The above image was created by DALL-E by reviewing the following Sequoia Capital’s articles on the Evolution of GenAI:

ARTICLE 1

In their article “Generative AI: A Creative New World,” Sequoia Capital describes the emergence of Generative AI, a powerful new class of large language models that enables machines to generate creative outputs in various fields, such as writing, coding, drawing, and more. The authors highlight the potential of Generative AI to revolutionize numerous industries by reducing the cost and increasing the efficiency of knowledge work and creative processes.

Sequoia Capital emphasizes the transformative power of AI and encourages businesses to embrace this technology to unlock unprecedented labor productivity and economic value. They envision a future where AI-driven systems will not only replace certain functions but also collaborate with humans in a tight, iterative creative cycle, leading to better, faster, and more cost-effective solutions across a wide range of end markets.

Here’s a summary of the four waves of Generative AI as defined in Sequoia Capital’s article:

Wave 1: Small Models Reign Supreme (Pre-2015) – During this phase, small, task-specific models dominated the field of Generative AI, enabling early automation and optimization of various processes. However, these models were limited in their capabilities and required extensive human input.

Wave 2: The Race to Scale (2015-2022) – This wave witnessed a race to scale up AI models, leading to the development of large language models like GPT-3 and DALL-E. These models showcased impressive generative capabilities across a wide range of tasks, thanks to their large-scale training on vast amounts of data.

Wave 3: Better, Faster, Cheaper (2023-2025) – In this wave, Generative AI becomes increasingly accessible, efficient, and effective. Technological advancements allow for faster and cheaper training of AI models, while their outputs continue to improve in quality and utility.

Wave 4: System 2 – Agentic Constructs (2025+) – In Wave 4, Generative AI evolves from reactive, pattern-based “System 1” thinking to a more deliberate, reasoning-based “System 2” approach. This enables AI models to better handle out-of-sample data and engage in more complex problem-solving across diverse fields. Moreover, this wave witnesses the emergence of “agentic constructs” – AI systems that possess agency and can autonomously adapt and make decisions in response to new situations and challenges.

ARTICLE 2

In the article titled “Generative AI’s Act o1: The Reasoning Era Begins,” Sequoia Capital describes the evolution of Generative AI from “System 1” to “System 2” thinking:

According to Sequoia, System 1 refers to pre-trained instinctual responses, such as those demonstrated by models like AlphaGo or Large Language Models (LLMs), which mimic patterns and provide powerful, but limited, capabilities.

In contrast, System 2 thinking involves deeper, deliberate reasoning that allows AI systems to pause, evaluate, and reason through complex, novel situations. Sequoia argues that this shift from System 1 to System 2 thinking is the next frontier in AI development, as it enables models to better handle out-of-sample data and engage in more complex problem-solving across various fields, including math, biology, and business strategy.

Additionally, Sequoia suggests that a new scaling law is emerging: the more inference time compute given to a model, the better it can reason. This breakthrough is leading to the rise of “wrapper” companies, which build on the foundation layer provided by major players like Microsoft/OpenAI, Google/DeepMind, Meta, and Anthropic.

Numorpho  is one of them “wrapper” companies that plans to leverage the power of GenAI by utilizing tools from provisioning companies to orchestrate process engineering and automation to enable the tenets of Industry 4.0 and beyond. Here is our take on the article based on a summary posted by Shish Shridhar:

Thanks for the shoutout to Numorpho Cybernetic Systems (NUMO), ShiSh Shridhar. You rightly state: It’s not enough for models to simply know things—they need to pause, evaluate, and reason through decisions in real time.

“Mulling it over” has not been the characteristic of System 1 generative AI systems wherein #LLM based prompt engines simply spit out aggregation of words (word salad?) which needed to be validated and verified for correctness. We called them #hallucinations when the output was totally off base but I think #confabulation is a better word. Check #halluconfabuphobia in the article linked in references.

In a series of 42 (yes, a homage to the Hitchhiker’s Guide to the Galaxy ) prompt engineered articles appropriately titled “Making Sense out of Nonsense”, we have worked with different System 1 LLM engines to tease out different dimensions for our progression with our process engineering platform, the Mantra M5 by theming out a set of questions for each topic, setting out scenarios and even inviting distinguished #avatars to partake in the conversation. It is the questions that we ask and how we stage them that has helped us in our endeavors (since the answer to life, the universe and everything is always 42!).

More recently, we are working on integrating Industrial Copilots to be agentic drivers for our forays into orchestrating automation in manufacturing and other domains. Here is where we are reviewing System 2 type GenAI systems like GPT-4o1 from OpenAI that Microsoft has been utilizing for their copilot interactions.

We have been following Yann LeCun, the chief scientist at Meta evolve Objective-Driven AI and are utilizing it as the basis for our Outcome Based Adaptive Engineering (OBAE) to enable the creation of smart and connected products and solutions.

NVIDIA’s NIM platform via our Inception partnership with them has also helped us explore, discover and validate different LLM systems like Llama and Mixtral to build out our Large World Models (#LWM) using NVIDIA Omniverse as the basis for defining #digitaltwins and interacting with their physical counterparts to enable actionable intelligence- what we call converting meaning to motion.

Sequoia Capital‘s article that Shish references appropriately summaries this progression from System 1 to System 2: Two years into the Generative AI revolution, research is progressing the field from “thinking fast”—rapid-fire pre-trained responses—to “thinking slow”— reasoning at inference time. This evolution is unlocking a new cohort of agentic applications.

