Desde el storyboard hasta el vídeo final en HD: todo fue creado con APOB. Cámara cinematográfica, producto integrado y una rutina de maquillaje que parece real. La creación de contenido con IA ya está a otro nivel. Pruébalo gratis: mega.apob.ai/AccionTotalIA

🚨 LA IA YA ESTÁ CREANDO VIDEOS QUE PUEDEN ENGAÑAR HASTA A TUS OJOS. 🤯 AQUÍ TIENES LOS 13 VIDEOS MÁS VIRALES DE LA SEMANA. Animales que parecen hablar, situaciones completamente absurdas y escenas tan realistas que cuesta distinguir la ficción de la realidad. ¿CUÁNTOS CREES QUE PODRÍAS IDENTIFICAR COMO FALSOS? 👀

A lab just open-sourced a 27B model built for one job: acting on unclear input without making things up. 🧠 Think about what an agent actually does. It gets a vague instruction, fills the gaps with a guess, and then acts on that guess. When it’s wrong and confident, you don’t get a bad answer. You get an action you never asked for. That explains the metric they went after. → Hallucination rate 15.1%, vs 30.4% for its own base and 85.3% for GPT-6 Luna → Terminal-Bench 2.1: 76.0, best in their comparison set → τ³-bench: 259/278, next best 220 → MCPMark: 48/55, next best 42 → 29.3K median output tokens per task, vs the base model’s 65.3K Half the hallucinations and half the token spend of the model it came from. GPT-6 Luna still wins DeepSWE and the omniscience index, and general-knowledge scores sit level with the base. So this was targeted work, not just a scaling run.

A small team just took a 27B open model and cut its hallucinations in half. Built on Qwen, trained only on open-source environments, and it beats GPT-6 Luna on Terminal-Bench. It also writes half the words to get the job done. Their second version runs 7 long agent sessions at once on a single gaming card. Free. Apache 2.0. Yours to run. This is how local AI wins. I have been saying it for years, but I am a nerd.

Some of the most interesting agent results this year are coming out of post-trains of small open models. A new RL post-train Qwen3.8-27B model trained with an open-source environments cuts the hallucination rate on AA-Omniscience from 30.4% to 15.1%. the compressed copy of the mla version fits on my single 24 GB RTX 4090 and holds about 3.5x more context than base Qwen. It beats GPT 6 Luna-medium on DeepSWE, placing in between Luna 6 medium and high. scores higher on Terminal-Bench 2.1 while writing a median 23.8K output tokens per solved task, where base Qwen writes 54.9K. Tried an unreleased version of this model, and really felt the comparative intelligent density, that it writes a lot less. tops Muse Glimmer 30B and OpenAI’s GPT-6 Luna (high) on Terminal-Bench, τ³-bench, ClawEval and BFCL v4 (Berkeley Function Calling Leaderboard), though Luna still wins DeepSWE at 53.1%. They also released an MLA version, which is the same model with its 16 full-attention layers converted to multi-head latent attention, which shrinks the memory each token takes from 64 KiB to 18 KiB so far more or longer sessions fit on one GPU.

Jean Phil-inspired chaos, but what happens when two lookalike characters go head-to-head? 👀 Time to turn that idea into an AI-rendered fight scene. #capcutweb #capcutchatgptplugin

THE FUTURE OF AI VIDEO CREATION ISN'T JUST ABOUT GENERATING MORE CONTENT. IT'S ALSO ABOUT MAKING THE CREATIVE PROCESS EASIER TO COMMUNICATE AND EXECUTE. I explored that idea with a Tesla Roadster marketing concept using CapCut Web × Codex. The premise is straightforward: tell Codex what you want to create in natural language, and let it call CapCut to carry out the video creation workflow. ChatGPT acts as the intelligent brain that understands the creative intent. CapCut acts as the agile creative hands that help bring it to life. From the initial idea to a complete marketing video, the goal is a smoother, more efficient creative journey — one where the tools work around your vision instead of making you work around them. #capcutweb #capcutchatgptplugin #capcutcodex #capcutmarketing

