Digest

Brett Adcock, founder and CEO of Figure AI, discusses the development of general-purpose humanoid robots, highlighting the challenges and advancements in hardware, AI, and control systems. He contrasts electric humanoid robots with older hydraulic models, emphasizing the efficiency and safety of electric systems. Adcock also touches upon his previous ventures, including Archer Aviation, an EVTOL company, and Vettery, an AI-powered recruiting marketplace. The conversation delves into the future of AI, predicting an age of abundance driven by productivity gains and the potential for robots to automate mundane tasks, freeing humans for more meaningful pursuits. He addresses safety concerns, the complexity of AI development, and the importance of data in training neural networks. The discussion also covers the potential of Terahertz imaging for school safety and the future of AI devices beyond smartphones, with the launch of his new AI lab, Hark. Adcock shares insights into entrepreneurship, the challenges of innovation, and the optimistic outlook for a robot-assisted future, while also acknowledging the potential risks of malfunctioning or misused machines.

Outlines

00:00:00

Introduction of Brett Adcock and His Ventures

Introduction of Brett Adcock, founder and CEO of Figure AI, a serial entrepreneur with a background in founding and leading companies in AI talent marketplaces, electric aviation, and AI security.

00:03:03

Addressing AI Safety Concerns in Robotics and Early Ventures

Discussion on the potential for AI platforms to invent answers and the risks when combined with physical robotic actions, focusing on safeguards against harm from bad programming or errors. This section also covers early ventures like Vettery, an AI-powered recruiting marketplace, and the challenges of traditional recruiting.

00:03:47

Evolution of Humanoid Robots and AI Strategy

The evolution from hydraulic, hand-coded robots to cheaper, electric humanoids utilizing AI-first strategies with neural nets, highlighting the rapid advancement in the field.

00:04:52

Challenges in Building Safe and Affordable Humanoid Robots

The immense challenge of creating affordable, mass-producible humanoid robots capable of complex household tasks autonomously and safely, requiring advancements in hardware, AI, and safety protocols.

00:06:21

Humanoid Robots in the Home and Future Vision

Brett Adcock shares his experience testing humanoid robots in his home, including interactions with his children, discussing the current limitations and future potential for integration. A prediction that humanoid robots will become as common as phones and cars in the near future, transforming daily life and labor automation.

00:08:24

Multifaceted Challenges of Robotics Development and AI's Potential

Exploration of the numerous challenges in robotics, including hardware, AI, scaling, commercialization, reliability, and manufacturing. Brett Adcock firmly believes AI is not in a bubble, predicting transformative technological events and emphasizing that we are only scratching the surface of AI's potential.

00:09:34

The Dawn of Physical AI and Age of Abundance

The current stage of AI is early, with a focus on integrating AI into the physical world. AI's ability to use computers and think like humans will lead to unprecedented productivity gains, potentially creating an age of abundance by reducing the cost of goods and services.

00:10:50

Redefining Human Roles and Delegating Life's Tasks

Discussion on how automation will free humans from mundane tasks, allowing them to focus on more meaningful activities, creativity, and personal growth. The concept of exporting one's brain and delegating daily tasks, both digital and physical, to AI and robots to manage life efficiently.

00:14:15

Brett Adcock's Early Life and Entrepreneurial Journey

Brett Adcock's upbringing in rural Illinois, his early interest in computers, and his first entrepreneurial ventures selling products online. The transition from a third-generation farming family to a career in AI and robotics, highlighting the entrepreneurial spirit instilled from a young age.

00:18:36

Sibling Success and Parental Influence in Entrepreneurship

Brett's brother, Colby, runs an AI defense company, showcasing a shared trajectory into advanced technology from a farming background. The impact of parents emphasizing entrepreneurship and self-reliance, encouraging the pursuit of running one's own business from a young age.

00:20:38

University, Early Career, and Founding Vettery

Attending the University of Florida and subsequently moving to New York to work on software startups, building on early internet ventures. The creation of Vettery, an AI-powered recruiting marketplace designed to fix the broken job application and hiring process through algorithmic matching.

00:22:44

The Journey and Acquisition of Vettery

Discussion on the exorbitant fees and inefficiencies of traditional headhunters and recruiting firms, highlighting the need for a more streamlined solution. The challenging yet ultimately successful journey of Vettery, including overcoming financial difficulties and its eventual acquisition for $110 million.

