GPT Engineer Generates An Entire Codebase Based On A Prompt

Mon Jun 19 2023

AI in the Workplace

92% of developers are using AI in the workplace according to a new GitHub survey.

GPT-4 performed best out of the box with 88.4% agreement with ground truth compared to skilled human annotators' 86% accuracy.

Refuel, an AI company that raised $5 million, found that LLMs can label data as accurately as humans but 20 times faster and 7 times cheaper.

The European Parliament passed a draft of the AIA which requires foundational models to create a comprehensive accounting of their training data.

The UK government allocated £100 million for a new UKAI task force led by entrepreneur Ian Hogarth.

The Grammys announced that songs fully generated by AI are not eligible for awards, but those with meaningful human authors can still be submitted.

Nikon released an advertising campaign called Don't Give Up on the Real World.
Nikon's new campaign called Don't Give Up on the Real World overlays mid-journey prompts on photos shot with Nikon cameras in the real world.

GPT Engineer is an AI agent on GitHub that can write an entire codebase with a prompt.
GPT Engineer is a new AI agent that can create entire code bases and has caught the attention of developers on GitHub.
Autonomous AI agents are being developed for specific types of tasks, such as GPT Engineer which generates an entire codebase based on a prompt.
GPT Engineer asks clarifying questions to generate technical specifications, writes all necessary code, and lets you finish a coding project quickly.

The process of interaction between GPT engineer and the user involves supplying a text file with the software requirement, identifying clarifying questions, refining requirements as system prompts, adding an additional set of instructions around what it wants to see as an output, receiving a response from GPT-4 and then creating the source code files for the software project that the user provided instructions for.

Potential improvements include iterative development and changes around the workspace project structure to allow for multi-level file directory organization.

Matthew Berman was impressed by the first output of GPT Engineer which created a game called Snake on its first pass without any edits or bug fixes.

Anton shared that over the last three days he had gone from around 500 to 12,000 github stars.