Brian Peña Calero (@brianmsm) profile picture

Studying Máster Metodología en Ciencias del Comportamiento y de la Salud - UCM. Interest in mental health issues, psychometric and empirical research.

Madrid, Spain
Joined January 2014
Photo by Brian Peña Calero (@brianmsm)Photo by Brian Peña Calero (@brianmsm)Photo by Brian Peña Calero (@brianmsm)Photo by Brian Peña Calero (@brianmsm)Photo by Brian Peña Calero (@brianmsm)Photo by Brian Peña Calero (@brianmsm)Photo by Brian Peña Calero (@brianmsm)Photo by Brian Peña Calero (@brianmsm)Photo by Brian Peña Calero (@brianmsm)Photo by Brian Peña Calero (@brianmsm)
Brian Peña Calero retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
274
675
394
6,743
225,432
Brian Peña Calero retweeted
Did you know the same intelligence can cost 10× more, need 6× more output tokens and take twice as long? In addition to task spend, Agent Arena now shows which agents give you best performance for every output token generated and every minute you wait. These real-world usage plots reveal that costs grow most steeply with intelligence, waiting longer pays off unevenly, and output tokens barely predict the bill. To find the best model for you, see the link in thread.
20
3
123
14,951
Brian Peña Calero retweeted
Introducing Phonic-1, the most advanced speech-to-speech model for agentic tasks — from @phonic_co, the speech-to-speech model company. We’ve had a breakout year and raised a $27.5M Series A led by @firstround with participation from @Lux_Capital, @felicis, and Definition.
146
197
134
1,313
1,074,840
Brian Peña Calero retweeted
From recognizing text to understanding documents. 📄 Meet Xiaomi-OCR-0, an open-source document understanding model with just 0.8B parameters. It brings document parsing, key information extraction, and OCR-oriented visual question answering into one lightweight model. It scores 95.24 on Real5-OmniDocBench and 83.2 on average across five OCR-oriented VQA benchmarks. On OmniDocBench v1.6, Xiaomi-OCR-0 ranks first in document parsing performance among models with fewer than 1B parameters. Model weights, inference code, an interactive demo, MCP service, and Agent Skill are now open source. 👇 🤗 Huggingface: huggingface.co/xiaomi-resear… 🔗 GitHub: github.com/xiaomi-research/X… 📄 Demo: huggingface.co/spaces/loongw… #XiaomiAI #OCR #OpenSource
57
309
53
2,847
148,938
Brian Peña Calero retweeted
Montar llamadas VoIP para un agente de IA en España es un rollo. Twilio te pide números, verificaciones y papeleo. Las APIs de WhatsApp y Telegram tampoco te lo ponen fácil. Y al final pagas por minuto por algo que solo quieres para ti. Así que me he hecho una app que lo resuelve: agentring. Le da a tu agente la capacidad de llamarte al móvil, y a ti la de llamarle a él. Cómo funciona: → Tu agente (Claude Code, Codex, OpenClaw, n8n o un script) tiene una herramienta "call". Cuando necesita que decidas algo, la usa: "la migración ha fallado, ¿hago rollback o lo arreglo y reintento?". → Tu móvil suena como una llamada normal, aunque esté bloqueado. Coges y una IA de voz te explica qué pasa. → Lo habláis, tú decides, y tu agente recibe la conversación escrita y sigue trabajando con tu respuesta. → Al revés también: abres la app, pulsas "Llamar al agente" y la IA le pregunta a tus agentes lo que necesites mientras habláis. → Si no lo coges, te llega como mensaje y respondes desde la notificación. Yo ahora mismo lo uso para que me llame si: • Sale en Twitter o en alguna fuente oficial una vulnerabilidad crítica que me afecte. • Detecta algún problema de seguridad en mis servidores. • Entra un correo crítico. • Se cae algo. Con todo lo que está pasando, creo que hace falta tener agentes levantados revisando tu seguridad, igual que va a haber cientos de miles intentando entrar. Y si encuentran algo, un mensaje no basta: tiene que sonarte el móvil. Por dentro: • La llamada va por VoIP desde tu servidor a tu móvil. Sin números de teléfono, sin Twilio y sin pagar por minuto. • La voz es Gemini Live: muy barato, y en AI Studio tienes capa gratuita para probarlo. • Es open source y lo montas tú, en un VPS o en tu propio ordenador. Se conecta a tu agente en un par de minutos. Por ahora es solo Android. ¿Qué se os ocurre montar con esto? Os dejo el repo en el siguiente comentario por si alguién lo necesita 👇
39
42
8
447
33,362
Brian Peña Calero retweeted
No digo nada pero Qué infravalorado está lo que ha hecho @mrm8488 con huggingface.co/mrm8488/embed… gracias por la ideaza, el provecho que le voy a sacar no tiene ni sentido ^^
5
10
135
8,800
Brian Peña Calero retweeted
Day 25/ We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today. We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering extended limits for dots for now, and usage of your dot will still continue to be separate from your default usage. And we'll of course also keep shipping better models and other efficiency improvements that increase your effective usage over time.
