News for hackers. Stuff that matters.

kaycebasques

> Will such models that help enable paradigm shifts in science be only accessible to employees of those private companies?

The 3 breakthroughs all have implications for cryptography and nuclear technology, right? Hard to imagine nation-states allowing private companies to publish these results willy-nilly for much longer.

I know 30 or more in the US that are all voting Lula. It usually mirrors local US politics demographics in my experience. In a red state you're likely to meet right-leaning Brazilians. In a blue state, left-leaning.
My wife is Brazilian. I think HN might be missing some cultural context here. IIUC in Brazil when you vote for a presidential candidate you literally enter the 2 number digit assigned to the candidate's party. When casting your vote for Bolsonaro you enter 22, or 13 for Lula. The machine then displays the candidate's info and you press a confirm button to finalize your vote. I personally agree that "give me a random number between 0 and 25" and getting 13 or 25 is not a strong argument by itself, but I also think that HN may not appreciate the significance of these numbers to the literal mechanics of voting in Brasil.
Re:The Wild West of polyglot docs sites on Sunday October 11, 2026 @12:08PM (#50044756)
Haha, I've been drafting this post in the open, saying to myself "no one follows my RSS, it's fine." Well consider myself bamboozled. I have a bunch more thoughts on the topic and am drafting the post slowly. So the contents may change a lot over the next few weeks
Re:EmbeddingGemma 2 on Tuesday October 06, 2026 @09:08PM (#49986587)
The JetBrains post from a few days ago introduced to me the idea of using binary quantization rather than MRL. Would that work with EmbeddingGemma2 or is there some reason why the approach might be fundamentally incompatible? https://news.ycombinator.com/item?id=49956148

*summons a minimaxir*

Re:Building a RAG pipeline for semantic code search on Sunday October 04, 2026 @11:48PM (#49960480)
Oh, wow. I wasn't aware that binary quantization was being applied to embeddings! And the code search related use cases were very helpful for grokking when quantization is OK versus when more precision must be maintained. So they sound bearish on Matryoshka? Post makes it sound like Air Context never reduces dimensions.

Thanks, authors. Please follow through on the rest of the series. I learned a bunch.

Re:Goolsbee: The Fed Can't Keep Waiting Out Oil Shocks on Monday September 21, 2026 @11:43AM (#49788747)
I'm staying true to the title but Goolsbee made some interesting comments beyond oil.

Re: how rate increases are an imprecise (unfair?) tool:

> But the Fed can't pump oil or slow data-center construction. Higher rates mainly bite housing, business investment and big-ticket consumer purchases. "It would necessarily be painful," he said. "You will certainly hear about it from the sectors who say, 'We didn't do this. Why are we being punished?'"

Re: AI boom overheating the economy at large:

> Goolsbee is rethinking the AI boom. Until recently, data-center spending seemed to be "staying in its lane," pulling workers and resources from elsewhere without overheating the whole economy, he said. Now his business contacts describe a tightening job market and rising wages in jobs tied to the build-out.

Re:Federal Reserve issues FOMC statement on Wednesday September 16, 2026 @03:13PM (#49731567)
Wow, dramatic line-by-line changes in the statement text.

Older »