What to know

  • Leading AI companies OpenAI, Google, and Anthropic face significant challenges in improving their latest AI models, with diminishing returns on their investments.
  • OpenAI’s new Orion model has fallen short of expectations, particularly in coding tasks, while Google’s Gemini and Anthropic’s Claude 3.5 Opus face similar hurdles.
  • The main obstacles include a shortage of high-quality training data, massive computing costs, and energy resource limitations.

The tech world’s AI giants have hit a stumbling block in their race to develop more advanced artificial intelligence models. OpenAI, the creator of ChatGPT, has discovered that its newest model, Orion, isn’t living up to the company’s high expectations. The model particularly struggles with coding tasks, failing to show the dramatic improvements that previous versions achieved, as per Bloomberg.

You’ll find similar stories at other AI powerhouses. Google’s upcoming Gemini software isn’t meeting internal benchmarks, and Anthropic has pushed back the release of its Claude 3.5 Opus model. These setbacks have sent ripples through Silicon Valley, challenging the long-held belief that throwing more computing power and data at AI systems would automatically yield better results.

The heart of the problem lies in the scarcity of quality training material. Think of it as trying to teach a student with increasingly limited textbooks – AI companies are running out of high-quality, human-created content to train their models. Some experts predict AI models might exhaust available training material by 2028, forcing companies to explore alternative solutions.

These challenges have prompted a shift in strategy. Rather than solely pursuing bigger models, companies are now exploring different approaches. They’re focusing on post-training refinements, incorporating human feedback, and developing specialized AI tools for specific tasks like flight booking or email management. It’s like switching from a one-size-fits-all approach to a more tailored, task-specific strategy.

The implications of these struggles extend beyond technical challenges. The sky-high valuations of AI companies have been built on the promise of continuous, rapid improvement. As the pace of advancement slows, investors who have poured hundreds of billions into these projects might start asking tough questions about their returns.

Via: MacRumors