7,383
2,676
3,626
38,693
6,980,157
Brian Peña Calero retweeted
claude designed a 3d part for my robotics experiment in freecad. i'd never used cad, so every time i wanted to change something i couldn't even say which part i meant so i gave my coding agent its own cursor. i circle the thing, ask, and it points back
3
4
14
1,157
Brian Peña Calero retweeted
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model. Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards. As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
2,083
2,750
1,918
34,086
6,898,749
Brian Peña Calero retweeted
I don't code anymore, but I miss looking at code. Yesterday I threw together a TUI that reminds me of Neovim, that watches the agents work. github.com/bjarneo/shoulder
29
9
3
289
20,656
Brian Peña Calero retweeted
OpenDisplay for Linux is now available, Sender & Receiver! It's still early, therefore happy to receive feedback and contributions to support as many linux setups as possible.
24
100
11
1,577
76,409
Brian Peña Calero retweeted
Introducing Real World Engineering Bench (RWE-Bench). AI isn’t great at designing rockets, engines, or chips yet. With better evals and RL environments, it will be. Software was just the beginning. The physical world is next.
86
152
57
1,226
158,397
Brian Peña Calero retweeted
🇵🇪 Ya está en línea: el Censo 2017 vs el Censo 2025, distrito por distrito. 👉 censo-2025.appsbi.net Empiezo con un dato que parece contradictorio. 🚗 En 2025 hay 1,285,772 hogares con auto o camioneta, 196,157 más que en 2017 (+18 %). Pero el porcentaje bajó de 13,2 % a 12,2 %. ¿Por qué? Los hogares crecieron más rápido que los autos: pasamos de 8,25 a 10,57 millones de hogares en solo 8 años. Hay más familias, más pequeñas. 📍 Y la brecha es enorme: • San Isidro: 70,9 % de hogares con auto • San Borja 64,1 %, La Molina 62,9 %, Miraflores 62,2 % • En distritos de Loreto como Jenaro Herrera, Yaguas o Yaquerana: 0 % Dos realidades en nuestro mismo pais. 📊 Qué puedes hacer en la app: ✅ Comparar 60 preguntas que se hicieron en ambos censos: auto, refri, internet, celular, SIS, EsSalud, agua, desagüe, piso de tierra, gas natural, leña, idioma materno, educación y más ✅ Verlas en el mapa de los 1,893 distritos: 2017, 2025 o cuánto cambió ✅ Ver cantidades y %, y el top 20 de distritos que más subieron y más bajaron ✅ Bajar a tu departamento, provincia o distrito ✅ Explorar todas las variables del Censo 2025 🔍 Por qué no es un simple cruce: El INEI cambió códigos y opciones de respuesta entre un censo y otro. Emparejamos pregunta por pregunta y documentamos qué se agrupó y qué no es exactamente comparable. Además, validamos que cada distrito cuadre con el total de su región, y los totales 2017 coinciden con los oficiales del INEI. 🔓 Todo es abierto: • Datos en CSV para descargar • Metodología completa • Código completo para clonar: extractores, endpoints del INEI, backend y frontend ℹ️ Los datos son de la población y los hogares censados (32,7 millones de personas), los mismos que el INEI publica en su página. Los 34,2 millones de habitantes que anunció el INEI suman además la población omitida que estima su encuesta postcensal. Cualquier mejora es bienvenida. ¿Encontraste un error, una pregunta mal emparejada o una idea para la app? Comenta y la implementamos. 📥 Y si quieres hacer tus propios análisis, eres libre de hacerlo: toda la información es pública. Descarga los CSV, la base y el código en 👉 censo-2025.appsbi.net/descar…
29
280
6
906
40,069
Brian Peña Calero retweeted
Flint makes a great teacher. Ran it locally on a 24GB M5 Pro, had it write + double-check 1.2k text-to-SQL pairs, then distilled a 331MB model with 0.2s per query (Flint: 7.4s). Model, data + demo: huggingface.co/datasets/alex…
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine. • Best-in-class fast local coding • Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark • Built from scratch in under 4 months Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
3
5
89
4,798
Brian Peña Calero retweeted
¡He creado una biblioteca para usar IA en tu navegador! Modelos de pesos abiertos y especializados. ✓ Gratis y Open Source ✓ Sin APIs ni servidores ni pago por tokens ✓ Tus datos nunca salen de tu dispositivo Se llama runonweb: → github.com/midudev/runonweb.…
66
343
13
3,044
131,520
Brian Peña Calero retweeted
Introducing epstein.md 👋 A 927,400+ Character .MD File that's specifically trained to skip every vibe coded AI Slop design principle. - no box shadows - no emojis - no '🟢 LIVE ' badges - 0 blue/purple gradients & religiously ignores 350 most common AI words (delve, vibe, seamless ...) It's the result of 11 months of 8 hrs/day Twitter doomscrolling and constant additions to the codebase of our vibecoding ui. Wholeheartedly hoping this finds the most design-needy indie hackers.