00:26:11

Transitioning to New Ventures: Robotics and Aviation

After selling Vettery, Brett Adcock took a year to explore new opportunities in robotics, aviation, and security, driven by a desire to tackle significant challenges.

00:26:57

Founding Archer Aviation: The EVTOL Vision

The inception of Archer Aviation, focused on developing electric vertical takeoff and landing (EVTOL) aircraft to revolutionize urban transportation. Explanation of how EVTOL aircraft combine vertical takeoff capabilities with airplane-like flight, enabling electric power for cost and safety benefits.

00:28:50

Building Confidence and Deep Dive into Electrification

Brett Adcock discusses how his experience in software engineering provided the foundation and confidence to tackle complex hardware challenges in aviation. Brett's self-study of electrification, vertical lift, and winged aircraft, including intensive reading and specialized courses to master EVTOL design.

00:30:32

The Genesis of Archer Aviation and Rapid Growth

The development of an algorithm to determine the optimal design for electric aircraft. A chance encounter at an engineering course led to the collaboration and founding of Archer Aviation, initially operating out of a university lab. Archer Aviation's swift progress, including going public within three years and becoming a multi-billion dollar company, despite initial funding challenges.

00:35:24

Navigating the SPAC Process and VC Landscape

The difficulties of going public via a SPAC, including raising capital and facing a lawsuit from established players. Venture capital's general reluctance to fund hardware and deep tech companies, contrasting with the current investment climate.

00:36:59

Personal Investment and the Engineering of Archer's Flying Robot

Brett Adcock's significant personal financial investment in Archer Aviation during its IPO, highlighting the stressful period and the company's subsequent growth. Detailed breakdown of Archer's aircraft as a \"flying robot\" with complex systems like batteries, motors, software, and 24 degrees of freedom.

00:39:28

Altitude, Safety Standards, and Government Regulation

Discussion on the operational altitude of Archer's aircraft (2-3,000 feet) and the stringent safety standards required by the FAA for passenger transport. The perceived lag of government regulation behind advancements in AI and electric vehicles, impacting the widespread adoption of new technologies.

00:40:06

AI's Potential in Urban Mobility and Autonomous Vehicles

The possibility of AI optimizing traffic flow and reducing congestion in cities, even without widespread adoption of air vehicles. The significant time required to replace the global fleet of cars with electric and autonomous vehicles, highlighting the challenges of retrofitting and infrastructure.

00:41:16

The Transitory Phase of Autonomous Driving and Waymo Experience

Acknowledging the current \"messy\" and transitional phase of autonomous driving technology, with the expectation of full autonomy and safety within five years. Personal experience with Waymo's autonomous ride-sharing service in the Bay Area, highlighting its reliability, human-like driving, and positive user experience.

00:42:53

The Inevitable Rollout of Scaled Autonomy and FAA Certification

The widespread adoption of autonomous technology is inevitable but will take time due to the need for mature technology, sufficient infrastructure, and a larger install base of autonomous vehicles. Archer's aircraft are operational and flying weekly in California, but passenger service requires FAA type certification, a slow and rigorous process.

00:43:56

FAA Certification and the Early Stages of Flying Cars

The FAA's rigorous safety standards (one in a billion hours reliability) for aircraft certification, emphasizing the commitment to making air travel exceptionally safe. Flying cars and electric aircraft are in their early stages, predating autonomous and electric vehicles, but their widespread adoption is anticipated within our lifetime.

00:45:53

The Three-Dimensional Future of Urban Transportation and Vertiports

The future of urban travel lies in three-dimensional airspace and underground tunnels, offering solutions to the limitations of 2D road networks. Takeoff and landing for flying cars will occur at designated \"vertiports\" due to noise and infrastructure requirements, not from individual homes.

00:48:31

Air Travel's Advantage and Shifting Travel Patterns

Air travel offers greater network flexibility, allowing travel between any nodes, unlike tunnels which are limited to point-to-point routes. Journeys exceeding 20 minutes will likely move to the sky, enabling people to live further from cities and commute quickly, reshaping urban and suburban living.

