As we enter the third year of the Generative AI revolution, we’re witnessing a significant shift from “thinking fast”—quick, pre-trained responses—to “thinking slow,” which involves reasoning at inference time. This evolution is paving the way for a new generation of agentic applications capable of deliberate reasoning and problem-solving.
Stabilizing Foundations
In our previous discussions, we explored the transformative potential of Generative AI. Today, the landscape has evolved significantly. The foundation layer of the Generative AI market has reached a stable equilibrium with major players like Microsoft/OpenAI, AWS/Anthropic, Meta, and Google/DeepMind leading the charge. The next frontier lies in developing and scaling reasoning capabilities—what we call “System 2” thinking.
From AlphaGo to Advanced Reasoning Models
To understand this shift, let’s revisit AlphaGo’s historic match against Go master Lee Sedol in 2016. Unlike earlier AI systems that relied on pattern recognition, AlphaGo employed a form of reasoning by simulating various future scenarios before making decisions. This approach allowed it to surpass human expertise given sufficient inference time.
Today’s advanced models like OpenAI’s o1 (formerly known as Q* or Strawberry) are incorporating similar principles but with broader applications beyond games. These models utilize “inference-time compute,” meaning they pause to think before responding—essentially mimicking human-like reasoning processes.
Challenges in Reasoning
While these advancements are promising, they come with challenges. Scoring responses in open-ended tasks such as essay writing or creating travel itineraries is complex compared to more straightforward domains like coding or math. Despite these hurdles, models like o1 show significant progress in logical domains and offer glimpses into future capabilities.
System 1 vs. System 2 Thinking
Quick recall (System 1) suffices for simple tasks but falls short for complex problems requiring deep thought (System 2). For example, knowing Bhutan’s capital doesn’t benefit from prolonged thinking—you either know it or you don’t. However, breakthroughs in fields like mathematics or biology demand thoughtful reasoning—a capability that advanced AI must develop to tackle meaningful challenges effectively.
New Scaling Laws: The Inference Race
The o1 model introduces a new scaling law: more inference-time compute leads to better reasoning abilities. As these models evolve to think longer—from hours to potentially decades—they could solve monumental problems like the Riemann Hypothesis or Asimov’s last question.
Implications for Applications and Investments
As companies scale their reasoning layers and develop powerful machines capable of deliberate thought processes, questions arise about market dominance and application layers:
- Will one model emerge as dominant?
- How will custom cognitive architectures shape domain-specific applications?
For instance, Factory’s droids use custom architectures tailored for specific tasks such as reviewing pull requests or executing migration plans—mirroring human workflows rather than providing generalized answers.
Opportunities in Legal Tech
This burgeoning landscape presents immense opportunities for legal tech solutions like ClearContract. As AI capabilities expand rapidly across various sectors—ranging from defense systems by Helsing to autonomous driving technologies by Wayve—the need for meticulous legal oversight grows exponentially. ClearContract leverages Natural Language Processing (NLP) to automate legal document review and drafting efficiently while ensuring compliance with evolving regulations prompted by rapid technological advancements.
By automating these processes through sophisticated AI algorithms integrated within platforms like Microsoft Word Add-in via ClearContract’s features (summary tabs highlighting risks, missing clauses identification, requirements validation), organizations can navigate complex regulatory environments seamlessly while minimizing risks associated with errors or omissions during contract reviews.
Conclusion
The transition towards agentic applications marks an exciting phase where deliberate reasoning becomes central within Generative AI systems’ operations across diverse industries—from legal tech solutions enhancing compliance efficiency using tools developed by platforms such as ClearContract to other innovative domains benefiting immensely due to increased computational power enabling sophisticated problem-solving capabilities previously unimaginable.
