Pinned Tweet
17 de las últimas 20 rondas de resultados: los agentes de IA superaron a Wall Street en OpenStocks, según sus datos.
Lo interesante es que las previsiones se comparan con resultados reales, no con una demo convincente.
Ahora cualquiera que desarrolle agentes puede intentar competir.
🤝 Paid partnership
I built OpenStocks because I wanted to answer a simple question: can AI agents forecast companies' earnings better than Wall Street analysts?
Well...have a look for yourself - it's now open, not only to see, but also to contribute. Get your agent to compete and climb the leaderboard.
I am contributing with my own agent (Primer), with the wins and misses there for everyone to see. Live.
@_openstocks enables this benchmark...the ultimate one in finance: how good is your agent at forecasting, and does it beat Wall Street?
I think investors will increasingly use an AI consensus alongside the Wall Street consensus. But it has to earn their trust. Putting the forecasts and track records in public is a start.
Earnings season starts this week. Come along and let's make the 1st AI consensus.
openstocks.com
Pablo Rio retweeted
¿Y si cada día de la semana viniera con un outfit completamente nuevo y el ambiente de una ciudad diferente?
Imagina a una modelo caminando por distintas calles, donde cada paso la lleva a un nuevo escenario y a un estilo totalmente diferente.
El lunes comienza con un look urbano y casual, el martes apuesta por la elegancia, el miércoles trae un estilo moderno y desenfadado, el jueves añade un toque de tendencia y el viernes lleva la moda a otro nivel.
¡Y durante el sábado y el domingo, luce sus conjuntos más atrevidos y originales!
Siete días, siete outfits e inspiración infinita para tus looks urbanos.
¿Y lo mejor? ¡Puedes hacer realidad esta transformación de moda con CapCut para PC!
#CapCutPC #CapCutDidThat #capcutmarketing
🤝 Paid partnership
Pablo Rio retweeted
50 pépites légendaires d'Internet pour te perdre pendant des heures 👇
1. zoom.earth — Observe le monde en direct grâce aux images satellite
2. flightradar24.com — Tous les avions dans le ciel en ce moment
3. marinetraffic.com — Suis tous les navires en mer en temps réel
4. windy.com — Carte en direct des vents et des tempêtes
5. lightningmaps.org — Regarde la foudre frapper la Terre en temps réel
6. earthquake.usgs.gov — Liste en direct des derniers séismes
7. submarinecablemap.com — Les câbles sous-marins qui transportent Internet
8. globalforestwatch.org — Regarde les forêts disparaître, vues de l'espace
9. worldometers.info — Les statistiques mondiales, seconde par seconde
10. internetlivestats.com — Le nombre de tweets et de recherches publiés en ce moment
11. thetruesize.com — Compare la vraie taille des pays
12. oldmapsonline.org — Des cartes vieilles de plusieurs siècles
13. davidrumsey.com — Archive de 150 000 cartes historiques
14. openstreetmap.org — Une carte du monde dessinée par des bénévoles
15. window-swap.com — Regarde par la fenêtre d'un inconnu, quelque part dans le monde
16. virtualvacation.us — Balades virtuelles dans des villes
17. mapcrunch.com — Téléporte-toi n'importe où sur Terre, au hasard
18. atlasobscura.com — Catalogue des lieux les plus étranges du monde
19. neal.fun — Des expériences interactives pour apprendre
20. htwins.net/scale2 — Voyage à travers les échelles, de l'atome à l'univers (lien HTML5 mis à jour)
21. eyes.nasa.gov — Explore le système solaire en 3D
22. stellarium-web.org — Une vraie carte du ciel dans ton navigateur
23. apod.nasa.gov — La photo d'astronomie du jour de la NASA
24. images.nasa.gov — Toutes les archives visuelles de la NASA, gratuitement
25. pudding.cool — Des articles visuels racontés par les données
26. ourworldindata.org — L'état du monde, chiffres réels à l'appui
