Research Summary 2024

Intro

This Research was also published on my medium account, please clap :)

Generative AI (Gen AI) burst onto the scene, captivating the tech world and beyond. Today, we’re witnessing its rapid evolution across various domains. 

The GenAI landscape continues to evolve at a remarkable pace, we can see this in weekly GenAI model and tool announcements. Large Language Models (LLMs) are emerging as the new “CPUs” of the digital age, revolutionizing how we interact with technology. Just as CPUs process machine code to execute complex tasks, LLMs process natural language to understand and generate solutions. 

This analogy goes beyond surface similarities; LLMs are becoming the central processing units for human-computer interaction, enabling individuals without complex high-level skills to harness the power of advanced AI computing.

Over the past year, I have explored hundreds of GenAI tools and developed various GenAI applications. In the last few months, I’ve also evaluated a great many startups, connected with industry leaders, VCs, and analyzed reports from various consulting firms to gain a comprehensive understanding of how the industry is about to shape.

This article is an overview of GenAI’s future evolution and impact, it draws on technological trajectories and industry insights from 2024 and examines key trends and potential developments that may shape GenAI in 2025. 

This research also serves as the foundation for the website stateofgenai.com, where I share my GenAI investment thesis, including recommendations on where to invest and what to avoid. I’ll update predictions as new information becomes available, so feel free to check back regularly!

Company Types

I identified several main classes of GenAI companies, such as system integrators, developer assistants, content creators, virtual agents, and GenBI. Their solutions address industry-specific challenges that are focused around leveraging Large Language Models (LLMs), and require various levels of expertise, which influence their business model and product.

There are more classes, but most startups center around these definitions. Others have reached a similar conclusion by asking company employees “What are the most interesting GenAI use cases in their company?” as seen in Figure 1.

Figure 1: Author figure, based on a presentation by stki.info.

In the following section, I’ll define the five classes that I mentioned above.

Virtual agents is a trend that highlights the growing demand for versatile AI-assistants that augment workers and automate routine tasks.

Generative Business Intelligence (GenBI) will augment workforces by using LLMs to transform large datasets into actionable insights. Streamlining data analysis improves operational efficiency and supports data-driven decision-making tailored to customized needs.

System integrators lack competitive moats — they face challenges in developing AI-core products quickly due to a lack of competitive moats. With open-source models widely accessible, companies can easily replicate each other’s innovations.

Developer assistance has become mainstream, with companies enhancing programmer productivity and minimizing debugging time. These tools empower developers to automate end-to-end or repetitive tasks, optimize workflows, and elevate code quality, ultimately speeding up delivery in software development.

Content creators enhance efficiency and quality by leveraging GenAI in their work, and SOTA advancements like Sora, Flux, and GameEnGen are streamlining creative applications.

Market Shaping Forces

McKinsey’s report that GenAI technologies will add between $2.6 trillion to $4.4 trillion of value across business sectors, educe the total amount of work required by all employees by 50%-70% and that the value in GenAI is 75% in Customer operations, marketing and sales, and software engineering.

I have identified several key areas of development that stand out as market-shaping forces that will impact current and emerging markets, workforces, and nations.

Agentic AI 

Agentic AI and AI governance are the two major domains that will create the most impact, globally, in 2025. 

In order to understand why AI agents will play a pivotal role in shaping the future, we need to start with the definition for AI agents. AI Agents need to be able to autonomously make decisions, and plan actions, based on an objective. This means that virtual workforces will be able to offload and augment human work, develop and manage more complicated projects, and therefore, increased productivity across organizations.

Augmentation vs Automation

In their report, the International Labor Organization (ILO) determines that GenAI augmentation, has the potential to affect 10.4 percent of global employment in low-income countries and 13.4 percent of global employment in high-income countries (Figure 2).


Figure 2. Global estimates for jobs with augmentation, ILO.

AI Governance

I see that the ethical considerations surrounding AI are becoming increasingly important to governments, particularly in the areas of privacy and security. This includes the development of enhanced data protection mechanisms, machine unlearning, guard rails, and privacy-preserving AI techniques, alongside robust security frameworks. There is also a growing emphasis on responsible AI development. Additionally, the regulatory framework is evolving, with international cooperation on AI standards becoming more prevalent.