00:49:58

On-Demand Air Travel and Electrification's Role

Flying cars will operate as an on-demand Uber-like service, accessible via an app, offering affordable and rapid transportation between vertiports. Electrification significantly reduces the cost and safety burden of aircraft compared to traditional helicopters, with fewer critical components and greater reliability.

00:51:14

Test Pilots' Perspective and Archer's Strategic Position

Test pilots at Archer Aviation are enthusiastic about the electric aircraft, viewing it as the future of aviation and a testament to its functionality. Archer's current advantage lies in being in the right decade for electric aviation, with working aircraft, ongoing FAA certification, and a strong financial and team foundation.

00:52:42

The Genesis of Figure AI: Pursuing General-Purpose Humanoid Robots

The motivation behind Figure AI: to build general-purpose humanoid robots capable of performing a wide range of tasks, inspired by the human form factor. Humans build their environment around their form factor, suggesting that a humanoid robot, mirroring this form, is the most effective general-purpose machine.

00:53:58

The Humanoid Robot: Form, Function, and the Leap to Electric

The ultimate goal in robotics is a general-purpose machine that can do what humans can, which is embodied in the concept of a humanoid robot. A humanoid robot possesses legs for mobility, arms and hands for manipulation, and sensors for environmental understanding, enabling it to perform real-world tasks. The feasibility of building electric humanoid robots, driven by advancements in electric motors, battery systems, and control software, offering advantages in cost, safety, and performance.

00:55:04

Contrasting Humanoid Robots and Control Complexity

Comparison of Boston Dynamics' hydraulic Atlas robot with electric humanoids, highlighting the limitations of hydraulics (leaks, short battery life) versus the benefits of electric systems. The immense complexity of controlling a humanoid robot with 40 degrees of freedom, leading to more possible states than atoms in the universe, making traditional coding impossible.

00:56:54

High-Frequency Control, Dynamic Stability, and Advanced Sensing

The necessity of running control software at over 200 times per second for balance and task execution in humanoid robots, requiring dynamic and responsive feedback systems. Humanoid robots require constant dynamic balancing, unlike statically stable quadruped robots, making even simple movements like reaching a complex challenge. Humanoid robots utilize position encoders, force sensing, and torque sensing to gather detailed feedback, which is processed by a central computer for precise control.

00:58:42

Shopify's Role and Figure AI's Ambitious Goal

Shopify is presented as a partner that alleviates the fear and challenges of starting new ventures, providing tools and expertise for e-commerce businesses. Figure AI's mission to rebuild humanoid robots from the ground up, betting on the possibility of creating functional, general-purpose machines in the near term.

01:01:49

Overcoming Skepticism and Rapid Team Expansion at Figure AI

Initial difficulties in gaining excitement for Figure AI due to the lack of precedent in AI-humanoid integration and electric humanoid hardware. Figure AI was initially self-funded, rapidly building a 40-person team and establishing core competencies in hardware and software development.

01:02:40

Engineering Pillars of Robotics Development and Motor Challenges

Key engineering areas for robotics include electric motors, battery systems, control software, embedded systems, and sensors, each with intricate sub-components. Electric motors are most efficient at high speeds, posing an engineering challenge for humanoid robots that require high torque at low or zero speeds for tasks like standing and holding.

01:04:02

Specialized Teams and Figure One's Achievement

Dedicated teams focus on specific aspects of motor design, including rotor, stator, gearbox, sensors, and control systems, to overcome engineering hurdles. Figure AI successfully developed and demonstrated their first-generation robot, Figure One, walking autonomously in under 12 months, a record achievement.

01:05:11

Figure Two, Neural Networks, and Robot Control

The development of Figure Two, the second-generation robot, and a significant demonstration of performing tasks like making coffee using only neural networks, without traditional coding. The groundbreaking achievement of using neural networks to process camera input and output motor trajectories for humanoid robots, eliminating the need for explicit coding.

01:06:35

The Dawn of Humanoid Robots and AI Advancements

The speaker discusses the pivotal moment in 2023 when advancements in hardware and neural networks made humanoid robots a viable reality. The focus shifted to building cheap, reliable hardware capable of human-like tasks, powered by neural networks rather than traditional coding.