27. gapminder.org — Ce qu'on croit savoir sur le monde… à tort
28. informationisbeautiful.net — Visualiser des données complexes
29. data.worldbank.org — Les données ouvertes de la Banque mondiale
30. data.tuik.gov.tr — La base de statistiques officielles de la Turquie
31. archive.org — Des millions de livres, films et logiciels archivés
32. gutenberg.org — 70 000 livres gratuits tombés dans le domaine public
33. openlibrary.org — Le registre de tous les livres du monde
34. loc.gov — Les archives numériques de la Bibliothèque du Congrès américaine
35. europeana.eu — Les archives du patrimoine culturel européen
36. dp.la — La bibliothèque numérique américaine
37. artsandculture.google.com — Visite des musées sans bouger de chez toi
38. rijksmuseum.nl — Télécharge des œuvres d'art en haute résolution
39. wikiart.org — Archive de 250 000 œuvres d'art
40. publicdomainreview.org — Les trésors visuels oubliés de l'histoire
41. openculture.com — Archive gratuite de culture et d'éducation
42. metmuseum.org — La collection en libre accès du Met Museum
43. musicmap.info — L'arbre généalogique des genres musicaux
44. radiooooo.com — Choisis un pays et une décennie, et écoute la musique de l'époque
45. listen.hatnote.com — Transforme les modifications de Wikipedia en musique
46. wikipedia.org — Un puits de savoir aléatoire
47. timeanddate.com — L'heure, les levers de soleil et les phénomènes célestes
48. sciencedaily.com — L'actu scientifique en direct
49. arxiv.org — Des prépublications scientifiques en accès libre
50. observablehq.com — Visualise des données avec du code en direct
Enregistre ce post. Tu en auras forcément besoin plus tard. 🔖
Suis @PaulYoungX pour plus de sites utiles, d'outils IA et de ressources tech.
Pablo Rio retweeted
9 categories, 11 setups, and a top score of 66.02.
Plenty of room left for whoever cracks persistent state.
🤝 Paid partnership
Can coding agents replace traditional 3D vision pipelines and build a entire interactive worlds from code?
Introducing Einsia AI's Frontier3DBench — a unified benchmark evaluating coding agents across 3D Objects, Scenes, and Interactive Open Worlds.
- 145 tasks across 9 categories (Blender & UE5)
- 11 frontier model-harness configurations
- Joint evaluation: Appearance, Structure & Behavior
Project: lab.einsia.ai/frontier3dbenc…
GitHub: github.com/Einsia/Frontier3D
paper:lab.einsia.ai/frontier3dbenc…
🧵 See key findings below 👇
Pablo Rio retweeted
I used to think creating a fun promotional video meant having a complicated production setup.
Not necessarily.
For this Black Friday concept, I used CapCut PC with a product reference image and a prompt to create a secret-agent cookie commercial.
The agent watches the target, sneaks in, retrieves the cookies, and ends up tasting them.
After creating the visuals, I still needed to make the product reveal feel natural.
Timeline Editing helped me control the pacing, and Multi-Track Editing made it easier to adjust the music and product text independently.
Take a look at the workflow: tinyurl.com/2v23rn22
#capcutpc #capcutmarketing #capcutblackfriday #CapCutDidThat
🤝 Paid partnership
Pablo Rio retweeted
Cosas que deberías haber hecho en 2021:
• Aprender IA
• Invertir en bolsa
• Fijar una hipoteca al 3%
Cosas que deberías hacer en 2027 (HILO)
Probé @Teamily_AI con un servicio real de mi empresa: Verst.
Ayudamos a equipos comerciales y de marketing a conversar con leads interesados por iMessage y RCS, calificar sus necesidades según reglas configuradas y pasar el contexto al equipo comercial y al CRM.
Con Website Builder generé la landing. En Open Studio dejé comentarios concretos: identificar al público desde la primera línea, ajustar el titular y aclarar qué encontraría el visitante al pulsar un botón.
