This year, AI became virtually synonymous with generative AI.
What happened: Launched in November 2022, OpenAI’s ChatGPT ushered in a banner year for AI-driven generation of text, images, and an ever widening range of data types.
Driving the story: Tech giants scrambled to launch their own chatbots and rushed cutting-edge natural language processing research to market at a furious pace. Text-to-image generators (also sparked by OpenAI with DALL·E in early 2021) continued to improve and ultimately began to merge with their text-generator counterparts. As users flocked to try out emerging capabilities, researchers rapidly improved the models’ performance, speed, and flexibility.
- Microsoft integrated OpenAI’s language models into its Bing search engine. Google, sensing a threat to its search business, leveraged its own formidable models into the Bard chatbot. These rapid-fire launches weren’t all smooth sailing — the AI-enhanced Bing exhibited bizarre behavior, while Bard’s debut was beset by hallucinations — but they set a new bar for search functionality and broad access to text generation.
- Pressing its lead, Microsoft added generative Copilot systems to its flagship applications: a code generator and chatbot for GitHub; a chat interface for Windows; and tools to summarize Word documents, craft Excel formulas, and draft emails in Outlook.
- Numerous teams built open source competitors, seeding an ecosystem of options that developers can download and run freely. Meta initially offered LLaMA for free to researchers, but it jumped the fence to make high-performance text generation available far and wide. Hot on its heels came Falcon, Mistral, and many others. Many open source models deliver performance comparable to that of GPT-3.5, although GPT-4 remains the leader.
- In the cloud, Microsoft Azure, Google Cloud, and Amazon AWS battled to deliver generative AI in the cloud. Amazon offered its own TItan models and a sampling of models from third parties, including Stability AI, Anthropic, and AI21. By the end of the year, many alternatives were available from a variety of cloud providers.
- Less than a year after ChatGPT, GPT-4 integrated DALL-E 3, giving it the ability to interpret images and prompt the image generator to produce them. In December, Google introduced Gemini: a family of language-and-vision models that process mixed inputs of text, images, audio, and video.
Gold rush: Generative AI didn’t just thrill customers and businesses; it generated a flood of funding for AI developers. Microsoft invested $13 billion in OpenAI, and Amazon and Google partnered with the nascent startup Anthropic in respective multibillion-dollar investments. Other generative AI startups raised hundreds of millions of dollars.
Where things stand: In the span of a year, we went from one chat model from OpenAI to numerous closed, open, and cloud-hosted options. Image generators have made strides in their ability to interpret prompts and produce realistic output. Video and audio generation are becoming widely available for short clips, and text-to-3D is evolving. 2024 is primed for a generative bonanza, putting developers in a position to build a wider variety of applications than ever before.