The Genesis and Impact of ChatGPT

Key Insight

OpenAI's progress with GPT-3 led to the development of Codex, a private test version in late 2021 designed to help engineers write code, which Microsoft later integrated into Github Copilot. The company also refined its image generator, releasing DALL-E 2 in spring 2022, a diffusion model capable of generating photorealistic images like 'a Corgi riding a skateboard through Times Square.' Concerns about deepfakes and the potential for GPT-3 to spread misinformation due to its convincing prose and tendency to 'hallucinate' or produce toxic responses led to continued safety work.

The core technology of ChatGPT, an aligned version of GPT-3.5, had been available in the API for nearly a year. However, a key figure realized the immense potential of a simple chat interface to the model, choosing to launch it separately from the more powerful GPT-4 to mitigate the simultaneous shock of both advances. Launched as a 'research preview' on November 30, 2022, with a humble tweet, ChatGPT's immediate viral success was unexpected, quickly reaching 100 million users and becoming the fastest-growing consumer tech product in history.

This launch triggered an 'AI arms race' in the tech industry, posing an existential threat to Google Search. Google declared a 'code red,' rushing to integrate generative AI and announcing its own conversational model, Bard, based on LaMDA. Microsoft, having invested a total of 13 billion dollars in OpenAI, aggressively integrated the technology into its Bing search and Edge browser to challenge Google's market dominance, openly stating their intention to make competitors 'dance.' In March 2023, OpenAI released GPT-4, estimated at 1.77 trillion parameters, capable of passing advanced exams like the bar, signifying a new level of AI capability.