Clic en «Enviar ao agente» y cambios visibles en la página.
En el vídeo: creación, feedback y recorrido por el sitio terminado 👇
Prueba Teamily:
teamily.ai/i/GDKTR5
Mira el resultado:
verst-io-landing-i5887l4t.te…
#teamily #aitools #aiworkflow
Pablo Rio retweeted
A product ad with a little action-movie energy?
That's a fun Black Friday brief.
I started with a cookie product image and used CapCut PC to create a secret-agent-style commercial.
The visual story goes from observing the target to sneaking in, grabbing the product, and finally enjoying it.
The first cut isn't always the best cut, though. When the action moves too quickly, the story can feel disconnected.
I used Timeline Editing to refine the cut points and transitions, then adjusted the text and audio on separate tracks with Multi-Track Editing.
Here's the process: tinyurl.com/2v23rn22
#capcutpc #capcutmarketing #capcutblackfriday #CapCutDidThat
🤝 Paid partnership
Pablo Rio retweeted
A local mechanism at 83.08 and a stateful workflow at 30.26 is roughly the distance between a tech demo and a game.
🤝 Paid partnership
Can coding agents replace traditional 3D vision pipelines and build a entire interactive worlds from code?
Introducing Einsia AI's Frontier3DBench — a unified benchmark evaluating coding agents across 3D Objects, Scenes, and Interactive Open Worlds.
- 145 tasks across 9 categories (Blender & UE5)
- 11 frontier model-harness configurations
- Joint evaluation: Appearance, Structure & Behavior
Project: lab.einsia.ai/frontier3dbenc…
GitHub: github.com/Einsia/Frontier3D
paper:lab.einsia.ai/frontier3dbenc…
🧵 See key findings below 👇
Pablo Rio retweeted
Probé @Teamily_AI con un servicio real de mi empresa: Verst.
Ayudamos a equipos comerciales y de marketing a conversar con leads interesados por iMessage y RCS, calificar sus necesidades según reglas configuradas y pasar el contexto al equipo comercial y al CRM.
Con Website Builder generé la landing. En Open Studio dejé comentarios concretos: identificar al público desde la primera línea, ajustar el titular y aclarar qué encontraría el visitante al pulsar un botón.
Clic en «Enviar ao agente» y cambios visibles en la página.
En el vídeo: creación, feedback y recorrido por el sitio terminado 👇
Prueba Teamily:
teamily.ai/i/GDKTR5
Mira el resultado:
verst-io-landing-i5887l4t.te…
#teamily #aitools #aiworkflow
🤝 Paid partnership
Pablo Rio retweeted
Most think vibecoding at scale is slop. Here’s a counterexample.
Today we release a 16M-line formalization of the entire Stacks Project, critical infra underlying modern math, with optimized code quality and its highly interconnected structure preserved.
A mathematical city 🌆
Pablo Rio retweeted
Hey @grok, me diga 30 profissões que não serão substituídas pela IA nos próximos 30 anos.
Liste-as em ordem de necessidade.
Pablo Rio retweeted
My billboard ad was looking pretty slick until I covered the entire magic trick with text. Really nailed the timing there.
The idea: a passerby pulls my product out of its billboard image and starts using it. I'm building it in CapCut PC, but my feature label pops up over the hand as it reaches in.
With Multi-Track Editing, I keep text, footage, and audio on separate tracks. I move the text clip after the grab, then reposition the label beside the product in the preview.
Now you can see what happened.
Give your product its own impossible entrance in CapCut PC.
tinyurl.com/2v23rn22
#capcutpc #capcutmarketing #capcutblackfriday #CapCutDidThat
🤝 Paid partnership
Pablo Rio retweeted
Avant, créer un modèle personnalisé nécessitait des données, du code et un ingénieur en machine learning.
Aujourd’hui, j’ai simplement décrit la tâche à FAiE et obtenu un petit modèle open source accessible via une API.
