
Tech, Innovation and Research in Artificial Intelligence, computer vision, deep learning and learning distance in @UTPL #AI #computervision #iHCI_UTPL #IA_UTPL
Joined April 2009
- Tweets10.5K
- Following982
- Followers691
- Likes19.8K
Rod BG 👨👩👧👧 retweeted
Reinforcement learning learns from rewards, not labels.
Each step the agent sees a state sₜ (a shape), picks an action aₜ (a bin), and the environment returns a reward rₜ₊₁: +1 if right, −1 if wrong.
The agent keeps a value Q(s,a) and updates it with Q ← Q + α[r + γ maxₐ′ Q(s′,a′) − Q]. Each sort ends the episode, so the max term is 0 and Q ← Q + α(r − Q). With α = 0.5 a correct pair goes 0.5, 0.75, 0.875, … → 1.
Rod BG 👨👩👧👧 retweeted
Toma el control total de tus presentaciones y videos. 🎬✨
Gemini Notebook estrena la creación guiada para que el resultado final sea exactamente el que buscas:
✔️ Elige el tema basándote en tus propios documentos
✔️ Define la duración, el idioma y el estilo visual
✔️ Edita el esquema preliminar antes de generar el archivo
Disponible para suscriptores Pro y Ultra.
Empieza a crear aquí 👇
notebooklm.google.com
Rod BG 👨👩👧👧 retweeted
🇪🇨🇸🇪 ¿Te mudas a Suecia? Empieza aquí.
Francisca Cortés e Isis Chustek, dos profesionales latinoamericanas que viven en Suecia, colaboraron con nuestra Embajada para crear “Rumbo a Suecia”: una guía en español claro para quienes llegan del Ecuador a este país.
Qué incluye:
📄 Permisos de residencia
🆔 Número personal
🏥 Salud, trabajo y seguridad social
🏠 Vivienda, educación y banca
📞 Números de emergencia y contactos consulares
También nos ayudaron a renovar la página web de la Embajada. Cada trámite consular muestra ahora sus requisitos, y las citas se agendan en línea.
Creo que una embajada puede ser un catalizador para quienes quieren aportar a su comunidad. Así se ve en la práctica.
La guía es un punto de partida, así que confirma siempre con las autoridades suecas.
📘 Guía: cancilleria.gob.ec/suecia/20…
🌐 Página web: cancilleria.gob.ec/suecia/
Si estás construyendo con IA, guarda esto.
GitHub → control de versiones (gratis)
Claude → coding ($20/mes)
Namecheap → dominio ($12/año)
Cloudflare → DNS (gratis)
Vercel → deploy (gratis)
Clerk → autenticación (gratis)
Supabase → backend + base de datos (gratis)
Upstash → Redis (gratis)
Pinecone → vector DB (gratis)
Resend → emails (gratis)
Stripe → pagos (2.9% por transacción)
PostHog → analytics (gratis)
Sentry → error tracking (gratis)
Costo total para correr tu startup: ~$20 al mes.
Sin servidores.
Sin equipo de DevOps.
Sin necesidad de funding.
Solo una idea y conexión a internet.
Nunca había sido tan barato construir con IA.
Guarda este post.
Te va a servir cuando quieras lanzar tu próximo proyecto.
Rod BG 👨👩👧👧 retweeted
Egg line anomaly detection, now running locally 🥚 with no cloud API calls:
• FoundAD (DINOv3-based) anomaly det model fine-tuned on @roboflow with clean ("good") eggs only, runs on the edge
• heatmaps show where egg shells looks off
• RF-DETR tracks every egg, scored on every frame
• operator sets threshold, frame votes and severity bands per line, with no retraining needed
Rod BG 👨👩👧👧 retweeted
this is fun!! building an ai workout companion that gives me real time feedback while I work out at home
Rod BG 👨👩👧👧 retweeted
Bread counting on a conveyor belt using @ultralytics YOLO26 🥪
This system can offer a detailed report on bread quantity, enhancing production and inventory management.
You can also use it for multi-object detection to distinguish among various products, sizes, and packaging types.
#MLops #MachineLearning #Logistics
Rod BG 👨👩👧👧 retweeted
A Curva da Felicidade.
O economista David G. Blanchflower analisou dados de 132 países para investigar a relação entre idade e bem-estar subjetivo. O trabalho reuniu 109 conjuntos de dados e encontrou evidências de uma curva em U em todos os países analisados.
