Post AGI Project: Starting 2025. Zurich, Switzerland

This project endeavors to develop Large Language Model (LLM) agents that emulate human senses by creating specialized AI systems, each dedicated to one of the five senses: sight, hearing, touch, taste, and smell. These agents will undergo a developmental process analogous to human learning, commencing with basic instincts and progressively acquiring knowledge and skills through experience and training.

Leveraging Cerebras Systems' Wafer-Scale Engine (WSE), which integrates over 1.2 trillion transistors on a single chip, provides a substantial computational advantage. In comparison, the human brain comprises approximately 86 billion neurons.

Furthermore, employing analog chips, which process real-world signals and are essential for interfacing between the physical environment and digital systems, offers significant benefits. Analog chips can directly handle continuous signals without the need for analog-to-digital conversion (ADC), thereby eliminating issues such as DC offsets, quantization errors, and component imperfections. This direct processing negates the necessity for auto-zeroing, reduces data loss, consumes less power, and operates at higher speeds compared to digital chips.

By integrating these technologies, the project aims to create AI agents capable of capturing human sensory evolution, achieving a new form of consciousness.

1. Visual Perception (Sight) Agent:

• Instinctual Foundation: Begin with fundamental pattern recognition capabilities, such as detecting light, dark, and basic shapes.

• Developmental Training: Expose the agent to a diverse range of images and visual stimuli, teaching it to recognize objects, understand spatial relationships, and interpret visual scenes.

• Advanced Learning: Introduce complex visual tasks like facial recognition, emotion detection, and scene understanding to enhance contextual awareness.

2. Auditory Perception (Hearing) Agent:

• Instinctual Foundation: Start with the ability to detect basic sounds and differentiate between various frequencies and amplitudes.

• Developmental Training: Expose the agent to a variety of sounds, including human speech, environmental noises, and music, enabling it to recognize patterns, languages, and intonations.

• Advanced Learning: Train the agent in speech recognition, language comprehension, and auditory scene analysis to interpret complex auditory information.

3. Tactile Perception (Touch) Agent:

• Instinctual Foundation: Implement basic sensitivity to pressure, temperature, and texture variations.

• Developmental Training: Simulate interactions with virtual objects of different materials and shapes to teach the agent about texture, hardness, and temperature.

• Advanced Learning: Develop the agent's ability to interpret complex tactile sensations, such as identifying objects by touch or assessing surface qualities.

4. Gustatory Perception (Taste) Agent:

• Instinctual Foundation: Introduce basic taste recognition for sweet, sour, salty, bitter, and umami flavors.

• Developmental Training: Provide data on various food compounds and their taste profiles to help the agent learn to identify and differentiate complex flavors.

• Advanced Learning: Enable the agent to predict flavor combinations, assess taste preferences, and suggest culinary pairings based on learned taste data.

5. Olfactory Perception (Smell) Agent:

• Instinctual Foundation: Equip the agent with the ability to detect and distinguish between basic scent molecules.

• Developmental Training: Expose the agent to a wide range of odor profiles, teaching it to recognize and categorize different smells.

• Advanced Learning: Train the agent to interpret complex scent mixtures, associate smells with specific objects or environments, and predict scent interactions.

Developmental Approach:

• Initial Instincts: Each agent starts with pre-programmed basic instincts relevant to its sensory domain, providing a foundation for further learning.

• Progressive Learning: Agents undergo a structured training regimen, gradually increasing in complexity, similar to human developmental stages from infancy to adulthood.

• Experiential Training: Utilize simulated environments and real-world data to provide diverse experiences, facilitating comprehensive sensory learning.

• Consciousness Simulation: Integrate the sensory agents into a unified system, allowing for cross-modal interactions and the emergence of higher-level cognitive functions, aiming to simulate aspects of consciousness.

This approach mirrors human sensory development, enabling each LLM agent to build upon foundational instincts through experience and learning, ultimately striving to achieve a new form of artificial consciousness.

Learn more »

Join Post AGI Project: ASI Starting 2025

Embark on a groundbreaking journey with the Post AIG Project, set to revolutionize artificial intelligence by developing Large Language Model (LLM) agents that emulate human senses—sight, hearing, touch, taste, and smell. Launching in 2025, this ambitious endeavor seeks to create AI systems that learn and evolve akin to human development, progressing from basic instincts to advanced cognitive functions.

