AI FORECASTING
Are AI Capabilities Increasing Exponentially? A Competing Hypothesis
Recent research suggests AI capabilities are growing exponentially. However, our analysis of the same data indicates that growth may be following a sigmoid curve, with an inflection point already passed or in the near future. We propose a multiplicative model that decomposes AI capabilities into base and reasoning components, supporting a more nuanced, plateauing growth trajectory.
Key Findings for Enterprise AI
Our detailed analysis provides critical insights into the real growth trajectory of AI capabilities, essential for strategic planning and investment.
2025-06-06
Overall Sigmoid Inflection Point
2024-11-21
Base Model Inflection Point
2026-06-06
Reasoning Model Inflection Point
203.69
MSE for Our Sigmoid Model
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Our Competing Hypothesis
Multiplicative Model
Data Analysis & Fit
Limitations & Future Work
Challenging Exponential Growth
The METR report suggests AI capabilities have grown exponentially since 2019, doubling every seven months. However, our analysis of the same data, when fit to a sigmoid curve, indicates that the inflection point for overall AI capabilities occurred on 2025-06-06, suggesting a plateau in the near future. This challenges the notion of continuous exponential growth and implies that recent rapid improvements may represent a mature phase of a growth curve rather than its beginning.
2025-06-06
Overall Sigmoid Inflection Point (already passed)
This finding suggests that relying on an exponential growth model for long-term strategic planning might lead to overestimations of future AI capabilities, particularly for tasks currently requiring significant human effort. Enterprise leaders should consider models that account for potential plateaus to avoid misallocating resources or setting unrealistic expectations for AI automation timelines.
| Approach | Key Findings |
|---|---|
| METR Exponential Fit |
|
| Our Overall Sigmoid Fit |
|
Decomposing AI Capabilities
We propose a theoretical model that decomposes AI capabilities into two primary components: a 'base model' capability and a 'reasoning' capability. This multiplicative approach posits that overall AI performance is a product of these two evolving technologies. We argue that the apparent exponential growth observed in recent years can be interpreted as staggered improvements across these distinct, yet inter-dependent, technological components.
Enterprise Process Flow
Base Model Development (e.g., Scaling Data/Size)
→
Integration of Reasoning Capabilities (e.g., CoT Finetuning)
→
Overall AI Capability
The introduction of reasoning capabilities into base LLMs has significantly boosted performance. Our model suggests that base capabilities, driven by scaling data and model size, likely plateaued around November 21, 2024. However, reasoning capabilities, a more recent development, are still showing rapid growth but are projected to reach their inflection point by June 6, 2026. This implies that future gains in overall AI capabilities will heavily depend on significant breakthroughs in reasoning or other component technologies.
2024-11-21
Base Model Capabilities Inflection Point (already passed)
2026-06-06
Reasoning Capabilities Inflection Point (projected near future)
Robust Statistical Fit
We fitted our models to the same experimental data shared by the METR study (HCAST, RE-Bench, SWAA benchmarks) using probabilistic modeling. Our methodology minimizes the mean-squared error (MSE) of the predicted 50% model horizon time. We compared sigmoid, exponential, and B-spline link functions for both base and reasoning components. The results demonstrate a compelling alternative to purely exponential projections.
| Model Specification | Mean Squared Error (MSE) |
|---|---|
| Our Sigmoid Link Model (Full Multiplicative) | 203.69 |
| B-Spline Link Model | 511.80 |
| Exponential Link Model | 2874.67 |
| METR Exponential Curve | 339.93 |
| Our Simple Sigmoid Curve (Overall Fit) | 27.37 |
The sigmoid link model, especially our simpler overall sigmoid fit (MSE: 27.37), achieved significantly lower MSE values compared to the METR exponential curve (MSE: 339.93) and other alternatives. This robust statistical fit reinforces the plausibility of a plateauing trajectory for AI capabilities, indicating that sigmoid growth is a better description of the current data than continuous exponential growth. This suggests that the current era of AI is less about unbounded exponential acceleration and more about navigating rapid improvements within a bounded growth curve.
Acknowledged Limitations & Future Work
Our analysis, while presenting a compelling alternative, has inherent limitations. These include reliance on in-sample evaluation due to the limited amount of available data, potential biases from comparing models optimized using different loss functions, and the critical multiplicative assumption of our model, which requires further validation. Furthermore, our decomposition of AI capabilities is currently limited to "base" and "reasoning" components, while other factors like data engineering, pre/post-training algorithms, and network architecture also significantly contribute to overall progress.
For enterprise leaders, understanding these limitations is crucial for interpreting AI forecasts. Future work should focus on developing more rigorous methodologies for forecasting and evaluation. This includes collecting more data to enable out-of-sample validation, standardizing evaluation metrics across different models to ensure fair comparisons, and further decomposing AI progress into more granular components to achieve more accurate long-term predictions. We invite the broader community to engage in this critical research to enhance the reliability of AI capability forecasts.
Calculate Your Potential AI Impact
Estimate the annual savings and reclaimed human hours by implementing advanced AI solutions within your enterprise.
Your AI Implementation Roadmap
Our structured approach ensures a smooth, impactful AI integration tailored to your enterprise's unique needs.
01. Discovery & Strategy
In-depth analysis of current workflows, identification of high-impact AI opportunities, and development of a custom AI strategy aligned with your business objectives.
02. Solution Design & Prototyping
Designing bespoke AI solutions, selecting optimal models and technologies (considering potential plateauing), and developing rapid prototypes for early validation.
03. Development & Integration
Building, training, and fine-tuning AI models, seamlessly integrating them into existing enterprise systems, and ensuring robust performance and scalability.
04. Deployment & Optimization
Phased deployment of AI solutions, continuous monitoring of performance metrics, and iterative optimization to maximize efficiency and ROI.
05. Training & Support
Comprehensive training for your teams to effectively leverage new AI tools, ongoing support, and expert guidance for long-term success and adaptation to evolving AI capabilities.
Ready to Transform Your Enterprise with AI?
Don't rely on outdated forecasts. Partner with Own Your AI to build robust, future-proof AI strategies tailored to the actual trajectory of AI capabilities.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat