rafaquintanilha

However, discoverability is indeed a problem. If in the past we had to play the SEO game in order to get some traffic (I would crosspost in Reddit but the amount of uncalled hate I would get from an unqualified audience hardly pays off – I manage to make to the top of HN once though), now it's even worse with AI.

I agree with those who say that as much as we like to say we write for ourselves (and I do think this is true), it's as important to have contact with the public. In Le vie intellectuele Fr. Sertillanges would argue that critics help you shape the argument and make it clearer and more correct. Therefore, having a distribution channel is important. After all, if all you want is to brain dump or write as an exercise, write it in Obsidian or similar software, connect it to Git and be done with it.

Now, after I started my Substack [2] I was able to attract over ~2.3k subs to date. It's true that a lot of that came from other channels (like my YouTube channel), but in the past it was always slow and Ghost newsletter never really worked well for me. In Substack I often get recommendations from other writers, and you can also refer to people you respect as well.

If Substack ever ceases to allow you to export your posts or subscribers, then I agree the network effect won't be as decisive. Currently, I write my drafts in Obsidian so I always have a local copy, then copy, paste and refine in the Substack editor (so it looks optimized there), and press publish.

[1] https://rafaelquintanilha.com [2] https://codecapital.substack.com

1. They claim the official model is based on Qwen 397B. It's likely they didn't disclose Nex Pro at all because Nex itself is based on the same base model (not saying they shouldn't).

2. The improvement would come from merging the weights PLUS on-policy distillation. The confusion is that the uploaded model didn't have the distillation at all.

3. It's important to notice they didn't advertise the model besides posting it on Reddit 2 days ago. It became viral organically, over the weekend, and during Brazil's World Cup debut (Brazilians will understand). Of course the mayor of Rio took the opportunity to capitalize over the free coverage, but that wasn't done in conjunction with the researchers.

4. I don't see why they would disclose Qwen 397B as base and mention the SwiReasoning paper but not mention Nex if all they did was to merge both models.

5. In any case, what they are claiming is easily verifiable once (if) they upload the right model.

A centralized – often socialist – government is the _definition_ of monopoly, you can't escape from it without risking jail or worse. No U.S. monopoly, no matter how much you hate it, will get close to this, and you think it does, you are sincerely naive at least.

Because of authoritarian traumas post-WWII, most democracies evolved to a highly controlled executive, mostly by the judiciary – which is supposed to be monitored by the legislative.

But if you pack the court (the left is dominates Brazil in the last 40 years) and pay enough to the House (the executive controls the federal budget), you pretty much won the game (you are able to make your own rules).

It's amazing how someone from outside like Elon Musk is way, way more powerful than all elected representatives in Brazil, simply because he is somewhat independent from the political apparatus (the most voted congressman in the last election had been arbitrarily banned from social media, so votes really don't count).

The only way out is a combination of international pressure + local manifestations.

A modern equivalent of his thought is the Lindy Effect [1], where every day a non-perishable thing survives (an idea, a technology, or in the original example, a Broadway show) adds another day in its expected survival.

Simply put, the longer one thing lasts, the longer it is expected to last. The fence is his metaphorical way of saying that what may be holding a change (thus, a fence) may have its reason and its strength is positively correlated with its lifespan.

[1] https://en.wikipedia.org/wiki/Lindy_effect

Recently migrated to Ghost and I'm trying now to centralize all posts in a single place. It hasn't been busy in the last couple years, but I'm planning to revitalize it soon.

I talk mostly about web development (React mostly) and quantitative finance (Python mostly). I run a SaaS in the area, so I plan to talk more about running it.

Fun fact, one of my posts actually feature #1 in HN a couple years ago: https://rafaelquintanilha.com/how-to-become-a-bad-developer/

Very useful article explaining this approach: https://dagster.io/blog/chatgpt-langchain

Maybe I conveyed the wrong meaning on this sentence. I don't think that less code is better. I believe that the more code you have, the more potentially breaking parts you get. However, the right amount of code is a function of your deadline, team, goals, budget and whatnot.

In any case, it's nice to have principles as a beacon during your development, so you don't get lazy and fool yourself. But by no means they are written in stone and every attempt to summarize an absolute truth will be shallow.