THE ADVENT OF REASONING ENGINES

The maturation of GenAI to System 2 has ushered in the era of LLM-based reasoning engines. With the advent of the -o series from OpenAI, Deep Think from DeepSeek, and updates from other key players like Claude, Meta AI, and Gemini, the landscape of AI systems has evolved significantly. These advancements are transforming the way we process information and make decisions, moving beyond simple pattern recognition and instinctual responses to more complex reasoning and logical thinking.

System 2 thinking called reasoning, characterized by slower, deliberate, and analytical processes, is being integrated into LLMs to create AI models that can better mimic human cognition. This development brings us closer to creating AI systems that can truly understand and interact with the world in a more human-like way.

The recently released Deep Research update to Perpelxity.ai (02/15/2025) does a thorough survey of the ask before reasoning about it (does not use DeepSeek, albeit a version of it is available in the Pro version) and could be the basis for conducting detailed research into a subject matter before delving into the details. We at Numorpho utilized it with fair success – it hallucinated a bit attributing us to tools that we have not (yet) created! Perhaps it already has future vision!!

At Numorpho, this maturation to System 2 will advance our Adaptive Response Engineering (ARE) templates by including varying viewpoints before summarizing the response to the query. This will help us create more comprehensive and nuanced solutions by taking into account diverse perspectives, leading to better decision-making and problem-solving capabilities.

By integrating System 2 thinking into our ARE templates, we can:

1. Enhance Critical Thinking: Incorporating multiple viewpoints encourages a deeper analysis of the query, promoting critical thinking and challenging assumptions.
2. Improve Decision-Making: With a more comprehensive understanding of the problem at hand, our AI systems can provide more informed and well-rounded responses, leading to better decision-making.
3. Foster Innovation: The consideration of diverse perspectives can inspire novel ideas and approaches, fostering innovation and creative problem-solving.
4. Increase User Satisfaction: By providing more accurate, well-informed, and balanced responses, we can enhance user satisfaction and trust in our AI systems.

Integrating System 2 thinking into our ARE templates will elevate the quality and effectiveness of our AI solutions, enabling Numorpho to remain at the forefront of the rapidly evolving AI landscape.

As these AI systems continue to evolve, it becomes increasingly important to maintain human involvement in the decision-making process. By ensuring that humans act as System 2, we can harness the strengths of both humans and AI, combining our unique abilities to make more informed and nuanced decisions.

Additional notes:

1. Perplexity AI’s Deep Research feature performs autonomous research, synthesizing information from multiple sources18.

2. Claude 3.5 Sonnet has a notably large context window of 200,000 tokens3.

3. Meta’s LlaMA 3.3 introduced multimodal capabilities, processing both text and images3.

4. Gemini’s Deep Research feature, like Perplexity’s, offers comprehensive research capabilities6.

5. ChatGPT Plus and Claude.ai offer persistent memory across conversations, while Perplexity AI does not maintain context between queries38.

6. Pi.ai is more focused on conversational AI and has limited features compared to the other models listed.

This comparison provides a general overview of the features, but it’s important to note that these AI models are rapidly evolving, and new features or improvements may be introduced frequently.

OUR TAKE

At Microsoft Ignite 2024 in Chicago, Satya Nadella defined Copilot as the UI for AI. We at Numorpho Cybernetic Systems (NUMO) intend to have the Industrial Copilot to be the driver for our intelligent process engineering platform, the Mantra M5 to facilitate make, manage, move, market and maintain functions for automating system interactions.

• https://www.linkedin.com/pulse/delivering-3-trusted-platforms-ai-age-satya-nadella-uijzc

Utilizing its agentic multi-modal architecture, we intend to:

1. Foster Responsible Innovation – the theming of new products and services with meaningful intent with focus on ethics, sustainability and resilience.
2. Enable Collaboration – our Linked Solutioning model would provide a framework for structured partnership and joint roadmap creation.
3. Enhance Productivity – serve as a productivity multiplier by optimizing and harmonizing processes and enabling proactive operations management.

We commend the work done by Siemens Digital Industries Software, Rockwell Automation, HARTING Technology Group and others in collaborating with Microsoft on defining their basis for the Industrial Copilot and plan to work alongside them as we institute our themed platform for engineering processes.

ShiSh Shridhar, Amanda Marx, Balakrishnen Varadarajan

Stay tuned to an exciting future by following us at https://numorpho.org where we are moving with alacrity to build out our thesis for process engineering using graded responses from our AI counterparts to enable appropriate outcome.

Numorpho has been a Microsoft for Startups partner for 3+ years, and we thank ShiSh Shridhar for his guidance and support as he advises us on our relationship with them and our forays with AI.

NITIN UCHIL Founder, CEO & Technical Evangelist
nitin.uchil@numorpho.com

REFERENCES:

• 20240429 – LLMs can’t compute – EVERYTHING CONNECTED – Numorpho’s Book of Business
• Large Language Model (LLM) Architecture – EVERYTHING CONNECTED – Numorpho’s Book of Business
• 20240303 – Futuring AI – EVERYTHING CONNECTED – Numorpho’s Book of Business
• Agentic AI – EVERYTHING CONNECTED – Numorpho’s Book of Business
• AI Agent Types – EVERYTHING CONNECTED – Numorpho’s Book of Business
• From Sequoia Capital:
  • Generative AI: A Creative New World | Sequoia Capital
  • Generative AI’s Act o1: The Reasoning Era Begins | Sequoia Capital