Overview. Reuters is reporting that leaked information from inside OpenAI points to a major effort to incorporate long inference reasoning inside ChatGPT. The effort is code named Project Strawberry and seems to be very similar to the Q* (pronounced Q star) effort which was the rumored basis for Sam Altman’s very short-lived firing back in November. If true, this would mark a major jump in LLM capabilities, far more significant than the incremental improvements in various benchmarks that have been the hallmark of recent LLM releases.

Long interference reasoning. Long inference reasoning in LLMs will significantly enhance business capabilities compared to current LLM technologies. While existing LLMs can process and generate text based on patterns and short-term context, they often struggle with complex, multi-step problems. LLMs with long inference reasoning, however, will provide businesses with a much more powerful tool for tackling intricate challenges. For example, in financial forecasting, current LLMs might analyze recent market trends and company data to make short-term predictions. In contrast, LLMs with long inference reasoning could integrate a broader range of factors – including long-term economic cycles, geopolitical events, and subtle market interconnections – to generate more accurate and far-reaching forecasts. In supply chain management, traditional LLMs might optimize based on current conditions, while advanced models could anticipate and plan for potential disruptions years in advance by considering climate change patterns, evolving global trade relationships, and technological advancements. For product development, current LLMs can suggest improvements based on existing features and immediate market feedback. LLMs with long inference reasoning could go further, synthesizing insights from emerging technologies, shifting demographic trends, and potential regulatory changes to identify entirely new product categories or markets. This leap in capability will enable businesses to make more informed, strategic decisions, potentially leading to better risk management, increased long-term profitability, and the ability to navigate complex global challenges more effectively than is possible with current LLM technology.

Business impact. One of the best examples of a business process that would benefit significantly more from an LLM with long inference reasoning is strategic corporate planning, particularly for large multinational corporations.

Strategic planning without long inference reasoning:

Current LLMs might assist in analyzing recent market trends, compiling industry reports, and generating short-term forecasts based on existing data. They could summarize competitor actions and help draft basic strategic documents. However, their analysis

would be limited to more immediate, obvious factors and struggle with complex interrelationships or long-term projections.

Strategic planning with long inference reasoning:

An LLM with long inference reasoning could revolutionize this process by:

1. Integrating multiple complex factors: It could simultaneously analyze global economic indicators, geopolitical trends, technological advancements, demographic shifts, climate change projections, and company-specific data over extended time horizons.
2. Scenario modeling: The LLM could generate and analyze numerous detailed, long-term scenarios, considering cascading effects and complex interactions between various factors.
3. Identifying hidden opportunities and risks: By processing vast amounts of data and understanding subtle relationships, it could uncover non-obvious strategic opportunities or potential risks that human analysts might miss.
4. Adaptive strategy formulation: It could propose dynamic strategies that adapt to changing conditions over time, rather than static plans.
5. Cross-industry insights: The LLM could draw relevant insights from seemingly unrelated industries or historical events, applying them innovatively to the company’s context.
6. Stakeholder impact analysis: It could model the long-term effects of strategic decisions on various stakeholders, including employees, customers, investors, and communities.
7. Regulatory foresight: The system could anticipate potential regulatory changes across multiple jurisdictions and model their impact on the business.
8. Competitive dynamics: It could simulate complex competitive scenarios, considering potential moves and countermoves of multiple players in the market over extended periods.

This level of comprehensive, nuanced, and forward-looking analysis would be extremely challenging for human teams or current LLMs to achieve. The insights generated could lead to more robust, adaptable, and successful long-term strategies, potentially providing a significant competitive advantage.

The difference in capability is stark: while current LLMs might help streamline the process of gathering and summarizing information for strategic planning, LLMs with long inference reasoning could actively contribute to the creative and analytical aspects of strategy

formulation, potentially uncovering insights and opportunities that would be extremely difficult to identify through traditional means.

Waiting for Gidot. The power and promise of long inference reasoning in LLMs is clear. But when will those capabilities be available? OpenAI isn’t saying. By the way … has Sora dropped in general availability yet? Infinitive believes that the first long inference reasoning capabilities from OpenAI will ship by the end of 2024 as a ChatGPT 4.5 release.