Why a Smaller AI Outperforming Bigger Rivals Matters

TechCrunch recently reported that a London startup called Inherent, founded by former Google DeepMind researchers, claims its AI agent named Faraday outperformed much larger models from Anthropic and OpenAI at replicating scientific research. The key twist is that Faraday runs on a model with only 27 billion parameters, while its competitors use far bigger systems. This flips the usual story that bigger AI always wins.

For Australian small and mid-sized businesses, this is significant because it suggests that powerful AI doesn't have to be expensive or require massive computing power. If a small team can build a focused AI that beats the giants at a specific task, the same principle could apply to areas like customer service, inventory management, or compliance checking. The era of "good enough AI on a budget" may be closer than we think.

The Real Breakthrough Is in How the AI Was Trained

Inherent’s approach uses reinforcement learning, which rewards the AI for good outcomes rather than giving it a rulebook. This is like teaching a junior employee by praising smart decisions instead of reading them a manual. The result is an AI that develops "research taste," or the instinct to choose the right experiments to run.

That matters because most business AI today is trained on past data and can only repeat what it has seen. An AI that learns from rewards can adapt to new problems, much like a human team member who gets better with experience. For Australian SMBs, this could mean AI that actually improves over time without constant reprogramming, making it a better long-term investment.

What This Means for Australian SMBs

Australian small businesses often struggle with tight margins and limited IT budgets. The Inherent story shows that cutting-edge AI isn't only for tech giants. If a small London team can beat OpenAI's newest model at a complex task, there's no reason an Australian company couldn't use similar techniques for niche problems like farming yield prediction, retail demand forecasting, or local regulatory compliance.

However, most SMBs don't have an in-house AI lab. The practical takeaway is to watch for AI tools that focus on specific outcomes rather than general-purpose chatbots. A tool that learns from your business results could outperform a massive model that knows everything but understands nothing about your industry.

What You Can Do Now

  • Identify one repetitive task in your business where a "right answer" is clear — like sorting invoices or answering common customer questions — and look for small, specialized AI tools that can be trained on your data.
  • Ask your IT provider about reinforcement learning applications for your industry. Even if you don't build your own AI, understanding the concept helps you evaluate vendor claims.
  • Start small. Run a pilot with a focused AI agent on a single process before rolling out a company-wide system. The Inherent team started with paper replication; you can start with one department.
  • Consider data quality over data quantity. The best AI for your business learns from high-quality examples, not just more data. Review your internal records for consistency before feeding them to any AI tool.

At MS&VG, we help Australian SMBs cut through the hype and find practical digital solutions. Whether you're curious about AI for your business or need a roadmap for digital transformation, our team can guide you through the options that actually make sense for your size and budget.