New AI models are falling short of expectations — here’s why

The rapid advancement of artificial intelligence models appears to be hitting unexpected roadblocks, with major tech companies struggling to achieve significant improvements in their next-generation AI systems.

Current challenges facing OpenAI: OpenAI’s newest language model, Orion, is showing less impressive gains over its predecessor compared to the leap from GPT-3 to GPT-4.

• Internal testing reveals minimal improvements in certain capabilities, particularly in coding tasks
• The underperformance suggests potential limitations in the current approach to AI development
• This setback represents a significant deviation from OpenAI’s historical pattern of achieving substantial improvements with each new model iteration

Industry-wide struggles: The challenges extend beyond OpenAI, affecting other major players in the AI field.

• Google’s next version of Gemini is not meeting internal expectations
• Anthropic has delayed the release of Claude 3.5 Opus, with references to its launch date removed from their website
• These setbacks indicate a broader industry trend rather than isolated incidents

Technical and resource constraints: The current approach to improving AI models through scaling is facing significant hurdles.

• AI models require increasingly massive amounts of computing power and data
• Microsoft is exploring nuclear power plant restoration to meet AI data center energy demands
• Training costs are escalating dramatically, with Anthropic’s CEO projecting individual model development costs could exceed $10 billion by 2027
• The scarcity of high-quality training data has forced companies to rely on synthetic, computer-generated alternatives

Data quality challenges: The industry faces growing difficulties in obtaining suitable training data.

• Companies struggle to acquire unique, high-quality datasets without human intervention
• The readily available supply of web-scraped data is becoming exhausted
• The reliance on synthetic data may not provide the same quality of training as authentic human-generated content

Expert perspectives: Industry leaders and academics are reassessing expectations for AI development.

• Margaret Mitchell of Hugging Face suggests the need for different training approaches to achieve human-like intelligence
• Noah Giansiracusa from Bentley University notes that the recent rapid progress was unsustainable
• These expert observations indicate a growing recognition that current AI development methods may need fundamental revision

Future implications: This technological plateau could reshape the AI industry’s trajectory and economic outlook.
The combination of diminishing returns and escalating costs may force companies to explore alternative approaches to AI development, while potentially cooling investor enthusiasm for the sector. The situation suggests that achieving artificial general intelligence may require more innovative methods beyond simply scaling up existing technologies.