Table 1: AI Act, risk levels.

Industry-specific regulations are being established, and the implementation of ethical AI guidelines is gaining traction to ensure that AI technologies are developed and used responsibly. Notably, the AI Act, was authorized in 2024 and unofficially dubbed “GDPR for AI”. It introduces risks (Table 1) and penalties for non-compliance (Table 2). Reinforcing the importance of adhering to these regulations and ethical standards. Historically, legislation and regulation often begin in the EU, setting a precedent that the rest of the world tends to adopt, with the first markets expected to be the financial and insurance sectors.

Table 2: AI Act Penalties

Influence On Job Markets

On a global scale, I see workforce transformation driven by AI is likely to lead to a major reshaping of job markets. Many traditional roles may be automated, resulting in job displacement [ref], in sectors such as manufacturing, retail, and customer service. However, this shift could also create new job opportunities in emerging fields [ref]. Countries and business markets that adapt swiftly through strategic change management by providing education and workforce retraining will likely see economic growth and enhanced competitiveness, while those that fail to adapt may struggle with higher unemployment rates and social unrest. Businesses that prioritize replacing human workers with AI risk falling behind those that integrate AI as a tool to enhance human capabilities, i.e., AI should complement human skills rather than replace them [ref]. Ultimately, consumers will experience improved services and personalized products due to AI adoption.

Trillion-Size datasets

LLMs are constrained by the knowledge they’ve been trained on; they will be equivalent, but won’t independently expand the boundaries of human understanding. The arrival of trillion-size datasets across various fields, similar to how large-scale internet data fueled the development of LLMs, will catapult advancements in domains like healthcare, drug discovery, and biology (e.g., AlphaFold), unlocking unprecedented insights and capabilities that were previously unimaginable.

Models That Understand

Neural-symbolic models (NSMs) represent an advancement beyond approaches like reasoning-and-actions (ReAct) and chain-of-thought (CoT) reasoning. By integrating logical reasoning with deep learning, NSMs have the potential to tighten knowledge gaps, enabling algorithms to grasp concepts not directly present in the data. This technology could help industries automate complex decision-making and streamline task optimization, driving productivity gains, reducing repetitive work, and fostering a more efficient and equitable job market.

MultiModal Capabilities

I expect foundation models to evolve significantly beyond the current LLMs. There will be an increase in multimodal capabilities, allowing for the integration of text, vision, audio, various signals, and interaction in multimodal domains such as robotics.

Small Language Models

Additionally, I anticipate improvements in efficiency and a reduction in computational requirements, along with enhanced reasoning and problem-solving capabilities. I also anticipate small language models (SLM) to gain more popularity due to their low computational requirements and high performance, again, especially in robotics.

AI SIlicon

In the realm of AI infrastructure, the development of specialized AI hardware is becoming increasingly prominent, fueled by advancements in AI chip manufacturing. This progress will enable distributed AI systems to scale significantly, making it possible for these systems to move away from the cloud and penetrate hardware markets, particularly in robotics, edge computing applications, automotive, and Internet of Things (IoT).

Conclusion

The state of Gen AI a year after its big breakthrough is one of rapid advancement of technology and models. From advanced prompting, fine tuning to hardware innovations, the field is evolving at an unprecedented pace. As we look to the future, it’s clear that Gen AI will continue to transform industries, challenge our understanding of technology, and open new frontiers of possibility.

This shift is transforming how non-technical users approach business and personal challenges. With LLMs, individuals can now describe complex problems, request analyses, or seek creative solutions using everyday language. The models can interpret these requests, drawing upon vast knowledge bases to generate insights, suggest strategies, or even produce code snippets and documentation. This accessibility accelerates problem-solving, improves the quality of solutions by leveraging broad-based knowledge, and significantly lowers the barrier to entry for engaging with advanced computational tools. As a result, businesses can innovate faster, individuals can upskill more easily, and a new era of human-AI collaboration is unfolding, where the power of sophisticated computing is democratized and available for everyone.

The key to harnessing this potential lies in staying informed, embracing innovation, and developing the expertise to not just use AI, but to shape it to our needs. As we navigate this new landscape, one thing is certain: the journey of Gen AI has only just begun.