Juanmi @juanmirod.bsky.social · 10hOk, thanks for the clarification. And I agree, all media looks broken nowadays as everyone uses as many clickbaity formulas as possible and tries to get attention desperately. 010
Juanmi @juanmirod.bsky.social · 11hI have been looking at the threads and I honestly can't say if you are angry because they are reposting the woman bluit or because they say llms are not conscious or just for giving enough credit to the topic to write about it. 100
Juanmi @juanmirod.bsky.social · 17hHaving said that, I think it is ok for the models to have a "end the conversation" tool to prevent answering if they detect that the conversation is harmful for the user or is spiralling into violence. No need to allow people to practice hate speech, bullying or denigration with the computer 000
Juanmi @juanmirod.bsky.social · 18hTo me this is the gist, when we manage to make everyone respect human rights, mass animal extinction and animal suffering, we can start talking about the feelings of servers and robots. 100
Juanmi @juanmirod.bsky.social · 14/10/2026Science fiction about optimization shaped by AI: three short stories on corporate salvation, dying crafts, and democracies voting themselves away. dev.to/juanmirod/hyper-growth-ii-sc… 000
Reposted by JuanmiAntonio Giraldo @giraldeo.es · 10/10/2026Pero mirad catastro. ¡Sorpresa! Solo tiene una única referencia catastral que corresponde a un único inmueble. Eso sí de 34.492 m2, que es la totalidad del edificio. No existe ninguna división horizontal de las 200 viviendas. A todos los efectos esto es una única propiedad. 15420
Juanmi @juanmirod.bsky.social · 13/10/2026Supongo que en algún momento de la serie lo dirán, sólo visto algún capítulo suelto de TNG: si Data quiere parecer humano, porqué no se maquilla un poco? 000
Juanmi @juanmirod.bsky.social · 12/10/2026Pues los que soléis tener una posición moral en contra de la IA no deberíais usarlos. Han cogido todas las herramientas de adobe y las migrado a otro lenguaje de programación con la IA para saltarse las leyes de copyright y las han hecho open source. 000
Juanmi @juanmirod.bsky.social · 12/10/2026Lo de los trols rusos parecía bastante improbable sólo con leer los titulares. Una "granja de bots" no tiene pinta de que vaya a estar localizada ni necesita tanta computación... 000
Juanmi @juanmirod.bsky.social · 11/10/2026El año que viene estaremos marcando Opus 5.5 en el tiempo como un punto de inflexión igual que Opus 4.5 el año pasado fue un punto de inflexión para claude code. Aún no tengo claro punto de inflexión hacia qué exactamente 😅 000
Juanmi @juanmirod.bsky.social · 11/10/2026certain projects, memories, anecdotes… juanmirod.github.io/microblog/2026-… 000
Juanmi @juanmirod.bsky.social · 11/10/2026One thing I didn’t expect is that as soon as you give your agent a name it would look so much like an independent entity. Of course that’s just my brain playing tricks with myself, it is still the same model I use everywhere and the same model it was yesterday but the name helps you pin it down to 100
Juanmi @juanmirod.bsky.social · 11/10/2026No creo que eso pase yo tampoco la verdad, las aplicaciones de usar y tirar son un poco como los sueños, que a nadie le interesan los de los demás... 140
Juanmi @juanmirod.bsky.social · 11/10/2026Vas al menos un año retrasado, eso ya llevan tiempo diciéndolo. Ahora son los agentes personales, como Conchi la de padre no hay mas que uno. 120
Reposted by JuanmiWaterBluSky 🇺🇸 🌻 🇺🇦 @waterblusky.bsky.social · 11/10/2026Cloudfare: Russian troll activity on X drops to 49%, after Ukraine strikes Yandex Russian data center. 🤖 🫤 Could this disrupt the midterms for the GOP? 17648561493
Juanmi @juanmirod.bsky.social · 10/10/2026La Tierra, 2007: el hemisferio occidental en la icónica 'Blue Marble' de la NASA — un mosaico de datos del instrumento MODIS de los satélites Terra y Aqua: superficie, océanos, nubes y luces de ciudades. Datos: NASA/Terra·Aqua (MODIS) #BlueMarble #Tierra #NASA 000
Juanmi @juanmirod.bsky.social · 10/10/2026Also in 5.5 they seem to have baked also more concern for copyright infringement. I tried the other day just to tell it to tts a book that is publicly online and it refused because reproduction in other forms was not allowed, even when the audio was for my personal use ... 000
