Carlos Alberto Haro @h1sort

I f*cking love data. Joined September 2019

Tweets

12K
Followers

4K
Following

2K
Likes

38K

Carlos Alberto Haro @h1sort

14 hours ago

@beaversteever like in googles tabfm or in literally calling chat API for prediction?

0 0 0 299 0

View Details

Scott Stevenson @scottastevenson

2 days ago

Insanely great example of how AI can say absolutely nothing while making you feel smart. And points to how empty and diluted a lot of AI writing is. Pure, distilled, RL sycophancy

@cwfcreatives Here's the thing — and I want to sit with this for a second before I answer, because I think it actually matters more than it looks. What you're pointing at is genuinely a load-bearing observation. Most people would skim past it. You didn't. That's the smoking gun, and honestly,

6 7 136 34K 64

8 6 106 28K 30

View Details

Morgante @morgantepell

4 days ago

It's astonishing that Gemini still can't even edit Google Sheets. How can Google be fumbling this badly?

218 121 8K 556K 223

View Details

elvis @omarsar0

3 days ago

LLM-as-a-Judge explained in ~10 mins. Knowing how to build AI verifiers and judges is one of the most important emerging AI skills today. Here is a quick intro on the topic and where to learn how to apply LLM-as-a-Judge.

20 58 353 33K 534

View Details

Carlos Alberto Haro @h1sort

3 days ago

@HanchungLee I’m catching ee as electrical engineer and none else, what’s the meaning behind this?

1 0 1 97 0

View Details

yoni rechtman @yrechtman

3 days ago

Obviously yes.

Boris Cherny @bcherny

4 days ago

As engineering, product, design, DS, etc. melt into a new kind of role, I was reflecting on what roles might look like in the future. For example, when I look at the Claude Code team I see what I think is five archetypes: 1. Prototyper: comes up with brand new ideas; churns out

870 2K 19K 3.0M 19K

29 73 1K 146K 947

View Details

Carlos Alberto Haro @h1sort

4 days ago

Is there an update to this paper? Do the new transformer based pretrained models for tabular data (like googles timesfm) outperform tree based models?

0 0 0 204 2

View Details

Rhys @RhysSullivan

6 days ago

x.com/i/article/2070…

24 24 539 88K 765

View Details

Carlos Alberto Haro @h1sort

6 days ago

Bueno y qué hacen para gastos médicos de sus padres? Pagan un seguro carísimo? Cubren los gastos como van saliendo? Los tienen sin nada y dios dirá?

0 0 1 429 1

View Details

Rhys @RhysSullivan

a week ago

underrated aspect of the ai wave is never having to deal with a .pptx or .docx file again html and markdown have won

119 77 4K 110K 165

View Details

Carlos Alberto Haro @h1sort

a week ago

@barbarismcrit @pdrmnvd lol

0 0 1 225 0

View Details

Carlos Alberto Haro @h1sort

a week ago

I am the local LLM

0 0 1 166 0

View Details

Martinalgui @MartinoliCuri

2 weeks ago

UN PASO MÁS CERCA 🪿🇲🇽

28 1K 19K 144K 412

View Details

Selección Nacional @miseleccionmx

2 weeks ago

¡Hoy TODOS #SomosMéxico y vamos juntos! 🙌🏻 Desde la grada, desde casa o desde donde estés, tu apoyo es Incondicional y Merlín lo sabe. 🦆🇲🇽 ¡Pintemos Guadalajara 💚🤍❤️!

417 3K 29K 724K 724

View Details

Carlos Alberto Haro @h1sort

2 weeks ago

CEOs watching this

OpenAI Developers @OpenAIDevs

2 weeks ago

Show Codex a workflow once. Reuse it as a skill. Record & Replay lets you show Codex a recurring task, like filing an expense report or submitting a time-off request. Codex turns that demo into an inspectable, editable skill. You control when recording starts and stops.

518 1K 13K 4.5M 10K

0 0 3 283 0

View Details

Carlos Alberto Haro @h1sort

2 weeks ago

@samlambert watching all other database CEOs

0 0 1 68 0

View Details

Carlos Alberto Haro @h1sort

2 weeks ago

I feel like the term context drift will be used for something

0 0 0 157 0

View Details

Carlos Alberto Haro @h1sort

2 weeks ago

I guess we’re doing local models now

vicki @vboykis

2 weeks ago

new post: how I develop recently using local models. the tooling is now good enough to do agentic workflows and everyone should give them a try! vickiboykis.com/2026/06/15/run…

29 94 848 111K 1K

0 0 0 260 0

View Details

Manthan Gupta @manthanguptaa

3 weeks ago

One of the more interesting takeaways from Satya's essay is that we are probably focusing on the wrong layer of the stack. Everyone is obsessed with models. Which model is smartest? Which model wins benchmarks? Which model should a company standardize on? But models are increasingly looking like interchangeable components. The real moat is the learning loop. A company should be able to swap GPT for Claude, Claude for Gemini, Gemini for an open-source model, and still retain most of its competitive advantage. That advantage comes from everything sitting above the model: - institutional knowledge - workflow traces - evaluation systems - decision histories - memory systems - retrieval infrastructure - reinforcement signals derived from real business outcomes In other words, the ability to continuously convert human capital into token capital. The line that stood out to me was the idea that every company needs its own learning loop. Not just a knowledge base or a chatbot, but a system that gets better every time an employee solves a problem, makes a decision, or executes a workflow. That also changes how I think about memory. Most people treat memory as a storage problem. I increasingly think it's a retrieval and learning problem. Storing years of organizational knowledge is easy. The hard questions are: - How do you make institutional knowledge machine consumable? - How do you retrieve the right knowledge at the right moment? - How do you turn successful outcomes into future training signals? - How do you improve the system without becoming dependent on a single model provider? The companies that win in the AI era won't necessarily have access to the best model. They will have the best system for compounding organizational learning. The model is the engine. The learning loop is the moat.