A curva sugere que:
- Na juventude: o bem-estar tende a ser relativamente elevado.
- Na meia-idade: a satisfação com a vida diminui, em média.
- A partir de aproximadamente 47 a 48 anos: o bem-estar tende a se recuperar, segundo as estimativas médias do estudo.
Créditos: IG / longevitydaily_
this is pure f*cking treasure
Leading AI researchers (Anthropic, Google, OpenAI) found how to orchestrate Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5, and GPT-6.1 Sol together, and it completely breaks the agentic context rot wall
most developers try to combine frontier models by chaining prompts in one context window or hardcoding rigid retry loops. you burn tokens on context bloat and freeze subagents into contradictory schemas that hallucinate within 5 iterations
their cognitive architecture eliminates monolithic prompt chaining and splits the agentic stack:
> syntax triage: Claude Haiku 5.5 elides raw AST syntax into typed interfaces (<45ms, -92.1% token compression) before prompt injection
> candidate proposer: GPT-6.1 Sol explores high-entropy code variations inside isolated execution sandboxes (+15.47% win rate)
> code synthesis: Claude Sonnet 5.5 acts as the dedicated execution workhorse (70.6% terminal-bench) generating verified AST diffs
> causal supervisor: Claude Opus 5.5 anchors the 1M context outer loop, coordinating multi-turn verification and semantic state
> zero context rot: Ebbinghaus exponential decay (tau=0.35) and Byte-0 prefix locks purge degraded noise while retaining 99.2% memory
the benchmark metrics outclass standard agent baselines:
> 91.4% reduction in token burn: Byte-0 KV-cache locking delivers 99.2% prefix hits
> +31.4% net increase in verified PR merge rate across 184,200 multi-turn sessions
> 78% lower latency by delegating pre-execution syntax checks away from the frontier reasoning models
> 0% context rot: non-volatile semantic ledger prevents hallucinations from polluting the persistent graph
Haiku strips the syntax. Sol explores the sandboxes. Sonnet writes the diffs. Opus guards the graph
you stop treating context windows like infinite scratchpads. you let frontier models orchestrate into a typed cognitive mesh
Rod BG 👨👩👧👧 retweeted
¿Se pueden usar robots inteligentes para ensamblar hardware de IA complejo?
En el Laboratorio de Robótica de NVIDIA en Seattle, están abordando esa pregunta mediante el ensamblaje de bandejas de pruebas del superchip GB300, desde el fijado de barras colectoras hasta la inserción de conectores delicados.
Para lograrlo es importante contar con que el ensamblaje robótico requiere más que solo precisión y aquí es donde la estimación de poses especializada entrenada con Isaac Sim y Reconstrucción Neural ayuda a los robots a localizar con precisión los conectores de cables. Las pinzas impresas en 3D personalizadas los guían hacia poses de agarre repetibles y el aprendizaje por refuerzo en NVIDIA Isaac Lab entrena las políticas de inserción de conectores.
Por lo tanto, la simulación inicia el aprendizaje y la interacción con el mundo real ayuda a refinarlo.
Una buena demostración de cómo la percepción, el diseño mecánico y el aprendizaje robótico se combinan mientras se trabaja para cerrar las brechas de confiabilidad y tiempo de ciclo para el despliegue de robots en fábricas.
Rod BG 👨👩👧👧 retweeted
Can you turn a regular camera view into a bird’s-eye view? 👀
Using @ultralytics YOLO26 + homography in Python, I built a demo that tracks vehicles and maps their movement onto a top-down view.
Useful for traffic monitoring and movement analysis.
#BirdEyeView #MachineLearning #AI
Rod BG 👨👩👧👧 retweeted
Pose inference just got more reliable.🧍
Ultralytics Inference v0.0.49 adds support for pose models that export 2-channel keypoints (x/y coordinates only), so they can now be processed and visualized correctly without requiring confidence scores.
Read more ➡️ github.com/ultralytics/infer…
#Pose #Ultralytics #Keypoints
Rod BG 👨👩👧👧 retweeted
Count vehicles in parking areas with Ultralytics YOLO26! 🚗
Detect and track cars in real time to monitor occupancy, measure flow, and automate parking analytics using region- or line-based counting.
Explore more ➡️ docs.ultralytics.com/guides/…
#Ultralytics #MachineLearning #ParkingLot
Rod BG 👨👩👧👧 retweeted
CPU vs GPU vs TPU vs NPU vs LPU, explained visually:
5 hardware architectures power AI today.
