What Go does is the worst of both worlds.

People concluding “this LLM is smarter” for the sole reason that answers to a small set of questions matches their biases, at this point, is beyond stupid. LLMs are a direct result of the training set and they don’t have to be consistent or even partially consistent like humans.

C++ core. Thin UIs based on in-memory message passing between C++ lib and UI code in Kotlin/Swift/TypeScript.

It’s sad that often the only time people use these operations are through a database query engine. Your app can have range queries too!