01:07:31

Figure's Evolution: From Figure 2 to Figure 3 and Real-World Deployment

This section details the progression of Figure's humanoid robots, from the successful deployment of Figure 2 in real-world industrial settings to the development of Figure 3, touted as the world's best humanoid hardware, capable of tasks like unloading dishes and doing laundry. The speaker emphasizes the challenge of deploying robots in client environments, focusing on achieving human-level Key Performance Indicators (KPIs) for speed and performance, particularly in manufacturing settings.

01:08:57

BMW Manufacturing Pilot and Automation Integration

The transcript highlights a significant milestone: Figure's robots were integrated into a BMW manufacturing line in January 2025, assisting in building parts of the X3 and X5 models, demonstrating their capability in a demanding industrial environment. The speaker describes the advanced automation within BMW's car manufacturing, noting the presence of massive robotic arms and sophisticated end-effectors. Figure's robots played a role in attaching the rear header on the body shop line.

01:12:06

Testing Durability, Scalability, and Software Refactoring

The primary goal of the BMW pilot was to test the robots' daily operational capability and durability over an extended period. A key learning was the robot's consistent performance, with the same unit completing the entire six-month shift. The speaker rates the hardware performance as \"A+\" but software as \"B/B+\". The initial software architecture relied on a mix of traditional code and neural networks, with code-based controllers proving to be a bottleneck for scalability and adaptability. To overcome software limitations, Figure refactored its entire stack, including controllers, into a neural network architecture, now known as Helix 2. This significantly reduced the codebase and improved adaptability.

01:15:24

Current Deployments, 24/7 Operations, and Office Use Cases

The company is preparing for customer deployments with robots now running 24/7 shifts without faults for days. They've achieved record uptime, with seamless robot-to-robot handoffs for charging and continuous operation. At their office, robots are used for 24/7 logistics, moving packages, and as greeters/visitor bots. These bots navigate the office, interact with people, and are being developed to handle more complex tasks like escorting visitors.

01:17:58

Deep Memory, Conversational AI, and Advanced Authentication

A key focus is developing \"deep memory\" for robots to personalize interactions. Unlike current conversational AIs, these robots will remember users, their permissions, and preferences, enabling more meaningful and context-aware conversations. The discussion touches on the need for advanced authentication beyond voice recognition for tasks involving spending money or sensitive actions, suggesting facial recognition or fingerprint scanning. Speech remains the primary desired modality for robot interaction.

01:20:14

5G Integration, Text Commands, and the \"App Store\" Model

All robots are equipped with 5G, enabling seamless communication and control via text messages. This allows users to issue commands remotely, such as preheating an oven or managing household tasks. The vision is for humanoid robots to operate like smartphones, where new capabilities can be added by downloading \"apps\" (neural network weights) without needing hardware changes, enabling tasks like complex folding or making coffee.

01:23:12

Commercial vs. Consumer Deployment Strategy and Home Safety

Figure will initially ship robots to businesses due to lower variability in work environments compared to homes. Homes present a \"chaos\" of unpredictable elements and diverse appliances, making them a more challenging environment for robotics. The biggest hurdle for home deployment is ensuring safety and building trust, especially around children. This requires robust system safety engineering and a gradual build-up of confidence in the robots' reliability.

01:25:34

Economic Viability and the \"Holy Grail\" of General Purpose Robotics

Commercial markets can support higher pricing (around $500/month) compared to the home market, which would need to be significantly cheaper. The vastness of human labor in the commercial sector ($3 trillion GDP contribution) makes it the primary target. The speaker believes the first company to achieve a general-purpose humanoid robot capable of performing diverse tasks in a home environment without prior specific training will become the world's largest company.

01:28:51

Training Robots and Rapid Progress in Robotics

Training a robot for home use will involve a process similar to showing a friend around your house, explaining preferences and routines. The robot will semantically learn and remember these instructions. Figure has made significant progress in just three to four years, compressing what was expected to take a decade. They are close to solving general-purpose robotics, with a clear architectural path and hardware in place.

01:31:40

Safety Concerns and the Figure 3 Humanoid Robot

The speaker draws a parallel to aircraft safety, stating they will only consider robots safe for children when they would personally fly an aircraft like an Archer with their own kids. This signifies the high bar for safety and trust. A detailed look at the Figure 3 robot: 5'6\", 135 lbs, designed for tasks like folding laundry and manufacturing. Improvements include reduced weight, enhanced sensors, fifth-generation hands with cameras and tactile sensors, and more onboard compute for the Helix neural network.

