AI experts suggest a ‘lighter’ approach is key to achieving AGI

The artificial intelligence industry stands at a crossroads, with the high costs of developing and deploying large language models (LLMs) creating significant barriers to widespread AI innovation and adoption.

Current market dynamics: The AI landscape is dominated by tech giants like OpenAI, Google, and xAI, who are engaged in a costly race to develop artificial general intelligence (AGI).

• Elon Musk’s xAI invested $6 billion in the venture, including $3 billion for 100,000 Nvidia H100 GPUs to train its Grok model
• The massive spending has created an unbalanced ecosystem where only the wealthiest companies can participate in advanced AI development
• High inference costs, which represent the expense of generating responses from AI models, are making it difficult for developers to create affordable applications

Technical barriers: The current state of AI development presents a significant challenge for application developers seeking to create viable AI-powered solutions.

• Developers face a difficult choice between using lower-cost, underperforming models or risking bankruptcy with expensive high-performance options
• Inference costs for top-tier models like OpenAI’s were approximately $10 per query in May 2023, compared to Google’s traditional search cost of $0.01
• By May 2024, OpenAI’s top model costs decreased to about $1 per query, showing promising cost reduction trends

Emerging solutions: A new approach to AI development is taking shape, focusing on creating more efficient and cost-effective models.

• Inference costs are declining by a factor of 10 per year, driven by improved algorithms, technologies, and more affordable chips
• Companies are beginning to prioritize building lightweight models that can achieve comparable results to top LLMs at a fraction of the cost
• This approach mirrors previous technology revolutions, such as the PC and mobile eras, where continuous improvements in performance and cost drove innovation

Innovation in practice: New companies are demonstrating the potential of this lighter approach to AI development.

• Rhymes.ai, a Silicon Valley startup, has trained a model comparable to OpenAI’s capabilities for just $3 million, versus the $100+ million cost of training GPT-4
• Their AI search application, BeaGo, operates at an inference cost of $0.03 per query, representing just 3% of GPT-4’s price
• The company achieved these results through vertical integration and holistic optimization of inference, model, and application development

Future implications: The shift toward more efficient AI development could reshape the industry’s trajectory and democratize access to AI technology, though challenges remain in balancing performance with cost-effectiveness while maintaining the pace of innovation needed to advance toward more sophisticated AI capabilities.