Rafael Marañón

AI Product Leader | Amazon | Zoox | MIT | Co-founder

[Rafael Marañón @rafaelmaranon]

Built AV Triage, find the driving scenarios worth labeling before you label them. 160+ scenarios from @Waymo , AV2, nuScenes. Preview in 3D. Send to @encord_team with pre-drawn cuboid predictions. One click. https://t.co/WKKIzsIyAB

[Rafael Marañón @rafaelmaranon]

Live demo:
https://t.co/1aI75RFekZ

[Rafael Marañón @rafaelmaranon]

I built a small scenario validation lab using the Waymo Open Dataset.

Goal: understand how raw autonomous driving logs become the signals engineers use to make safety decisions.

Project:
https://t.co/r2uxbGGn84

A few lessons from building it ↓

[Rafael Marañón @rafaelmaranon]

Building safe autonomous systems isn't just about training models.

It's also about building validation infrastructure that turns driving data into decisions.

[Rafael Marañón @rafaelmaranon]

2. Metrics are product decisions

A “risk score” isn’t just math.

Someone has to decide:
• which thresholds matter
• how to measure exposure
• which signals engineers trust when reviewing scenarios

[Rafael Marañón @rafaelmaranon]

3. Evaluation tools must enable fast scenario scanning

The workflow becomes two-speed:

1) scan highest-risk scenarios
2) inspect them with playback + metrics

[Rafael Marañón @rafaelmaranon]

Check out my latest article: 💡 5 Hackatones = 5 Lecciones sobre IA https://t.co/03vzxmZ4Nm via @LinkedIn

[Rafael Marañón @rafaelmaranon]

🎬 AI is transforming Hollywood & robotics—reshaping storytelling, mobility & UX.

Robert Zemeckis' Here (2024) de-ages actors with AI, mirroring robotics' evolution. Novelty fades, real value wins. Ready for the next era?

🔗 Read more: https://t.co/0Ud0jI22wK

[Rafael Marañón @rafaelmaranon]

Over 200 creators from 40 teams pushed boundaries at the Multimodal AI Agents Hackathon finale!

🔍 Video-first AI apps are on the rise!
🔍 LLMs with voice, OCR & sensors.
🔍 Enterprise focus, now for more consumer innovation! #AI #Hackathon #Multimodal
https://t.co/IsABLnA7ob

[Rafael Marañón @rafaelmaranon]

One of the clearest explanations I’ve come across on how LLMs (DeepSeek, ChatGPT, Llama 3.1) work. Would love to see @karpathy cover VLMs and physical world foundational models next!

[Andrej Karpathy @karpathy]

New 3h31m video on YouTube:
"Deep Dive into LLMs like ChatGPT"

This is a general audience deep dive into the Large Language Model (LLM) AI technology that powers ChatGPT and related products. It is covers the full training stack of how the models are developed, along with mental models of how to think about their "psychology", and how to get the best use them in practical applications.

We cover all the major stages:
1. pretraining: data, tokenization, Transformer neural network I/O and internals, inference, GPT-2 training example, Llama 3.1 base inference examples
2. supervised finetuning: conversations data, "LLM Psychology": hallucinations, tool use, knowledge/working memory, knowledge of self, models need tokens to think, spelling, jagged intelligence
3. reinforcement learning: practice makes perfect, DeepSeek-R1, AlphaGo, RLHF.

I designed this video for the "general audience" track of my videos, which I believe are accessible to most people, even without technical background. It should give you an intuitive understanding of the full training pipeline of LLMs like ChatGPT, with many examples along the way, and maybe some ways of thinking around current capabilities, where we are, and what's coming.

(Also, I have one "Intro to LLMs" video already from ~year ago, but that is just a re-recording of a random talk, so I wanted to loop around and do a lot more comprehensive version of this topic. They can still be combined, as the talk goes a lot deeper into other topics, e.g. LLM OS and LLM Security)

Hope it's fun & useful!
https://t.co/75mXcUBI8L

[Rafael Marañón @rafaelmaranon]

Check out my latest article: From Oil Painting to AI: 3 Key Lessons on Iteration, Perception, and Creativity https://t.co/kvXe4f1TWM via @LinkedIn

[Marco Pavone @drmapavone]

This week we will be presenting 3 papers @eccvconf, on online mapping (@iamborisi), traffic scenario generation, and VLM-based AV stacks (@ChaoweiX):

Paper 1: https://t.co/fMcYLSIHeL
Paper 2: https://t.co/2dCMpE8CFr
Paper 3: https://t.co/aEqTZ3KHCF

More on online mapping 👇

[Magnific @magnific]

“Redesigning design with AI” it’s happening at our San Francisco office ✨

[Greg Brockman @gdb]

GPT-4o for education:

[Greg Brockman @gdb]

ChatGPT tabular data support:

[Jeff Bezos @JeffBezos]

Big milestone. Kudos and congrats to the whole Zoox team!

[Zoox @zoox]

Hello, Las Vegas! 👋 @CES is underway! We’re at booth 3574 in the West Hall. Stop by to see our purpose-built robotaxi and learn more about how we’re reimagining the future of mobility-as-a-service.

#CES2023 #Robotaxi

[Andreas BubenzerPaim @andreasbubenzer]

@amazon’s @zoox showing its autonomous robotaxi at #CES2023. I had seen this before at “TechCrunch Mobility 2022” but it’s even closer to commercialization now, having passed federal motor vehicle safety standards and awaiting approval from the DOT and California’s DMV.

[Rafael Marañón @rafaelmaranon]

Check out my latest article: It's Time for More System Dynamics https://t.co/T095KNtyAF via @LinkedIn