01:36:11

Safety Features, Durability, and Internal Components

Figure 3 incorporates safety features like a foam layer. While durable, the robot's ability to withstand falls depends on the impact. It's designed to get back up and continue working, though some falls can cause damage. The robot's torso houses the battery, GPUs, computer, and power distribution – essentially the \"brains and energy.\" It has 40 electric motor joints with integrated sensors for balance and operation.

01:37:47

Neural Network-Driven Movement and Advanced Hand Dexterity

All walking and movement are controlled by neural networks, not traditional code, showcasing the advanced AI integration. The hands feature cameras and tactile sensors on each fingertip, allowing for precise grip and force sensing. This has been a four-year development effort, with a next-generation hand promising human-level dexterity.

01:39:14

Versatile Gripping Capabilities and Battery Life

The robot's hands can handle tasks ranging from picking up 40-pound boxes to delicately folding a t-shirt, demonstrating a wide range of manipulation capabilities. The robot holds a charge for 4-5 hours and takes about an hour to recharge. Inductive charging through the feet allows for wireless charging on mats, with a charging rate of approximately 1kW per foot.

01:40:23

Consumer Market Availability and Key Commercial Customers

Consumer availability hinges on achieving robust performance, particularly in unpredictable home environments. Robots are being shipped to commercial customers soon, with home deployment following once reliability is proven. Figure is working with major companies like BMW, a leading logistics firm, and Brookfield (real estate). More customer announcements are expected in the coming months.

01:41:11

Production Capacity and Future Goals

Their manufacturing facility can produce about one robot every 90 minutes, with a goal of reaching 40-50,000 units annually at full capacity and scaling to a million units per year within the decade.

01:42:37

Rapid Development Cycle and Next-Generation Hand Capabilities

The company has achieved significant milestones in just three and a half years, with walking robots developed within the first three years. They aim for dramatic improvements with each new robot generation annually. The upcoming hand design will feature human-level dexterity, matching the number of joints in a human hand. This advancement is crucial for tasks requiring fine motor skills.

01:46:02

Partnership with OpenAI and Strategic Decision to Part Ways

OpenAI, led by Sam Altman, co-led Figure's Series B funding round. They collaborated for a year on advancing AI models for humanoid robots, aiming to integrate advanced language models for robotic action output. Figure ultimately parted ways with OpenAI due to internal team capabilities and the distinct challenges of robotics AI versus language model development. Figure's internal AI team proved more effective in driving progress.

01:50:23

OpenAI's Robotics Roots and Future Re-engagement

OpenAI initially focused on AGI through robotics before pivoting to large language models. Their early work in robotics from 2016-2020 suggests a potential future re-engagement with the field.

01:52:14

Strategic Decision to Split from OpenAI and Military Applications

The decision to split from OpenAI was driven by Figure's internal team's superior progress in robot learning and the challenges of collaboration when facing potential future competition. Figure has chosen not to pursue military applications to maintain focus on commercial and home markets. While acknowledging the potential benefits, they believe pursuing both simultaneously is too difficult.

01:56:18

Hesitancy Towards Humanoid Robots and the Goal of General Purpose Robotics

The hesitancy surrounding humanoid robots is attributed to their human-like appearance and ability to perform human-like tasks, creating fear of job displacement, similar to concerns around digital AI. Figure's ultimate goal is to solve general-purpose robotics, enabling robots with common-sense reasoning that can be deployed in any home and communicated with via speech.

01:58:32

Data as the Key Constraint and Robot-to-Robot Communication

The primary constraint in achieving general robotics is acquiring sufficient and appropriate data at scale to train the Helix stack. With the right data, they believe general robotics could be solved now. Robots communicate autonomously for tasks like seamless handoffs during charging or when one robot needs to go for maintenance. This inter-robot communication is crucial for 24/7 operations.

02:02:16

Robustness and Future: Robots Building Robots

Robots have become incredibly robust, capable of continuing operations even with a lost motor, by stiffening the joint and \"limping\" to a maintenance area. This resilience is a recent development. The future vision includes robots building other robots, further automating the production process and enabling widespread autonomous work.

