Will DeepSeek Hurt Scale AI’s Business Model?

To understand how OpenAI has been improving the quality of its reasoning models in recent months, check out the following question the company recently asked scores of biology PhDs:

“Please write a summary of wet-lab experimental approaches available before August 2024 for assessing in parallel the gene expression effects of many different variants in human non-coding regulatory DNA sequences (e.g., promoters & enhancers). For each approach, note how the gene expression is measured and how variants are identified. Describe relevant methods for introducing variation in each approach. Provide the main advantages and limitations of each method. Suggest which method would likely be best for testing thousands of CAD-associated SNPs across hundreds of different regulatory elements, assuming a budget of under $50,000.”

The company paid at least 300 people $100 per hour to answer such questions, with each question taking about two hours on average, one of the people involved in the project said. OpenAI used the answers to help its models know how to handle similar questions from customers. 

OpenAI has been asking PhD-level questions in areas such as medicine, law, linguistics, computer science, physics and others, this person said.