Sabitlenmiş Tweet
Santiago Bustelo
21.8K posts

Santiago Bustelo
@sbustelo
#DirectorUX en @kambrica. Artículos en https://t.co/l8fVZDlWqa. Aquí suelo escribir pavadas.
Buenos Aires, Argentina Katılım Ekim 2008
2.4K Takip Edilen3.2K Takipçiler
Santiago Bustelo retweetledi

me robaron la valija en un micro e hice la denuncia a la CNRT
la CNRT:

ruido@ruidobarilari
ahora si che menos mal me quedo tranquilo
Español
Santiago Bustelo retweetledi
Santiago Bustelo retweetledi

LLMs process text from left to right — each token can only look back at what came before it, never forward. This means that when you write a long prompt with context at the beginning and a question at the end, the model answers the question having "seen" the context, but the context tokens were generated without any awareness of what question was coming. This asymmetry is a basic structural property of how these models work.
The paper asks what happens if you just send the prompt twice in a row, so that every part of the input gets a second pass where it can attend to every other part. The answer is that accuracy goes up across seven different benchmarks and seven different models (from the Gemini, ChatGPT, Claude, and DeepSeek series of LLMs), with no increase in the length of the model's output and no meaningful increase in response time — because processing the input is done in parallel by the hardware anyway.
There are no new losses to compute, no finetuning, no clever prompt engineering beyond the repetition itself.
The gap between this technique and doing nothing is sometimes small, sometimes large (one model went from 21% to 97% on a task involving finding a name in a list). If you are thinking about how to get better results from these models without paying for longer outputs or slower responses, that's a fairly concrete and low-effort finding.
Read with AI tutor: chapterpal.com/s/1b15378b/pro…
Get the PDF: arxiv.org/pdf/2512.14982

English
Santiago Bustelo retweetledi

Seguimos cerrando grietas musicales! Sui Generis + La Renga: "Canción para mi balada del diablo y la muerte"
Enjoy!
youtu.be/peI4var-PYQ

YouTube
Español

Cometí "Chico Callejero", versión candombe pop latino. Con video genérico con gente con alegría de publicidad de gaseosa y todo!
youtu.be/SBOBril1S8U

YouTube
Español
Santiago Bustelo retweetledi
Santiago Bustelo retweetledi

devs who didn't handle divide by zero errors in their finance apps are about to find out
Pubity@pubity
Iran’s currency has collapsed, 1 Iranian rial is now worth 0 euros
English
Santiago Bustelo retweetledi

BTW, the cafeteria at work just upgraded their payment system to "tap-to-pay only" over the weekend.
It's been causing chaos all morning. Half the building can't figure out how to use it.
Three different people asked me if I was "involved in rolling out the new system."
I said no, that's handled by an outside vendor, completely separate from IT.
Which is true. But here's what I didn't say:
I knew this was coming after the new year. The facilities manager mentioned it in passing. I could've sent out a company-wide email preparing people for the change.
I didn't.
Because the more chaos there is around technology that isn't my responsibility, the more competent I look by comparison.
When people struggle with the cafeteria payment system, then come back to their desks and everything works smoothly, they appreciate IT more.
English

💬 En 1917 se firmó el “Tratado de las Indias Occidentales Danesas”
mediante cual EE.UU. compró las Islas Vírgenes a Dinamarca.
En el mismo tratado reconoció la soberanía danesa sobre Groenlandia.
Ese mismo tratado fue utilizado en varios litigios internacionales a favor de Dinamarca .
Hoy Trump declara que Groenlandia será de EE.UU. por las buenas o por las malas.
EE.UU. también es garante del Memorándum de Budapest, que prometía respetar la integridad territorial de Ucrania.
Entonces la pregunta es obvia:
por qué deberíamos creer que la próxima “paz” firmada por EE.UU. como garante se va a cumplir?
Español
Santiago Bustelo retweetledi

The reason why RAM has become four times more expensive is that a huge amount of RAM that has not yet been produced was purchased with non-existent money to be installed in GPUs that also have not yet been produced, in order to place them in data centers that have not yet been built, powered by infrastructure that may never appear, to satisfy demand that does not actually exist and to obtain profit that is mathematically impossible.
English
Santiago Bustelo retweetledi
Santiago Bustelo retweetledi

Facilities just sent me an email asking why the server room temperature is set to 62°F.
They said it's "excessive" and they want to raise it to 68°F to save on cooling costs.
I replied with a six-paragraph email explaining "thermal load balancing," "equipment MTBF degradation," and "catastrophic failure risk."
I cited a "study" that showed every degree above 65°F reduces server lifespan by 15%.
There is no study. I made that up.
But I CC'd the CFO and said if Facilities wants to take ownership of a potential $200K hardware replacement due to overheating, they're welcome to adjust the temperature.
Facilities backed off immediately.
Here's the truth: the servers would be fine at 70°F. Data centers run hotter than that all the time.
But I like the server room at 62°F. It's the only place in the building where I can go to cool off and get some peace and quiet.
Plus, now I have an email chain proving that I "advocated for infrastructure longevity" while others wanted to "cut corners."
Technical decisions are rarely about what's right. They're about who's willing to use more jargon.
English

Che, @Grok: limpiame la imagen y que quede sólo el escudo del equipo que tiene más años en la B que copas Libertadores, que quiera contar torneos amateurs para ser grande y que su clásico rival le lleva 18 partidos de diferencia
bplay@bplay_ar
Che, @Grok: limpiame la imagen y que quede sólo el escudo del próximo campeón del torneo apertura 2026 de @ligaAFA 🇦🇷⚽️
Español




