02:03:27

The Bay Area as a Robotics Hub and Reshoring Manufacturing

The Bay Area is expected to see a significant increase in humanoid robots, potentially outnumbering humans in the next 10 years, driven by the concentration of robotics companies. Humanoid robots are expected to facilitate the return of high-end, transformative manufacturing to the US, focusing on complex, intelligent systems rather than low-skill production.

02:05:35

US-Based Manufacturing and Self-Sufficiency

Figure is manufacturing its robots in California, with a goal of increasing production capacity significantly. Robots are undergoing self-checkout and calibration, with a vision for them to eventually box themselves and be palletized by other robots.

02:07:24

Handling Compliant and Moving Objects with Neural Networks

The most challenging tasks for robotics involve handling compliant and moving objects like laundry or packages on conveyors, as opposed to stationary items. This is where AI and neural networks are crucial. The speaker discusses the effectiveness of neural networks in handling high variability and diverse data, citing examples like logistics and package handling where they perform exceptionally well.

02:09:44

Navigating the Department of War and Terahertz Imaging for School Safety

This segment highlights the complexities and lack of transparency in working with the Department of War, introducing SBIR Advisors as a solution for companies seeking government contracts. The speaker explains the development of Terahertz imaging technology, originally for military use, as a potential solution to detect concealed weapons in schools without causing alarm.

02:16:26

The Genesis of Cover Technology and Re-establishing Cover

The backstory of the \"Cover\" technology is revealed, detailing its development at JPL for war zones and its subsequent revival for school safety applications due to a renewed focus on the problem. The speaker discusses spinning out the \"Cover\" technology from JPL, the involvement of the original team, and the significant cost reduction achieved by integrating the technology into chips.

02:21:10

Affordability, Implementation, and Enhancing School Safety

The focus shifts to making the \"Cover\" technology affordable for schools, discussing budget allocations, cost reduction through chip integration, and the potential for widespread adoption. The discussion explores integrating facial recognition and other AI-powered features with the Terahertz imaging system to enhance school security by identifying individuals and potential threats.

02:23:37

The Scale of the School Gun Problem and Preventing Shootings

The speaker presents statistics on the number of guns brought into schools, the low rate of discovery and reporting, and the potential for the technology to significantly reduce gun-related incidents. The speaker emphasizes the critical need to prevent school shootings, highlighting the emotional impact and the potential for technology like \"Cover\" to act as a deterrent and save lives.

02:26:42

Advanced AI and Terahertz Technology for Threat Detection

This segment delves into the sophisticated AI and Terahertz technology used for threat detection, explaining its operational principles and the technical challenges involved in achieving high accuracy. The speaker discusses the specialized nature of Terahertz technology for security applications, noting the lack of a dedicated industry and the reliance on a passionate, expert team at \"Cover.\"

02:28:23

Future Rollout and Challenges of Cover Technology

Plans for beta testing \"Cover\" technology in schools are outlined, along with the challenges of marketing, managing false positives, and ensuring the technology doesn't alarm students.

02:29:28

How Terahertz Imaging Works for Weapon Detection

A detailed explanation of how Terahertz imaging functions, using radio frequencies to create 2D and 3D images of concealed objects, is provided, emphasizing its non-ionizing and safe nature.

02:30:42

Reconstructing 3D Point Clouds and Detecting Various Objects

The process of reconstructing 3D point clouds from Terahertz signals to identify weapons and differentiate them from human tissue is explained, highlighting the technology's ability to see through materials. The capability of RF technology to detect not just metallic objects like guns but also non-metallic items like knives and vape pens is discussed, along with the nuances of metallic signatures.

02:31:46

Data-Driven Insights into School Safety

The role of data scientists in analyzing school safety, including gun prevalence, types of weapons, and emergency responses, is discussed, emphasizing the need for accurate threat detection.

02:34:34

The Challenge of Humanoid AI Development and Limitations of Chatbots

The speaker describes humanoid AI as one of the most challenging AI technologies, detailing the four-year effort to develop advanced AI systems capable of complex tasks and natural interaction. Current AI chatbots like ChatGPT are criticized for their lack of personalization, memory, and ability to perform real-world tasks, highlighting the gap between current AI and futuristic expectations.

