OpenAI directs resources from long-term research to focus on improving ChatGPT

OpenAI is shifting its focus to improving its flagship chatbot, scaling back long-term research efforts, a move that has led to the exit of several senior employees. The strategy change comes as the roughly $500 billion company faces mounting competition from rivals such as Google and Anthropic.

To illustrate the seriousness of the situation, ten current and former staff members confirmed that the San Francisco-based firm has pivoted, reallocating resources from experimental projects to improve the core large language models (LLMs) driving its key chatbot.

Recently departed employees include vice president of research Jerry Tworek, model policy researcher Andrea Vallone, and economist Tom Cunningham amid this strategy shift.

These developments at OpenAI signal a significant transformation for a team that introduced ChatGPT through a research preview in 2022, sparking the rise of generative AI. 

OpenAI shifts its focus towards chatbot enhancements amid the AI boom era 

OpenAI is shifting its focus from being a research lab to a key player in Silicon Valley under the leadership of CEO Sam Altman. However, to achieve this success, the tech giant must convince investors that it can generate sufficient revenue to support its $500 billion valuation. 

One individual with knowledge of OpenAI’s research goals anonymously disclosed that, “OpenAI is viewing language models as an engineering challenge now. They are increasing computing power and refining algorithms and data, achieving significant improvements through these efforts.” 

Nonetheless, the individual warned that pursuing original blue-sky research is becoming increasingly challenging. If someone is not part of a core team, the environment becomes a contentious battleground between competing interests.

Mark Chen, OpenAI’s chief research officer, expressed disapproval of this viewpoint. Based on his argument, “long-term foundational research remains essential to OpenAI and still represents most of our computing resources and investment. We have numerous grassroots projects exploring important questions beyond any single product.” 

Apart from this explanation, Chen also argued that integrating this research with practical applications boosts their scientific impact by accelerating feedback and learning processes. “We have never felt more assured about our long-term research plans aimed at creating an automated researcher,” he added.

Meanwhile, as with other tech giants, OpenAI researchers must obtain senior leadership’s approval for technology credits before beginning their initiatives.

Regarding this requirement, several individuals associated with the firm alleged that researchers whose primary focus did not lie within the field of large language models (LLMs) frequently faced denied requests or insufficient support to conduct their research effectively.

For instance, sources close to the matter said teams such as Sora and DALL-E, which focus on video and image generation models, felt undervalued and lacked the resources for their initiatives because they were viewed as less crucial than ChatGPT.

Altman calls for ChatGPT improvements 

Some employees said multiple non-language-model projects were shut down over the past year, while teams were reorganized to concentrate on improving ChatGPT, which is now used by an estimated 800 million people.

These individuals made these remarks after Altman issued a code red alert on the need to improve ChatGPT in December. Meanwhile, it is worth noting that Altman’s alert came after Google introduced its Gemini 3 model, which surpasses OpenAI’s in independent evaluations, and after Anthropic’s Claude model improved its code-generation capabilities.

Following this finding, a previous worker remarked that there is intense competitive pressure in the tech industry, particularly for growing firms aiming to deploy top-tier models every quarter.

Another former senior employee mentioned that, in theory, there is a willingness to explore various research approaches.

If you're reading this, you’re already ahead. Stay there with our newsletter.