Why Invest in the Post AIG Project?

• Pioneering Technology Integration: By harnessing Cerebras Systems' Wafer-Scale Engine, which boasts over 1.2 trillion transistors on a single chip, alongside cutting-edge analog chips capable of processing real-world signals without the need for analog-to-digital conversion, we are poised to achieve unprecedented computational efficiency and speed.

• Redefining AI Capabilities: Our approach transcends traditional AI limitations by enabling direct interaction with the physical environment, thereby eliminating common issues such as data loss and high power consumption. This positions our AI agents to attain a new form of consciousness, fundamentally transforming human-AI interaction.

• Market Potential: The AI industry is experiencing exponential growth, with projections indicating significant revenue expansion in the coming years. Investing in this project aligns with market trends and offers substantial returns as we pioneer advancements in AI technology.

Join Us in Shaping the Future

We invite visionary investors and passionate innovators to join us in this transformative venture. Your investment will not only fuel technological breakthroughs but also contribute to redefining the boundaries of artificial intelligence.

Together, let's create AI that truly understands and interacts with the world as humans do.

Learn more »

LLM Agents (2022: LangChain) and AIG in 2025

Artificial General Intelligence (AGI) refers to a theoretical form of AI capable of performing any intellectual task that a human can, exhibiting flexibility and adaptability across various domains. Unlike narrow AI, which is designed for specific tasks, AGI would possess the ability to generalize knowledge and apply it to unfamiliar situations.

Retrieval-Augmented Generation (RAG):

RAG is an AI framework that enhances the capabilities of language models by integrating external information retrieval mechanisms. This approach allows AI systems to access up-to-date and specific data beyond their training corpus, improving the accuracy and relevance of generated responses.

Decision Trees in AI:

Decision trees are a method used in AI for decision-making and classification. They operate by making sequential assumptions based on input data, leading to a conclusion or decision. Each node in the tree represents a decision point, with branches indicating possible outcomes.

Combining RAG with Decision Trees:

Integrating RAG with decision tree methodologies can enhance AI decision-making processes:

1. Assumption Formation: The AI uses a decision tree to make an initial assumption or hypothesis based on the input data.

2. Data Retrieval: Leveraging RAG, the AI retrieves relevant external information to support or refute the initial assumption.

3. Comparison and Analysis: The AI compares the retrieved data against different potential outcomes or decisions outlined in the decision tree.

4. Optimal Decision Selection: Based on the analysis, the AI selects the most appropriate decision or generates the best response, informed by both the decision tree logic and the external data accessed via RAG.

This integration allows AI systems to make more informed and contextually appropriate decisions by combining structured decision-making frameworks with dynamic information retrieval capabilities.

Learn more »

LLMs (2017 Google: Transformer architecture, 2018: OpenAI released GPT-1)

Large Language Models (LLMs) are a subset of artificial intelligence (AI) designed to understand and generate human-like text. They operate by processing vast amounts of textual data to learn the statistical relationships between words and phrases.

How LLMs Work:

1. Training: LLMs are trained on extensive datasets comprising text from books, articles, websites, and other sources. This training enables the model to learn grammar, facts about the world, reasoning abilities, and some level of common sense.
2. Architecture: Most LLMs utilize transformer architectures, which allow them to process and generate text efficiently. Transformers use mechanisms like self-attention to weigh the importance of different words in a sentence, enabling the model to understand context and relationships within the text.
3. Inference: Once trained, LLMs can generate coherent and contextually relevant text based on a given input or prompt. They predict the probability of the next word in a sequence, allowing them to produce human-like responses in tasks such as translation, summarization, and question-answering.

Applications of LLMs in AI:

• Natural Language Processing (NLP): LLMs enhance various NLP tasks, including sentiment analysis, entity recognition, and language translation, by providing more accurate and context-aware results.
• Content Generation: They are used to create articles, stories, and even code, assisting in automating content creation while maintaining human-like quality.
• Conversational Agents: LLMs power chatbots and virtual assistants, enabling them to engage in more natural and contextually appropriate conversations with users.