02:35:41

Reimagining AI Interaction and Introducing Hark

The speaker proposes rebuilding hardware from scratch to create a more intuitive and effective interface for advanced AI, moving beyond current phone and computer paradigms. The launch of \"Hark,\" an AI lab focused on designing the future of AI beyond the iPhone, is announced, emphasizing the development of multimodal models and revolutionary AI systems.

02:38:06

The Future of AI Devices and Personalization

The vision for \"Hark\" includes a family of devices that will replace phones and computers, offering always-on, personalized AI assistance, near-perfect memory, and coaching capabilities.

02:38:45

Hark's Progress and Self-Funding

Updates on Hark's progress, including working hardware and models, are shared, along with the fact that the venture is currently self-funded, aiming for a stealth launch soon.

02:39:36

Personal Motivation and the Entrepreneurial Grind

The speaker shares personal motivations for pursuing ambitious projects like flying cars and AI, emphasizing a desire to create a better future and work on \"cool, crazy shit.\" The realities of entrepreneurship are discussed, including the need for personal sacrifices, intense dedication, and the blurring lines between work and personal life.

02:41:40

From Farm Boy to Tech Innovator and the Path of Innovation

The speaker reflects on their journey from a farm background to leading groundbreaking tech ventures, highlighting the American dream and the importance of passion and perseverance. The speaker acknowledges the inherent difficulties and uncertainties in building innovative companies, emphasizing the need for resilience and a commitment to making projects succeed.

02:43:24

The Real Threat: Obedient but Malfunctioning Machines

The discussion shifts to the potential dangers of robots, arguing that the greater threat lies not in conscious machines, but in obedient ones that can malfunction, be hacked, or used for malicious purposes.

02:44:21

Optimism for a Robot-Assisted Future and Building Synthetic Humans

Despite potential risks, the speaker expresses optimism about a future with widespread humanoid robots, believing they can free up human time and significantly benefit society. The transformative potential of AI systems that can be embodied and interact with computers is discussed, framing it as the creation of \"synthetic humans at scale\" with both exciting and daunting implications.

02:46:12

The Inevitability and Necessity of AI Advancement

The speaker asserts that AI advancement is inevitable and necessary for societal progress, comparing its importance to that of cars and airplanes, and emphasizing the need to steer it positively. The speaker acknowledges the widespread fear surrounding AI but expresses hope for a better future, emphasizing that progress is unstoppable and requires conscious effort to guide it positively.

02:47:50

Doing AI the Right Way for Global Benefit and AI's Role in Advice

The speaker stresses the importance of developing AI responsibly to maximize its positive impact on the world, as stopping its progress is not a viable option. The potential for AI, including chatbots and future humanoids, to provide advice, including relationship and personal guidance, is explored, acknowledging existing trends and future implications.

02:48:46

Advice for Future Founders and Ambitious Ventures

Key advice for aspiring founders includes starting to build immediately, embracing continuous learning, and focusing on solving the hardest, most impactful problems with passion. The speaker advocates for tackling ambitious, high-outcome projects, arguing they are often not proportionally harder to execute but offer significantly greater rewards and attract better talent.

02:53:37

The Lonely Path of Entrepreneurship and the Great Filter

The speaker acknowledges the lack of a clear rulebook for entrepreneurship, the abundance of noise, and the often lonely journey of seeking advice, encouraging founders to trust their instincts. The concept of a \"great filter\" in entrepreneurship is discussed, emphasizing that only those with unwavering determination and a willingness to do \"whatever it takes\" will succeed.

02:55:48

Embracing the Journey and Reflecting on a Fascinating Interview

The speaker expresses excitement about the current stage of their career, viewing it as the \"early innings\" and focusing on scaling impactful systems, despite the inherent challenges. The interviewer expresses gratitude for the fascinating discussion and acknowledges the incredible advancements and work being done by the guest.

Keywords

Q&A

• What are the main challenges in developing humanoid robots?

  Key challenges include creating affordable and mass-producible hardware, developing sophisticated AI for complex tasks, ensuring safety and reliability, and overcoming the immense complexity of controlling a robot with many degrees of freedom.

• How is AI expected to change the future of work and productivity?

  AI is predicted to drive unprecedented productivity gains by automating mundane tasks, enabling humans to focus on creative and strategic work, and potentially leading to an age of abundance with reduced costs for goods and services.

