Klu raises $1.7M to empower AI Teams  

Who is George Hotz?

by Stephen M. Walker II, Co-Founder / CEO

Who is George Hotz?

George Hotz, also known by his alias geohot, is an American security hacker, entrepreneur, and software engineer born on October 2, 1989, in Glen Rock, New Jersey. He gained notoriety for being the first person to unlock the iPhone, allowing it to be used with other cellular networks, and later for reverse engineering the PlayStation 3, which led to a lawsuit from Sony.

Hotz is also the founder of comma.ai, a company focused on developing vehicle automation using machine learning, which he has been working on since September 2015. Additionally, he has been involved with tinygrad, a deep learning framework. His work and contributions to the field of technology have made him a notable figure in the hacking and software engineering communities.

Despite his technical achievements, Hotz has a playful side to his public persona, as seen in his LinkedIn profile, where he humorously critiques the platform and expresses his interests in a tongue-in-cheek manner. His influence extends beyond his technical work, as evidenced by his significant following and the interest in his career and personal projects.

What is a person of compute?

George Hotz, known for his work in hacking and AI, proposed a new measure of computational power called the "person" unit. This unit is designed to relate computational effort to human intelligence. The "person" is defined as 20 PetaFLOPs (PFLOPs), and the "person-year" is a corresponding temporal unit. The idea is to translate the immense computational capacity of AI models into something more relatable to human intelligence.

Hotz's proposal has sparked discussions in the AI community. For instance, Andrej Karpathy, an AI researcher at OpenAI, questioned the efficiency of such a computation unit and the assumptions made by Hotz.

In a separate discussion, Hotz compared the computational power of the human brain to that of a computer. He estimated that if the brain is drawing 20W, it results in 81 million neuron firing events per second. Assuming each neuron has a fan-out of 1000, that's 2000 FLOPs per neuron, which totals to 162 GigaFLOPs (GFLOPs). This is far from the 20 PFLOPs, which is the classic estimate of brain compute.

Hotz also applied his "person" unit to real-world AI training. For example, he calculated that the AI model LLaMA-65B was trained for 1,022,362 GPU-hours, which equates to 6329 PetaFLOP/s-days. Comparing this to a "person" who has trained for 30 years (at 20 PFLOPs), which is 219,000 PetaFLOP/s-days, there's a significant difference in scale.

What is George Hotz's current project?

George Hotz's current project is Tiny Corporation, which he has been working on since November 2022. The aim of Tiny Corporation is to develop a new framework for machine learning that is more efficient and less complex than existing frameworks like PyTorch and TensorFlow. Hotz's vision is to enable machine learning models to be trained at the edge, rather than in the cloud, to facilitate faster and more responsive AI applications.

The goal of Tiny Corporation, founded by George Hotz, is to "commoditize the petaflop". In other words, the company aims to make high-performance computing more accessible and affordable. This is in line with Hotz's vision of preventing large entities from controlling the majority of the world's compute power.

Tiny Corporation's short-term goal is to get AMD on MLPerf using the tinygrad framework. Tinygrad is a project started by Hotz in October 2020, which has found a niche in the inference space and is expected to become a serious competitor to PyTorch.

In the long run, Tiny Corporation aims to sell computers for more than they cost to make, with the ultimate goal of becoming a chip company. The company's vision is to have compute power available from 50 different companies, all competing to drive the price to zero, and to have an open-source framework to run cutting-edge AI on any one of those 50 chips.

What is the tinygrad framework and how does it relate to tiny corporation's goal?

The tinygrad framework is a minimalistic, open-source deep learning library developed by George Hotz, also known as geohot. It is designed to be simple and easy to understand, with the entire codebase being less than 1000 lines. This simplicity makes it an ideal platform for testing new deep learning hardware, including GPUs, TPUs, and custom ASICs.

Tinygrad breaks down complex networks into four operation types, and it's designed to be the easiest framework to add new accelerators to, with support for both inference and training. It's also been used in openpilot to run the driving model on the Snapdragon 845 GPU, replacing SNPE, and it supports loading ONNX files.

The tinygrad framework is central to the goals of Tiny Corporation. It provides a consistent software layer across Tiny Corporation's hardware ecosystem, enabling the deployment of AI models across the company's hardware products. This aligns with Tiny Corporation's goal of democratizing access to high-performance computing and fostering rapid hardware innovation.

Tinygrad's efficiency and portability are critical in achieving these goals. By keeping the framework simple and lightweight, it allows for easy integration with various hardware backends and promotes the development of performant AI across Tiny Corporation's products.

In the short term, Tiny Corporation aims to get AMD on MLPerf using the tinygrad framework. In the long run, the company aims to use tinygrad to help it become a chip company, selling computers for more than they cost to make. This is part of Tiny Corporation's broader vision of having compute power available from 50 different companies, all competing to drive the price to zero, and having an open-source framework (like tinygrad) to run cutting-edge AI on any one of those 50 chips.

What did geohot do at twitter?

George Hotz, also known as geohot, volunteered for a 12-week "internship" at Twitter in November 2022, with the goal of improving the platform's functionality. His tasks included fixing Twitter's search function and addressing a persistent login popup issue. However, Hotz resigned from his position at Twitter after approximately five and a half weeks.

In his resignation announcement, Hotz stated that he didn't believe he could make a significant impact at Twitter. He also expressed disappointment about his GitHub "withering" during his time at Twitter. Despite his early departure, Hotz stated that he was still rooting for the success of "Twitter 2.0".

It's worth noting that some changes Hotz made, such as removing the login/signup popup when viewing Twitter posts, were reportedly reverted. This could have contributed to his decision to leave the company.

Is geohot a baller?

Yes.

George Hotz, known as geohot, exemplifies the term "baller" in the tech industry through his notable successes and bold actions. His achievements include unlocking the iPhone, reverse engineering the PlayStation 3, founding comma.ai for vehicle automation, and establishing Tiny Corporation to democratize high-performance computing.

His work is not only technically impressive but also disruptive, as seen in his legal battle with Sony and his brief, impactful stint at Twitter. Hotz's influence is underscored by his significant following and his willingness to take risks, such as his short-lived "internship" at Twitter to enhance its functionality.

Hotz's reputation as a "baller" is recognized within the tech community, a testament to his audacious actions and substantial contributions to the industry.

More terms

What is quantum computing?

Quantum computing represents a significant leap from traditional computing by utilizing quantum bits (qubits) instead of classical bits. Unlike binary bits which are either 0 or 1, qubits can exist in multiple states simultaneously (superposition), enabling quantum computers to process vast amounts of information concurrently and solve complex problems rapidly.

Read more

What is action model learning?

Action model learning is a form of inductive reasoning in the field of artificial intelligence (AI), where new knowledge is generated based on an agent's observations. It's a process where a computer system learns how to perform a task by observing another agent performing the same task. This knowledge is usually represented in a logic-based action description language and is used when goals change. After an agent has acted for a while, it can use its accumulated knowledge about actions in the domain to make better decisions.

Read more

It's time to build

Collaborate with your team on reliable Generative AI features.
Want expert guidance? Book a 1:1 onboarding session from your dashboard.

Start for free