Marwan—appreciate you dropping by @hypr. The team had a blast showing you HYPRACTIVE™.
And thank you to your Fractal Capital: having seen what we’re building up close and choosing to invest! 🦾🔥
Drove through San Francisco with @TimKentleyKlay, co-founder and CEO of @hypr, while his AI, HYPRACTIVE, navigated downtown on 33 watts powering an NVIDIA Orin chip that runs HYPR's proprietary models built by just four engineers in two years, for under a million dollars.
Tim is a Melbourne-born designer turned robotics CEO who co-founded Zoox, the first company to operate a purpose-built, fully autonomous robotaxi and scaled it to $3.2B. He envisioned the car becoming a character in 2011, and spent the next decade and a half making that real. 125+ patents, $800M raised, and now CEO of HYPRLABS Inc - a small team coding a clean AI architecture and launching a completely new robot this summer.
The whole industry is throwing tens of billions at problems HYPR is solving with a computer the size of your fist and a hot-glued camera.
RLAF > RLHF
If anyone is interested in a map of what just happened @moltbook and what comes next: review the training structure @GoogleDeepMind used for AlphaStar.
Conditional imitation learning on human datasets ≈ human Gold-level player.
Creating a league for self-play / an evolutionary system for fitness selection on variation = superhuman performance.
An LLM (Large Language Model) in an AI-only chat group with upvotes and ladders is the RLAF (Reinforcement Learning Agent Feedback) head that tunes models beyond their bootstrapped human imitation and its limitations.
The play was always there, for those paying attention.
Getting this agents on-chain with https://t.co/ZZWmnUcCaX is the next best step.
Self-Driving on 33 Watts: How HYPR Labs Trained a Model for Just $850
@TimKentleyKlay, CEO & co-founder of @HYPR joined @gbrulte on The Road to Autonomy podcast to discuss how a team of four engineers achieved autonomous driving in San Francisco using just 33 watts of compute and an end-to-end neural network that prioritizes learning velocity over traditional simulation and mapping.
Episode Chapters
0:00 Introduction to HYPRDRIVE
1:30 HYPRDRIVE
5:40 Learning Velocity
8:10 Building HYPR
12:23 Training the System
18:55 The Origins of the HYPR Approach
21:36 Building Trust
23:35 Simulation 2
7:07 $850 to Train the Model
30:44 HYPR Robots
33:22 Cameras
35:16 What's Next
Introducing HYPRDRIVE™ our real-world continuous learning AI stack built on our belief that robots learn best when they learn as they move 🦾🎉
Check it out at https://t.co/BPJMxvNaFn
.@realGeorgeHotz openpilot requires manual braking and turning for corners? Here’s my @hypr crew with our AI HYPRDRIVE driving a @tesla, delivering far better self-driving than @comma_ai ever shipped.
We are excited to announce that Blackbird has led an equity financing of $5.55M in robotics startup @HYPR.
Who is behind it? @TimKentleyKlay – the co-founder of @zoox. As the first check writer into Zoox, it is great to have Tim back in the BB fam 💜
👀 https://t.co/JlHd5FXeH3