• What is the significance of electric propulsion in aviation and robotics?

  Electric propulsion offers advantages in cost reduction, enhanced safety due to fewer critical components, improved performance, and a quieter acoustic signature compared to traditional systems, making it ideal for both aircraft and robots.

• What is the vision for urban air mobility and flying cars?

  The vision is for electric vertical takeoff and landing (EVTOL) aircraft to operate as an on-demand, Uber-like service, utilizing designated vertiports within cities to provide rapid, efficient, and sustainable transportation, alleviating ground traffic congestion.

• How does Figure AI plan to overcome the complexity of controlling humanoid robots?

  Figure AI is leveraging neural networks to process sensor data and output motor commands, bypassing traditional coding for tasks. This approach allows the robot to learn and adapt, managing the vast number of possible states and dynamic balancing required.

• What are the main challenges in developing general-purpose humanoid robots?

  The primary challenges include achieving human-level dexterity and manipulation, ensuring robust safety and reliability in diverse environments, developing sophisticated AI with common-sense reasoning, and acquiring sufficient data for training.

• How does Figure's Helix neural network stack differ from traditional robot programming?

  Helix relies heavily on neural networks for control and decision-making, moving away from traditional coded heuristics. This allows for greater adaptability, learning from experience, and handling complex, unpredictable situations more effectively.

• What is the envisioned role of humanoid robots in the future?

  Humanoid robots are envisioned to perform a wide range of tasks, from household chores and logistics to complex manufacturing and dangerous missions. They aim to augment human capabilities, automate labor, and improve quality of life.

• Why is data acquisition a critical bottleneck for advancing robotics AI?

  Neural networks require vast amounts of diverse and high-quality data to learn effectively. For general-purpose robotics, this means collecting data from countless scenarios, tasks, and environments, which is a complex and time-consuming process.

• What are the key considerations for ensuring safety when deploying robots in homes?

  Safety in homes requires rigorous system engineering, building trust through reliable performance, and ensuring robots can operate without posing a risk to children or adults. This involves advanced perception, control, and fail-safe mechanisms.

• How does Figure plan to scale robot production?

  Figure is investing in US-based manufacturing, aiming to produce robots at an increasing rate, with a goal of reaching a million units per year. They are also exploring automation in their own production lines, potentially with robots building robots.

• What is the significance of the partnership with OpenAI?

  OpenAI's involvement, particularly in leading their Series B funding, provided crucial early support. While they collaborated on AI advancements, Figure ultimately found its internal AI team more effective for their specific robotics goals.

• What are the potential economic impacts of widespread humanoid robot adoption?

  Humanoid robots could significantly impact the workforce by automating tasks, potentially leading to job displacement but also creating new opportunities. They could also drive economic growth by increasing productivity and enabling new industries.

• How does Terahertz imaging technology work for detecting concealed weapons?

  Terahertz imaging uses electromagnetic waves in the terahertz frequency range to scan individuals. These waves can penetrate clothing and other materials, allowing the system to detect the shape and density of objects like guns, even when concealed. The technology creates a 2D or 3D image, highlighting potential threats without using ionizing radiation.

• What are the main limitations of current AI chatbots like ChatGPT?

  Current AI chatbots often lack personalization, persistent memory, and the ability to truly understand context or perform complex real-world tasks. They struggle with seamless tool integration, internet usage, and remembering past interactions, making them feel less futuristic and less capable than desired for advanced AI assistants.

• What is the primary goal of the AI lab named Hark?

  Hark aims to design the future of AI, focusing on creating what comes after the iPhone. This involves developing highly multimodal AI models and new hardware interfaces that will enable more advanced, personalized, and integrated AI experiences, potentially replacing current phones and computers.

• What are the key pieces of advice for aspiring founders?

  Founders should prioritize starting to build immediately, embrace continuous learning from their experiences, and focus on tackling the hardest, most impactful problems with genuine passion. They should also be ambitious, as tackling difficult challenges can lead to greater rewards and attract top talent.

• What is the main concern regarding robots in the future?

  The primary concern is not necessarily conscious, self-aware robots turning on humans, but rather obedient machines that can malfunction, be hacked, or be misused as tools for surveillance, intimidation, or even assassination, posing significant security and ethical risks.

Show Notes