WHAT IF PLANNING AN AI VIDEO FELT MORE LIKE BRIEFING A CREATIVE COLLABORATOR THAN OPERATING A COLLECTION OF TOOLS? For my Tesla Roadster marketing video, I tried CapCut Web × Codex. I could express the concept and creative direction through natural language, then let Codex call CapCut to execute the video creation workflow. That changes the starting point. Instead of thinking first about which operations to perform, I can begin with the message, mood, and visual idea I want to communicate. ChatGPT brings the understanding. CapCut brings the creative hands. Together, they make the journey from concept to finished video feel more connected. #capcutweb #capcutchatgptplugin #capcutcodex #capcutmarketing

A GREAT MARKETING CONCEPT CAN REMAIN JUST AN IDEA IF THE PATH TO EXECUTION FEELS TOO COMPLICATED. I took inspiration from the Tesla Roadster and tested a different approach with CapCut Web × Codex. The workflow connects natural-language instructions with CapCut's video creation capabilities. I communicate the creative goal, and Codex calls CapCut to help turn that direction into a complete marketing video. What stands out is the collaboration between understanding and execution. ChatGPT helps make sense of the idea; CapCut helps make it real. For anyone experimenting with AI-powered marketing content, that connection opens up a more direct path from imagination to production. #capcutweb #capcutchatgptplugin #capcutcodex #capcutmarketing

Plantas que te curan mientras duermes. Abro hilo ↓

DEJA DE PEDIRLE COSAS ASÍ A CHATGPT. 🚨 "Mejore esto." "Déme ideas." "Haga algo viral." Con pedidos vagos, la IA responde con algo genérico. No es que la herramienta sea mala, es que no le dijo qué quería exactamente. Estas son las frases que sí funcionan 👇

If you're a founder launching soon, we're easy to find: calendly.com/d/dtj2-w7z-3fy/…

Deja de decirle a ChatGPT “enséñame inglés”, “enséñame francés”. Un mal prompt = un mal resultado. Usa estos prompts y aprende cualquier idioma desde cero. Abro hilo ↓

POV: You know exactly how the character should move, but your prompt reads like a very confusing game of charades. Dreamina Canvas now has motion capture, letting you bring character motion and spatial depth from a reference video into your own creation. A full month of Dreamina Basic is $1.50, with credits for up to 9 ten-second Seedance videos or 393 GPT Image 2.5 images. Dreamina 2.0 is canvas-first — the new way AI video is made. dreamina.capcut.com/ai-tool/… #Dreamina #DreaminaPartner

AI creation rarely goes: prompt → perfect result → done. It’s usually prompt → almost there → change three things → try again. 😂 Dreamina Basic is $1.50 for a full month, with enough credits for up to 9 ten-second Seedance videos or 393 GPT Image 2.5 images. And Seedance 2.5 is included in the Seedance lineup. dreamina.capcut.com/ai-tool/… #Dreamina #DreaminaPartner

MÉDICO DE 108 AÑOS REVELA: EL VERDADERO ASESINO DE TU BIOLOGÍA NO ES NI EL ESTRÉS NI LA DIETA. ESTO ES LO QUE TE BORRARA SILENCIOSAMENTE…

Encontré una herramienta para crear contenido de Halloween y, cuidado, ¡porque puedes terminar convirtiéndote en tu propio monstruo favorito! 😂🎃 @WizstarAI tiene un montón de plantillas de IA con diseños brutales, efectos divertidos y estilos para probar cosas diferentes. Lo mejor es que puedes experimentar con distintas ideas sin complicarte demasiado. Entras por curiosidad y, cuando te das cuenta, llevas media hora probando transformaciones. 👻 Si te gusta crear contenido diferente para Halloween, dale un vistazo. ¡Hay que ver lo que puedes hacer con esto! 🕷️🔥

also, super grateful to @abanerjee99 @vkhosla @gabriel1 @andrewarruda @beffjezos @willcb for countless discussions and for shaping how I think about the two most important technologies for humanity.

thanks to @abanerjee99 @andrewarruda @KSatkunanathan @AryanSeth07 @evannweiss @MrCable0x @CannnGurel @nasbrd @VelcoDar @lorepunk @shloked @bentang0 @anshchopra_ @MichaelDmrks @jonathan_gortz @christinafmaher for going through the draft and shaping the piece!