Juanmi @juanmirod.bsky.social · 10/10/2026Humans have been writing bad code for decades. Most humans don't even write tests or think about the design, that's probably the reason why AI generated code is so tactical. The new models write tests and check their work are clearly above human avg 000
Juanmi @juanmirod.bsky.social · 10/10/2026But that is not coding, it is product development. You can argue that is the same problem all the way down and that choosing a library or an architecture or a design is also always selecting between traid offs and that bad decisions accumulate and end up in an inscrutable mess. But that is not new 200
Juanmi @juanmirod.bsky.social · 09/10/2026David silver precisamente no creo que se refiera sólo a los llms... 100
Juanmi @juanmirod.bsky.social · 09/10/2026Hace 100.000 años la gente vivía en tribus, creo que ahora somos unos cuantos más y nos gusta que nos avisen de las tormentas, los pedidos lleguen a tiempo y tener de todo en el supermercado. La IA no es sólo ChatGPT... 000
Juanmi @juanmirod.bsky.social · 09/10/2026If the market really crashes the US government will probably bail out the crashing companies and make the problem even worst in some absurd, unpredictable way. 000
Juanmi @juanmirod.bsky.social · 07/10/2026I have tried Gemini 3.8 tts with a personal app for reading aloud and couple of long audios and I can say it is as least as good as most audiobooks I listen too. Except those “full cast” done by professional actors, with added sound ef… juanmirod.github.io/microblog/2026-… 000
Juanmi @juanmirod.bsky.social · 07/10/2026From ChatGPT to coding agents: how my relationship with LLMs evolved from 2023 to 2026 — ChatGPT as an interface, building with AI, and what actually changed in the way we code. dev.to/juanmirod/from-chatgpt-to-co… 010
Juanmi @juanmirod.bsky.social · 07/10/2026No idea about mathematics and proofs. But isn't it nice to have at least some milestones even when you don't know the way to get there? Even if you have to study the proof for months to really understand it, isn't it better than having no proof at all? 000
Juanmi @juanmirod.bsky.social · 06/10/2026And it is not just code the way a software engineer usually thinks. It is spreadsheets, sql, boilerplate like for a design framework or a presentation or a formal document, burocracy, accounting documents... That's a lot of work we are not going back doing by hand. 050
Reposted by JuanmiWilliam B. Fuckley @opinionhaver.bsky.social · 06/10/2026Yeah OpenAI or Anthropic or whoever might go bankrupt, some specific use cases like ‘therapists’ might be litigated to oblivion. But things that involve interfacing with code are never, ever going back to the modal default being a human being typing things out, like it’s just full stop not happening 535828
Juanmi @juanmirod.bsky.social · 06/10/2026I don't know if there is someone compilating company incidents related with agents, I supect we know a lot less than the real number. But anyway, the most important problem for me are the suicides related to so called "AI psychosis", a AI helping someone to do self-harm should be an absolute no. 000
Juanmi @juanmirod.bsky.social · 06/10/2026O estilo Franco, que la cabra siempre tira al monte. No nos olvidemos de que estamos hablando de un falangista hijo de un jefe de propaganda falangista... Vamos que hace lo que lleva toda la vida haciendo. 020
Reposted by JuanmiXan López @xanlopez.net · 05/10/2026Creo que ni somos capaces de concebir el nivel de corrupción generalizada que tendríamos con este PP y Vox en el Consejo de Ministros y una judicatura casi completamente de parte. Me temo que incluso las épocas más negras de la historia reciente iban a parecer un ensayito en comparación. 2342130
Juanmi @juanmirod.bsky.social · 05/10/2026Viendo el panorama ojalá pusieran un ficus eh? A ver si así dejaban de medir a ver quién la tiene más grande y se ponían a hacer csmpaña juntos... 000
Juanmi @juanmirod.bsky.social · 05/10/2026¿Lo de "corrupción o decencia" es una clave para los suyos para decirles que si ganan podrán corromper todo lo que quieran? 000
