How OpenAI's Smaller Rivals Are Developing Their Own AI That ‘Reasons’

To expand the market for artificial intelligence, developers must create large language models that can reason through customers’ problems, not just parrot answers based on what’s in the models’ training data.

While OpenAI and Google are throwing tons of capital at developing AI that solves problems it hasn’t seen before, small startups and individuals say they’ve come up with cheap hacks to achieve some forms of AI reasoning.

Many of these shortcuts involve breaking down a complex problem—such as drafting a blog post about a new product—into simpler steps and asking a model dozens of questions to help it analyze those steps.

For example, after a customer asks an AI app to draft the aforementioned blog post, the app could automatically trigger extra queries that the customer wouldn’t see, such as asking the LLM behind the app to rate how well it did and where it could improve. That’s similar to the Socratic method of teaching a student to think critically about their beliefs or arguments.

The app could then ask the LLM to rewrite the blog post, taking into account the feedback it gave to itself. This process, known as reflection, often leads to a better result, one AI founder told me.