Research at microGPT

Over the past ten years, the advancement of AI and its impact on society and economy have been largely driven by advancements in neural networks. Advances in deep learning from years of research and development in both industry and academia have led to the development of numerous useful technologies, such as machine translation, protein folding, self-driving cars, and programming assistants.

The combination of large-scale models and simple training methods has also enabled a single model to perform multiple tasks, a departure from the previous practice of creating bespoke models for specific tasks.

microGPT is developing advanced natural language interfaces to the digital world, allowing people to interact with machines as they would with a human colleague. The creation of these human-machine partnerships presents some of the most challenging research areas in AI.

We believe that many of the key research problems on the path to creating intelligent interfaces can only be solved by involving real people in the testing and improvement of our models.

In order to act on the user's behalf in the digital world, our models must possess several skills: the ability to process large amounts of contextual information, understand and manipulate multimodal information, learn from user feedback, and teach themselves to use software tools with minimal human supervision.

At microGPT, you will have the opportunity to work with leading researchers in fields such as modeling, systems, reinforcement learning from human feedback, data curation, and user interfaces. By developing useful products from existing capabilities, our fundamental research will open the door to new possibilities and further empower users. At microGPT, we have the unique opportunity to shape how people interact with the digital world and are seeking individuals with passion and brilliance to join us on this journey.


Introducing medGPT, an innovative application that leverages the power of GPT (Generative Pre-training Transformer) to extract valuable insights from electronic medical records (EMR). This cutting-edge app is designed to analyze and interpret large amounts of complex medical data, providing healthcare professionals with a more comprehensive understanding of their patients' health status and treatment options.

medGPT utilizes GPT to process EMR data and extract meaningful information, such as patient demographics, diagnosis, treatment plans, and lab results. The app's advanced natural language processing capabilities allow it to understand and interpret the data in a way that is easily understandable by healthcare professionals.

One of the key features of medGPT is its ability to identify patterns and trends in the data that may be difficult for a human to spot. This allows healthcare professionals to make more informed decisions about patient care and treatment. For example, the app can analyze a patient's lab results over time and alert a doctor to any abnormal patterns or changes that may indicate a potential health concern.

medGPT also has the ability to learn from user feedback, allowing it to improve its performance over time. As healthcare professionals use the app, they can provide feedback on the results, allowing the app to adjust its algorithms and fine-tune its performance. This makes the app increasingly accurate and efficient as it is used.

In addition, medGPT can assist healthcare professionals in the use of software tools with minimal human supervision. The app can help healthcare professionals navigate and use EMR systems more effectively, reducing the time and effort required to access and interpret patient data. This can lead to more efficient care and improved patient outcomes.

We believe, medGPT is a powerful tool that can help healthcare professionals make more informed decisions and improve patient care. Its ability to extract valuable insights from EMR data using GPT makes it a valuable asset for any healthcare organization.


The development of AI has progressed rapidly in recent years, with the scaling up of Transformers leading to exceptional abilities in language processing (such as GPT-3, PaLM, and Chinchilla), code generation (such as Codex and AlphaCode), and image creation (such as DALL-E and Imagen). microGPT is now pushing the boundaries even further by building models that can take action in the digital world, which is why we are excited to announce our first large-scale model, appGPT.

appGPT is a powerful Transformer model that has been trained to interact with digital tools based on text to action and/or voice to action user interface(UI) design. Recently, it has been taught how to use a web browser. At present, it is connected to a Chrome extension that enables appGPT to monitor the browser and perform actions such as clicking, typing and scrolling. The monitoring process uses a custom-designed "rendering" of the browser viewport that is intended to work across different websites. The model can perform actions based on the user interface elements present on the webpage.

appGPT can understand and complete a task based on a user's request. The user inputs a command into a text box and appGPT will execute it. An example of this is repeatedly taking actions and monitoring progress over a prolonged period to achieve a specific goal.

Sign up to be added to the waitlist for the upcoming alpha release.

txt to action, voice to action.