136
86
81
2,117
445,534
Brian Peña Calero retweeted
Introducing GLM-Legal Built on GLM-5.3-Flash, GLM-Legal matches frontier models on mainstream legal benchmarks. Designed for real legal workflows: contract review, evidence analysis, compliance, and more. Tech Blog: z.ai/blog/glm-legal Apply for API Access: z.ai/consultation/legal Try AutoClaw: autoclaw.z.ai
350
574
321
7,770
855,752
Brian Peña Calero retweeted
Introducing RustGIS v0.1.0, the first release! RustGIS is a fast, free and open-source GIS in pure Rust for mapping, analyzing and sharing geospatial data. It runs natively on Windows, macOS and Linux, and in your web browser. Your data stays on your machine unless you send it to an AI model you choose, and every action is a command you can script. This first release brings a full desktop GIS with a familiar ribbon, over 1,000 geoprocessing tools, and an AI Assistant that works the map for you. What it does - Your data, any map: GeoJSON, CSV, Shapefile, GeoParquet, Flatgeobuf, PMTiles, GeoTIFF and COG, web tiles, vector tiles and MapLibre styles, on a 3D globe or in any projection. - Cartography and layouts: thematic symbology, collision-free labels, page layouts, map series, PDF export, reports and presentations. - Analysis: overlay and proximity tools, guided workflows, ModelBuilder, and 1,000+ GeoLibre and Whitebox tools for terrain, hydrology and LiDAR, even in the browser. - Editing: snapping, templates, domains and attribute rules, with undo for every change. - AI Assistant: ask questions or give instructions in plain language, using the model you choose (Anthropic, OpenAI, Gemini, Ollama and more). - Scriptable and extensible: run any command from the command line, and add tools with sandboxed WebAssembly plugins. - Ready to install: Windows installers, a signed macOS app, Linux .deb, .rpm and AppImage, and 15 interface languages. Try it out - Launch the web app: web.rustgis.app - Download: downloads.rustgis.app - Documentation: rustgis.app - GitHub: github.com/opengeos/RustGIS - Release notes: github.com/opengeos/RustGIS/… #GIS #Rust #OpenSource #Geospatial #RemoteSensing #AI
67
285
26
1,831
79,175
Brian Peña Calero retweeted
Introducing Microsoft-Decision-1, our new model for fast decision-making. It delivers top performance on structured decision tasks, outperforming both LLMs and other decision models in latency and quality. We’re already testing it across Microsoft for everything from incident response and quality control to scientific discovery.
710
1,067
604
11,371
1,706,759
Brian Peña Calero retweeted
Introducing 𝙾𝚙𝚎𝚗 𝙸𝚗𝚝𝚎𝚕𝚕𝚒𝚐𝚎𝚗𝚝 𝚄𝙸 A template for bringing interactive, generative UI to your own product. Works with ANY agent. Supports: → Your design system → Your components → Your real data and actions Repo: github.com/CopilotKit/OpenIn… Available on Web, Mobile, and Slack. Powered by @CopilotKit and AG-UI.
GPT-6 and Intelligent UI, now rolling out in ChatGPT for everyone. Intelligent UI in ChatGPT delivers fast, interactive answers that make everyday questions more visual, complex topics easier to grasp, and interactive tools for your task available on the spot.
29
43
18
662
75,238