Overall, LLMs represent a significant advancement in AI, enabling machines to process and generate human language with remarkable proficiency, thereby enhancing various applications across different domains.

Learn more »

Neural Networks (Google: CNNs early 2010s)

Neural networks are computational models inspired by the human brain's interconnected neuron structure. They consist of layers of interconnected nodes, or artificial neurons, that process input data to produce an output.

Structure of Neural Networks:

• Input Layer: Receives the initial data.

• Hidden Layers: Intermediate layers that perform computations and feature extraction.

• Output Layer: Produces the final result or prediction.

How They Work:

1. Data Input: Each neuron in the input layer receives a numerical value corresponding to a feature of the input data.

2. Weighted Sum: Each input is multiplied by a weight, and the weighted inputs are summed.

3. Activation Function: The weighted sum passes through an activation function, introducing non-linearity and enabling the network to capture complex patterns.

4. Output Generation: The processed information is transmitted to the next layer, repeating the process until reaching the output layer, which provides the network's prediction.

Learning Process:

• Training: Neural networks learn by adjusting weights based on the error between predicted and actual outputs.

• Backpropagation: A common learning algorithm where errors are propagated backward through the network to update weights, minimizing prediction errors over time.

This iterative learning process enables neural networks to model complex relationships and make accurate predictions across various tasks.

For a more detailed explanation, you can refer to this resource.

Learn more »

Machine Learning (1940s: During World War II, Alan Turing, 1990s: The Cross-Industry Standard Process for Data Mining: CRISP-DM)

Before the widespread adoption of neural networks, predictive analytics primarily relied on statistical methods and traditional machine learning techniques to analyze data and forecast future outcomes. Key approaches included:

1. Regression Analysis

Regression analysis was fundamental in predictive modeling, enabling analysts to understand relationships between dependent and independent variables. Techniques such as linear regression and logistic regression were commonly used to predict continuous outcomes and binary classifications, respectively.

2. Time Series Analysis

Time series analysis involved statistical techniques that utilized historical data to forecast future events. Methods like moving averages, exponential smoothing, and autoregressive integrated moving average (ARIMA) models were employed to identify trends and seasonal patterns in data over time.

3. Decision Trees

Decision trees provided a visual and analytical method for decision-making and classification. By segmenting data into branches based on feature values, they facilitated straightforward interpretation and were instrumental in various predictive tasks.

4. Clustering Techniques

Clustering algorithms, such as k-means clustering, grouped similar data points together based on feature similarities. This unsupervised learning method was valuable for market segmentation, anomaly detection, and exploratory data analysis.

5. Bayesian Inference

Bayesian methods applied probability distributions to model uncertainty in predictions. By updating prior beliefs with new evidence, these techniques offered a probabilistic approach to predictive modeling, enhancing decision-making under uncertainty.

These methodologies laid the groundwork for predictive analytics, enabling organizations to derive insights and make informed decisions based on historical data. The evolution of computational power and data availability eventually paved the way for more complex models, including neural networks and deep learning algorithms, which have further advanced the field.

Learn more »

Learn more:

IA Evolution

• ASI: Artificial Superintelligence (ASI): What It Is and Why It Matters
• Moravec's paradox: What is Moravec's paradox?
• AGI: What is Artificial General Intelligence (AGI)?
• RAG: Reality Augmented Generation (RAG): Bridging the Gap Between Large Language Models and Real-World Data
• LLM Agents: LLM Agents: A Technical Deep Dive
• LLMs: Large Language Models (LLMs): A Technical Deep Dive
• Neural Networks: Neural Networks: A Technical Deep Dive
• Machine Learning: What is Machine Learning? A Technical Perspective

Human Evolution

• Consciousness: Human Evolution and Consciousness: A Complex Interplay
• Instinct vs. Rational Meaning: Instinct vs. Rationality in the Search for Meaning in Human Evolution
• Sense of Life: The Evolution of "Sense of Life" in Humans
• Cogito Ergo Sum: I Think, Therefore I Am: Human Evolution and the Implications of "Cogito Ergo Sum"
• Non-Human Intelligence: Human Evolution and the Search for Diverse Intelligences

Contact »