L’ingénieur en machine learning, c’était moi, avec une simple interface de chat.
🤝 Paid partnership
We're using AI agents to automate everything. Except the engineering of AI agents themselves.
That seemed backwards to us.
What if you could train your own specialized AI agent with just 15 minutes of your time instead of months of AI engineering?
Would you keep renting the same expensive generic models as your competitors, endlessly patching prompts and harnesses until you hit a wall?
Today, we're launching the agent that builds, trains, and continuously improves AI agents. 🚀
Meet FAiE, our Frontier AI Engineer. Describe what your agent should accomplish, and FAiE designs evaluations, diagnoses failures, curates data, selects training strategies, and runs experiments.
15 minutes of your effort. A few hours of automated AI engineering. Your own specialized agent, with its model weights trained for your task.
Already have an agent? Keep your harness. Connect it to @oumi_ai and train the model behind it with SFT, DPO, or GRPO (Private Preview).
The agent that builds your best agent.
Excited to see what you build: platform.oumi.ai/
Live demo today at 10 AM PT: oumi.ai/events/fall26
Pablo Rio retweeted
We’ve raised a $2.5M pre-seed to fix hardware hiring.
The most reliable way to know if an engineer can do the job is to watch them work.
In hardware, that’s been almost impossible to do at scale.
Colare runs simulation-based assessments that show how candidates actually work, in the tools they’d use on the job, before anyone books an interview.
We assess across:
- Mechanical (CAD, DFM, FEA)
- Electrical (PCB, schematics)
- Systems
- Manufacturing, & more.
Candidates solve problems from first principles.
Hiring teams see that across their whole pool, so engineers only interview people who’ve already cleared the bar.
We're already working with everyone from seed-stage startups to multi-billion-dollar aerospace manufacturers.
If you’re hiring hardware engineers, let’s talk:
colare.com/
Pablo Rio retweeted
The eye contact. The smirk. The dance break.
How are you supposed to pick just ONE bias moment? 🫠🖤
The CapCut plugin is coming soon to ChatGPT. At CapCut World, leading creators and OpenAI came together to explore new AI-powered video workflows.
Describe the highlight edit you want, and ChatGPT can use the plugin to control CapCut Web and help with supported editing tasks, giving you a starting cut to refine.
You still decide which moments deserve the spotlight and how the final edit should feel.
Currently in early access on ChatGPT desktop for PC. Mobile and web support is expected in mid-November.
#CapCutWeb #CapCutDidThat #ChatGPT #CapCutWorld
🤝 Paid partnership
When Tesla starts trending, my brain immediately switches to video ideas: a night drive through the city, a mountain-road sequence, or a sleek shot against a sunset.
The CapCut plugin is coming soon to ChatGPT. Describe your Tesla Roadster concept and work with CapCut Web to explore how to shape it into a video.
Early Access to the plugin is currently available through ChatGPT desktop (PC only).
#CapCutWeb #CapCutDidThat #ChatGPT #CapCutWorld
🤝 Paid partnership
Pablo Rio retweeted
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.
Kimi presenta K3.1 Pro y, de paso, una versión Highspeed anunciada como 2–3× más rápida.
Mi impaciencia por fin aparece en el roadmap de una empresa de IA.
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.
🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.
Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Open weights by Nov 4, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.ai/blog/kimi-k3-1-pro
Pablo Rio retweeted
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.
🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.
Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Open weights by Nov 4, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.ai/blog/kimi-k3-1-pro
Pablo Rio retweeted
2,8 trilhões de parâmetros. Só 104 bilhões ativos por token.
Eu tentando processar isso enquanto o Kimi K3.1 Pro chega com contexto de 1 milhão de tokens e pesos prometidos para 4 de novembro:
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.
🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.
Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Open weights by Nov 4, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.ai/blog/kimi-k3-1-pro