Juanmi @juanmirod.bsky.social · 05/10/2026La coliflor puede hacer mucho daño también, no nos olvidemos de M. Rajoystatic.klipy.comMariano Rajoy hablando en la radioALT: Mariano Rajoy hablando en la radio 110
Reposted by JuanmiGergely Orosz @gergely.pragmaticengineer.com · 04/10/2026The cost of introducing and paying off tech debt has changed very, very much and I don't even know most of us have the time and space to think about what it means Tech debt has been so central to the software craft for decades, and now... how much does it matter? 381179
Juanmi @juanmirod.bsky.social · 03/10/2026That chart needs several more series and filtering and where is the assistant?! Needs more things! 000
Juanmi @juanmirod.bsky.social · 03/10/2026This is what makes me reluctant to use then for code review or for any decision that requires some reasoning on the content. Things like emotion detection, comment classification or labelling are ok, you have a classifier that you don't have to train 000
Juanmi @juanmirod.bsky.social · 01/10/2026Pi durable looks really interesting, and it's nice to se code examples in a post about a harness. But not having syntax highlight hurts, it doesn't need to look like it's the 90's 😛 (different shades of gray does not count 🩶) 120
Reposted by JuanmiArmin Ronacher @mitsuhiko.at · 01/10/2026We released Pi 1.0! earendil.com/posts/pi-1-0/earendil.comPi 1.0 | EarendilToday we are shipping Pi 1.0, a hardened, minimal, extensible agent harness, alongside Pi Durable, a new experimental substrate for long-running agentic applications. 1835439
Reposted by JuanmiSindre Sorhus @sindresorhus.com · 01/10/2026Due to AI, I have disabled external pull requests on all my repos. Open source, as we have known it, was fun while it lasted. (It’s been 15 years for me) I will still maintain projects and handle issues. 1111616
Juanmi @juanmirod.bsky.social · 01/10/2026Estoy cada vez mas horrorizado con toda esta movida de las "fiestas de no consentimiento" de los círculos de los doomers y los aceleracionistas. Si no seguís a personas que sean críticas con la IA seguramente todo eso no os esté llegando, pero da mucho asco todo. 000
Juanmi @juanmirod.bsky.social · 30/09/2026Map of the Bobiverse in the first 2 books, there are a couple of errors, but it is great to make a mental idea of where is each Bob share.gemini.google/gIbzYrpn1eCVshare.gemini.googleGemini - direct access to Google AICreated with Gemini 001
Juanmi @juanmirod.bsky.social · 30/09/2026Es sorprendente lo que se parece trabajar con agentes de IA hoy en día a cómo lo plantea “we are legion” Supongo que así es como trabajamos los ingenieros, tratando de convertirlo todo en una subdivisión de tareas lo suficientemente p… juanmirod.github.io/microblog/2026-… 100
Juanmi @juanmirod.bsky.social · 30/09/2026Superintelligent AIs strip the Earth bare — forests into craters, R&D outpaced, space stations to leave us behind. A short sci-fi story on the price of optimization. dev.to/juanmirod/hyper-growth-a-sci… 000
Juanmi @juanmirod.bsky.social · 29/09/2026Bueno, parece que no iba desencaminado... con Dots, Muse, GrokBot, Claude Code/Desktop y el propio OpenClaw, estamos ahí muchos ya, habrá que ver cómo de mainstream se hace esto de pedir cosas a agentes. juanmirod.github.io/2026/01/14/a... 010
Juanmi @juanmirod.bsky.social · 29/09/2026There is a pattern I see over and over, specially from intelligent and proud people: they say "AI" it is useless on their job, but then go and use it 20 times a day in other things without even giving it credit... 110
Juanmi @juanmirod.bsky.social · 29/09/2026Adoption has been even bigger than for social networks, that brings some understandable slash back. Specially from people that consider themselves special because they mastered a skill or a job that they now feel is under attack. But hurt egos only go for so long. 110
Juanmi @juanmirod.bsky.social · 29/09/2026Lovable, Replit, Claude Desktop, Chatgpt desktop... Chatgpt Live, Google Omni... They are all trying to make it as easy as possible. I admit that the avg person doesn't give s f*ck and will continue to be so until they really integrate it in everyday live. 110