Each one makes a fundamentally different tradeoff between flexibility, parallelism, and memory access.
> CPU
It is built for general-purpose computing. A few powerful cores handle complex logic, branching, and system-level tasks.
It has deep cache hierarchies and off-chip main memory (DRAM). It's great for operating systems, databases, and decision-heavy code, but not that great for repetitive math like matrix multiplications.
> GPU
Instead of a few powerful cores, GPUs spread work across thousands of smaller cores that all execute the same instruction on different data.
This is why GPUs dominate AI training. The parallelism maps directly to the kind of math neural networks need.
> TPU
They go one step further with specialization.
The core compute unit is a grid of multiply-accumulate (MAC) units where data flows through in a wave pattern.
Weights enter from one side, activations from the other, and partial results propagate without going back to memory each time.
The entire execution is compiler-controlled, not hardware-scheduled. Google designed TPUs specifically for neural network workloads.
> NPU
This is an edge-optimized variant.
The architecture is built around a Neural Compute Engine packed with MAC arrays and on-chip SRAM, but instead of high-bandwidth memory (HBM), NPUs use low-power system memory.
The design goal is to run inference at single-digit watt power budgets, like smartphones, wearables, and IoT devices.
Apple Neural Engine and Intel's NPU follow this pattern.
> LPU (Language Processing Unit)
This is the newest entrant, by Groq.
The architecture removes off-chip memory from the critical path entirely. All weight storage lives in on-chip SRAM.
Execution is fully deterministic and compiler-scheduled, which means zero cache misses and zero runtime scheduling overhead.
The tradeoff is that it provides limited memory per chip, which means you need hundreds of chips linked together to serve a single large model. But the latency advantage is real.
AI compute has evolved from general-purpose flexibility (CPU) to extreme specialization (LPU). Each step trades some level of generality for efficiency.
The visual below maps the internal architecture of all five side by side.
Notice the thread connecting all five. Every generation exists to move data less, because the math was never the hard part. Feeding the math units fast enough is.
The same battle plays out one layer up in software. During LLM inference, a single GPU produces terabytes of KV cache per day, and nearly all of it gets thrown away and recomputed, which is a big reason agent workloads cost what they do.
I wrote a full breakdown of how a disaggregated caching layer fixes this, with up to 14x faster time-to-first-token. The article is quoted below.
You should also check the @lmcache GitHub repo: github.com/LMCache/LMCache
(don't forget to star 🌟)
👉 Over to you: Which of these 5 have you actually worked with or deployed on?
Rod BG 👨👩👧👧 retweeted
自動運転・運転支援向けに、CPUだけで周囲の状況を認識し、注意すべき対象を判断するAI🚗💻
🔹歩行者・自転車・車両や走行可能な範囲を認識
🔹対象の位置や動きから危険度を評価
🔹処理ごとに頻度を調整し、CPUだけで効率よく動作
開発:Affan Khan氏
情報元はリプ欄。
Rod BG 👨👩👧👧 retweeted
New tutorial | Household object detection with Ultralytics YOLO26 🏠
Watch how to train a custom YOLO26 model to recognize everyday household objects, then take it from the Ultralytics Platform to local Python inference.
Watch here ➡️ youtu.be/D3W9SuEWDOk
#YOLO26 #ObjectDetection #Indoor #Research
Rod BG 👨👩👧👧 retweeted
標準CCTV映像から交通流を解析するAIシステム「PathVision」🚗🚦
🔹YOLOv11+ByteTrackで車両を検出・追跡
🔹実速度・OD行列・混雑状況をリアルタイム分析
🔹遮蔽時のID維持やトリップ完了判定にも対応
開発:Anmar Nihad氏
情報元はリプ欄。
Rod BG 👨👩👧👧 retweeted
Graph Neural Networks for Molecular Discovery with Python is a 326-page resource focused on applying Graph Neural Networks (GNNs) to molecular discovery.
It covers topics such as:
Graph representation of molecules
Geometric deep learning
Molecular property prediction
Molecule generation
Python-based implementations
Applications of GNNs in drug and chemical discovery
Free PDF: clcoding.com/2026/07/graph-n…
Rod BG 👨👩👧👧 retweeted
15 melons counted in seconds using @ultralytics YOLO26! 🍈
A satisfying look at computer vision in fruit sorting. Automated counting can help packing teams track quantities and reduce manual work.
Small task, practical value.
#ComputerVision #AI #Fruits





















