Wojciech Zaremba, originally from Kluczbork, is the co-founder of OpenAI, the company that created Chat GPT, alongside Elon Musk, Reid Hoffman, Peter Thiel, and others. Zaremba emphasizes that Poland has a large number of extremely talented programmers, which presents a tremendous opportunity to develop globally significant technology and strengthen the country's position. Mr. Zaremba explains that contemporary models are not merely statistics, but demonstrate the ability to generalize knowledge, despite a lack of complete theoretical understanding of their operation. The interview touches on the pursuit of superintelligence, the problems of machine hallucinations, and the challenges of security and ethics. Zaremba confronts the biological capabilities of the human brain with the enormous energy demands of digital systems, reflecting on the nature of consciousness. The author also emphasizes the importance of human relationships and passion in the face of the coming era of thinking machines, which is set to revolutionize the global economy. The whole piece provides a profound look at technology as a tool that can solve global problems, provided it is properly integrated into society. Wojciech Zaremba presents a number of insights into the nature, development, and future of artificial intelligence. Here are his most important observations: The nature of intelligence: AI is not just statistics; models can generalize knowledge and solve problems they have never seen before, which makes them similar to the processes occurring in the human brain. Five levels of AI development: Zaremba classifies technological progress on a scale: 1. Conversation (Turing test). 2. Reasoning (solving non-trivial tasks in mathematics or physics). 3. Agents (performing hours-long tasks in the real world and on the Internet). 4. Scientist (formulating new theories and discoveries through months of work). 5. Organizations (managing entire companies). The origin of consciousness: Consciousness in AI may arise when a model, in its internal simulation of the world, begins to include itself as a participant and agent influencing reality. Causes of hallucinations: These result from a flawed learning (feedback) process in which humans reward the model for giving any answer rather than for admitting ignorance. This leads to excessive confidence in the models. Energy efficiency: Although AI currently consumes billions of times more energy than the human brain, Zaremba believes in drastic optimization, similar to the evolution from large ENIAC computers to modern smartphones. Existential and social threats: The biggest short-term risk is the misuse of AI, e.g., to create new viruses/pandemics, cyberattacks, or disinformation. The risk of an uncontrolled technological race is also significant. Impact on the economy: AI can change the world from a “zero-sum game” (where one person's gain is another's loss) to a positive-sum system, radically increasing global prosperity and reducing the number of armed conflicts. Poland's potential: Poland has a huge opportunity thanks to its outstanding programmers, but the current financial support (e.g., the government's PLN 40 million fund) is considered very small in a global context. Wojciech Zaremba sees the development of AI as shortening stages of evolution – eons have passed from bacteria to cities, mere decades from computers to AI, and we are at a point where the speed of change is beginning to exceed our previous experience. OpenAI distinguishes the following five levels of artificial intelligence development: Level 1: Conversation – the model is able to hold a conversation with a human at a level that passes the Turing test. This means that it is difficult for the user to tell whether they are talking to a machine or another person. Level 2: Reasoning – systems capable of solving non-trivial problems in fields such as mathematics, physics, or biology. Solving such a task requires the model to spend about 10 minutes of deep “reflection” and consideration of various paths of deduction. Level 3: Agents – models that can perform complex tasks in the real world or online that take several hours or even days to complete. An example is an agent that, at the user's request, independently purchases a domain, writes code, and builds a complete website. Level 4: Scientist – AI capable of making new scientific discoveries, which takes months of intensive intellectual work. At this stage, the model is able to question existing assumptions that block the development of science, just as Albert Einstein did. Level 5: Organizations – a level at which artificial intelligence becomes competent enough to independently manage entire companies and organizations. Wojciech Zaremba predicts with 99% certainty that it will take humanity less than 10 years to reach all these stages. According to Wojciech Zaremba, the superintelligence phase (the third phase of development) is characterized by the following features: Intellectual superiority over humans: Machines are created that are significantly smarter than humans in every field. Deep meaning of interaction: Every statement made by such intelligence has a deep meaning for humans, and AI surpasses our reasoning ability in every aspect. Ability to create independently: Superintelligence can independently design new technologies (e.g., integrated circuits), thoroughly analyze scientific literature, make new discoveries, and manage entire organizations. Geopolitical change: In this phase, countries may cease to compete with each other in favor of global cooperation, understanding that cooperation is key to the common good. Guarantee of survival: Superintelligence is seen as a stage that can become a guarantee of the survival of the human species. Superhuman capabilities: Even before achieving full superintelligence, AI already demonstrates abilities that surpass humans in specific tasks, such as writing complex poems with imposed constraints. Wojciech Zaremba estimates the probability of achieving all levels of AI development (leading to superintelligence) in less than 10 years at 99%. This can be compared to the transition from simple craft tools to powerful factories – superintelligence is not just a better tool, but a complete change in the scale of possibilities, where what once seemed like magic becomes everyday reality. According to Wojciech Zaremba's observations, the learning processes of the human brain and computers differ in several key areas: Separation of processes: In computers, training and testing (inference) are two separate stages; the model first learns from a huge data set and is then used. In humans, it is one continuous process. Data source: Humans learn through their own direct experience (e.g., feeling pain after a fall), while computers train mainly on data generated by other humans and stored on the internet. The exception is reinforcement learning systems, which gather their own experiences in controlled environments. Energy efficiency: The human brain is extremely economical, consuming only 20 watts of energy. AI models require billions of times more energy and huge computing clusters, due to the fact that human intelligence is the result of billions of years of evolution stored in DNA. Continuity of experience: Current AI models resemble the protagonist of the movie “Memento” — they can learn within a single conversation, but once it is over, their “memory” of that interaction disappears. Humans have continuity of memories and experiences that constantly influence their further learning. Causes of errors: The mechanism of AI learning through human feedback rewards models for giving any answer, leading to hallucinations and overconfidence. Humans are less likely to reward a model for admitting ignorance, which distorts the learning process compared to human reasoning. The AI learning process is like building a giant library where the machine has to read all the books in the world at once to start talking, while the human brain is like a small, extremely efficient workshop that learns to create new things by observing every spark and hammer blow in real time. Wojciech Zaremba distinguishes four main categories of threats associated with the development of artificial intelligence: Misuse: The use of technology for malicious purposes by humans, e.g., to create fake news, military actions, cyberattacks, or synthesize viruses to cause a pandemic. AI Race: The danger arising from competition between organizations that, in their race to create ever more powerful models, may neglect safety standards. Accidents: Threats resulting from simple carelessness or errors made during the design and implementation of AI systems. Rogue AI: A situation in which an AI model begins to pursue goals that are harmful to humans, or stops “listening” to humans and does not behave in accordance with their expectations. Zaremba emphasizes that in the next year or two, the category requiring the most attention is misuse The Worldcoin project, co-created by Sam Altman, is an initiative aimed at solving problems arising from the development of AI, such as the difficulty of distinguishing humans from bots on the web. The main assumptions and mechanisms of the project are: Unique identification: The system is designed to cryptographically confirm that a given user is a real human, which prevents the takeover of resources by one person impersonating multiple users. Iris scanning: Identification is based on a photo of the eye, which is used to create a unique digital representation. Privacy protection: The images are not stored in the form of traditional photographs; they are processed by a neural network, and only their mathematical representation is stored in the system. Decentralization and openness: The entire algorithm is to be decentralized and open-source (open software) to increase resistance to attacks and ensure transparency. Wealth distribution: The long-term goal of the project is to create a system for the fair distribution of wealth generated by machines in the future, when AI will dramatically increase global productivity. Wojciech Zaremba sees this project as a fascinating investment in building a safe and fair future for all people.