AI Roundup December 2022: ChatGPT Goes Supernova, Google Hits Code Red, OpenAI Open-Sources Point-E

TL;DR

December was the month generative AI stopped being a niche and became the whole conversation. ChatGPT, launched on the last day of November, crossed a million users in five days and never slowed down. The shockwave hit Google hard enough to trigger an internal code red. OpenAI also quietly open-sourced Point-E, a fast text-to-3D model, while Lensa's AI avatars put Stable Diffusion in millions of phones and reignited the fight over training data and artist consent.

ChatGPT Becomes the Fastest-Growing App in History

OpenAI shipped ChatGPT on November 30, and December is when the world noticed. It hit one million users in five days, a pace that made it the fastest-growing consumer product anyone could remember. By the end of the month it was a global phenomenon: people were writing code, drafting emails, debugging stack traces, generating boilerplate, and roleplaying their way through it at a scale nobody had budgeted for.

For builders, the important part was not the chat UI. It was the demonstration that a well-tuned instruction-following model wrapped in a dead-simple interface could pull tens of millions of non-technical people into using an LLM directly. That single product reset everyone's expectations for what a usable AI assistant felt like, and it did it for free.

Why it matters to you

If you build with language models, the ChatGPT moment is your tailwind. It normalized the idea that text in, useful text out, is a product feature people will pay attention to. It also made the GPT-3.5 family the default mental model for what an LLM can do, which is the baseline every local and open-weight project would spend the next year trying to chase.

Google Declares an Internal Code Red

The New York Times reported in late December that Google management had declared a code red over ChatGPT, with CEO Sundar Pichai reportedly reorganizing teams across research, trust and safety, and other groups to accelerate AI products. The framing was blunt: a chat interface that answers questions directly is an existential threat to a search and ads business built on ten blue links.

This is the moment the incumbent panic became public. Google had the research, the talent, and arguably better models sitting in the lab, but it had been sitting on them out of caution about accuracy, safety, and reputational risk. ChatGPT made caution look like a liability overnight, and the pressure to ship would define the search wars that followed.

OpenAI Open-Sources Point-E, Fast Text-to-3D

Amid the ChatGPT noise, OpenAI dropped Point-E in mid-December: a system that generates 3D point clouds from text prompts. The headline number was speed. Point-E produces a 3D point cloud in roughly one to two minutes on a single GPU, compared to optimization-based methods like DreamFusion that could burn hours of GPU time per sample.

The point clouds are low resolution and not production assets, so nobody was shipping game models from it on day one. The real story is two-fold. First, OpenAI open-sourced the code and the pretrained weights, which is not the company's usual move. Second, it pointed at a near future where text-to-3D becomes fast and cheap enough to be a normal tool in a content pipeline, the same way text-to-image had just become.

• Speed over fidelity: seconds-to-minutes generation instead of hours of per-prompt optimization.
• Open weights: code and a pretrained model you could actually run and build on.
• Direction of travel: a clear signal that 3D was next after images.

Lensa Avatars Put Stable Diffusion in Everyone's Camera Roll

The other consumer AI explosion of the month was Lensa's Magic Avatars. The app used Stable Diffusion to turn a handful of selfies into stylized portraits, and it went viral across social feeds. For a lot of people it was their first hands-on contact with a diffusion model, even if they had no idea that is what was under the hood.

It also poured fuel on an already burning fight. Artists pointed out that these models were trained on enormous scraped datasets including their work, with no consent and no compensation, and now an app was charging users to generate work in those same styles. The backlash over training data, attribution, and the economics of generative art moved from artist forums into mainstream coverage, setting up the lawsuits and licensing debates that would land the following year.

Key Takeaways

• The interface won: ChatGPT proved that distribution and a simple UI can matter more than a marginally better model, which is the lesson every builder should internalize.
• Incumbents are reactive now: Google's code red showed that a small lab with a sharp product can force a trillion-dollar company to scramble in public.
• Open weights keep showing up: Point-E being open-sourced gave tinkerers something real to run, and that pattern is exactly what fuels local and homelab AI.
• 3D is the next frontier: fast text-to-3D arrived in research form, so start watching the pipeline that text-to-image already walked.
• The data fight is here to stay: Lensa made the consent and training-data argument unavoidable, and that legal and ethical overhang now hangs over everything generative.