Carpathian Andrei (Andrej Karpathy)

Carpathian Andrei (Andrej Karpathy)

Biography

Andrei Karpaty was born in 1986 in Slovakia. Since 2017 - Director of Development of Computer Vision and Artificial Intelligence Technologies at Tesla.

The carpathian author of an amazing and promising idea, which can be interpreted as follows: if you embed the trained model inside a computer, you will be able to make it a learner and thus bring it out of the state of a stupid software machine.

In 2017, Andrei Karpaty, formerly a research scientist at Stanford University, took over as head of AI at Tesla, responsible for developing software for computer vision and auto-driving. The rapid career jump of an academic scientist far from the industry, a fan of the high-speed assembly of the Rubik cube, was preceded by the resignation of the previous leadership of this area in full force caused by a number of automobile incidents.

A couple of months later, Carpathi published a rather unexpected post entitled "Software 2.0"^[1]Later in 2018, he commented on it in a half-hour talk^[2] at] the 2018 Train AI conference (San Francisco, May 2018).

In his post, he stated that the current programming paradigm, which after the advent of Software 2.0 has to be called Software 1.0, has remained unchanged for 70 years, despite the emergence of a colossal number of hardware and software new technologies over the years. Its essence is that a strictly deterministic solution is first described by a person in a programming language, then it is compiled (less often interpreted) in machine codes and executed.

The current paradigm arose together with the Von Neumann architecture, which is based on the storage of data and programs in one common memory. It perfectly corresponds to the calculation tasks for which, in fact, it was created, as well as for other works with accurate data. The need for an alternative did not arise until the computers came into contact with the complexities of the world around them. In the vast majority of cases, real-life information is blurred, so greater measurement accuracy and a subsequent solution are not required. But the "sensory revolution" and the big data problem that arose later changed circumstances markedly.

Implementing the Software 2.0 paradigm can be the first practical step towards working with fuzzy data. If we cannot algorithmize the task and write a program due to the fuzziness of the source data, then it is worth providing the computer with some initial knowledge, something like subroutine libraries and functional modules in combination with the mechanism for choosing the one you need from the library.

Then you can enter into a dialogue with the computer, submit data to the input and look at the computer's reaction, adjust the input data and evaluate the results again. As a result of this iterative process, over time, the necessary code will be added to the computer.

According to Carpathian, the Software 2.0 development procedure should be similar to paired programming, where two are working on the code at the same time, sitting at the same workplace. One of them (the performer) writes code, and the other (the observer) monitors the process and focuses on strategy. Periodically, these two change roles. There will also be a couple in the development process of Software 2.0. In it, the performer will be a computer with a machine learning system preinstalled on it, and a person will get the role of an observer.

Carpathian proposes to completely transfer the formation of codes to the machine, and assign control of the human-machine development process to the person. The goal of Software 2.0 is to create a model that can generate codes itself, it learns which codes in accordance with the specified rules should be created to obtain certain results. From the programmer's point of view, when developing within Software 2.0, the writing processes and debugging of the activity program give way to working with data and the iterative process of learning a model based on neural networks.

For this approach, you can use the well-known abbreviation AI (Artificial Intelligence), but decipher it as Augmented Intelligence, understanding by this the ability of the machine to serve as an intellectual assistant to a person. The most important component of this AI approach is the iterative mode of operation with permanent testing, what is called test-driven. A person writes a task and criteria for its assessment, and the machine looks for a solution and presents the result. In this process, the developer is freed from routine activities and gets the opportunity to focus on the essence of the task being solved. That is, he must be a qualified person in the application field.

In the current conditions, using Software 2.0 methods, it will be possible to create codes for background Neumann machines, there is no other choice. But this solution opens up interesting prospects for the created neuromorphic processors, which have elementary abilities for more complex actions than the execution of a given program.

Learn more about Software 2.0 here.

2024: Leaving OpenAI

On February 13, 2024, a Canadian scientist of Slovak origin Andrei Karpaty, one of the co-founders of OpenAI, announced his departure from this company. He intends to focus efforts on his own projects.

Carpathi has extensive experience in fields related to artificial intelligence and earned his PhD from Stanford University in computer vision. He previously completed an internship at Google DeepMind, specializing in deep learning, and before that he studied at the University of British Columbia and the University of Toronto.

𝓲

karpathy.ai

Carpathian leaves OpenAI for the second time. He worked at this startup until 2017, and then decided to move to Tesla, where he took up autopilot technology. And there he reported directly to Elon Musk. In 2022, Carpathi left Tesla, and in 2023 he again joined OpenAI.

Hello everyone, yes, I left OpenAI yesterday [February 13, 2024]. Nothing extraordinary has happened and it is not the result of any particular event, issue or drama (but please keep pushing conspiracy theories as they are very entertaining). My closest plan is to work on personal projects and see what happens, "Karpaty wrote on the social network X.

According to him, while working at OpenAI, he communicated with wonderful people, and the program implemented by the company involves many promising developments. What Karpaty himself plans to work on is not specified as of mid-February 2024.

Andrei leaves to take up personal projects. We are deeply grateful for his contribution and wish him all the best. His duties passed to a senior scientist who interacted closely with him, OpenAI said in a statement.[3]

Notes