Challenges and opportunities for the future of AI
The development of artificial intelligence (AI) is advancing in leaps and bounds. Yet even the most advanced large language models (LLMs) keep running into unexpected limits.
A striking example of this is the Strawberry Test – a seemingly simple but revealing challenge that clearly exposes the weaknesses of current AI systems. Such tests matter not only to developers but to anyone who relies on AI-based solutions and needs to question their reliability and accuracy.
The Strawberry Test: a stumbling block for LLMs
The Strawberry Test confronts LLMs with a simple question: “How many letters ‘r’ are in the word ‘strawberry’?” For humans, the answer is obvious: three. Many LLMs, however, fail and wrongly answer “two.” This seemingly trivial problem reveals a fundamental weakness: while LLMs can generate impressive texts and grasp complex relationships, they often fail at simple, precise tasks. The Strawberry Test shows us that these models still need considerable improvement to sharpen their attention to detail and logical reasoning – an essential prerequisite for becoming reliable tools across a wide range of applications.
Numbers and counting should really be child’s play for AI, yet what sounds simple – counting the number of letters “r” in the word “strawberry” – poses a real challenge for many advanced models.
OpenAI’s strategy: from theory to practice
OpenAI is pursuing a long-term development strategy based on a five-level AI scale. The current focus is on reaching level 2 – the “Reasoner.” At this level, AI models should not only understand texts but also apply logic and solve problems at a human-like level. The further levels range from Strategist (level 3), where the AI makes strategic decisions, to Autonomous Agent (level 4), which acts without human supervision, all the way to Superintelligence (level 5), a hypothetical AI that surpasses all human cognitive abilities.
Project Strawberry and Orion: OpenAI’s next steps
To reach the next level of AI development, OpenAI has launched two major projects: Project Strawberry and the Orion model. Project Strawberry, announced for fall 2024, focuses on improving the capabilities of LLMs, particularly with regard to precise logical reasoning and exact problem solving. The goal is to create AI systems that are better prepared for challenges like the Strawberry Test and can handle similar tasks precisely. OpenAI has chosen a clever marketing strategy to spark interest in these new models: images of strawberries have been popping up more and more on social media, subtly hinting at the upcoming Project Strawberry. This has given the community plenty to talk about.
But the marketing goes even further. OpenAI reposted an image of the Orion constellation with the text “Let’s conquer the cosmos,” clearly alluding to the Orion model. The Orion model is expected to integrate contextual and reasoning-based capabilities more deeply, ushering in a new era of AI development. While Project Strawberry is meant to bring short-term improvements to existing models, the Orion model could be seen as a long-term evolutionary step – with the potential to fundamentally change our understanding of AI and how it can be applied.
Industry insiders also report that the Strawberry model was presented to high-ranking government agencies before its official release in order to demonstrate its capabilities and possibly secure regulatory support. This step shows how serious OpenAI is about introducing models that work more precisely and logically than their predecessors, and underlines how much importance the company attaches to trust in its AI systems.
Why this matters to us as a foresight consultancy and platform developer
For us as a foresight consultancy and platform developer, these advances are of particular importance. The growing integration of AI models into digital platforms and applications requires reliable and precise LLMs. Improvements through Project Strawberry and Orion could make our platforms more powerful and significantly improve specific use cases in futures studies, market analysis or customer service. These developments open up decisive opportunities to create innovative, AI-powered tools that effectively master complex challenges.
Conclusion: a look into the future of AI
The announcement of Project Strawberry and the Orion model marks a turning point in AI development. It shows that the industry has reached a point where it aims not only to solve complex tasks but also to address the fundamental weaknesses of current LLMs. For us, this means the future remains exciting and full of possibilities. The coming advances could soon bring us models that are more precise, more contextually informed and able to overcome hurdles that were previously insurmountable. This is a development that we, as foresight experts and platform developers, will be watching closely.
Sources & further reading on the topic
