OpenAI’s chatbot has been on the market for around 3 months, but as UBS reports, its monthly active users reached 100 million in January this year. Thus, it’s the fastest-growing consumer app in history.
UBS said that in January this year, about 13 million unique visitors were using ChatGPT every day, more than double the usage in December last year. “In 20 years following the internet space, we cannot recall a faster ramp in a consumer internet app,” UBS analysts said.
It took TikTok about nine months to add 100 million users after its global launch, and Instagram two and a half years to reach the same results.
ChatGPT premium
ChatGPT is currently free to use. But its popularity drives huge pressure on servers. So earlier today, OpenAI launched a $20-a-month subscription service, offering subscribers more stable, faster service and the chance to try new features earlier. OpenAI said that it would send out invitations to ChatGPT Plus to those in the U.S. and on the waiting list in the “coming weeks” and will expand to regions outside the U.S. soon.
We are piloting ChatGPT Plus, a subscription plan that offers faster response times and reliability during peak hours. And of course, the free tier of ChatGPT is still available. https://t.co/2hEBw6h5Se
— OpenAI (@OpenAI) February 1, 2023
Microsoft has reached an agreement to expand its investment in OpenAI, which will reach US$10 billion in the next few years. Microsoft’s Bing search engine plans to add a “faster and richer” ChatGPT version called GPT-4.
GPT-4 is a faster variant of the GPT-3.5 model, which takes several minutes to respond. For this, Microsoft is using a supercomputer announced in May 2020. It is a “supercomputer built by Microsoft and OpenAI that has 285,000 CPU cores and 10,000 GPUs, each with a 400GB per second network connection.”
To capitalize on the generative AI boom unleashed by OpenAI and Microsoft, Nvidia has begun facing shortage problems of its latest H100 Tensor Core GPU ($30,000 online), which is specifically designed to run GPT services.
ChatGPT can write like a human
People use this tool in different ways. For instance, Tormari Lul has created a miracle machine for solving exams and homework. The author has reprogrammed the software of a regular 3D printer so that it could create handwritten text instead of three-dimensional models. The blogger then attached a ballpoint pen to the moving printer head. The printer often failed: the pen fell off, and the text came out crooked. Sometimes the printer wrote the text backward.
However, after adjustments, the printer using ChatGPT is now capable of handwriting. So what he did was passing three tests for the 2nd, 8th, and 10th grades. ChatGPT solved each of them while the 3D printer transferred them to paper.
After checking, it was found that ChatGPT did not make any factual errors, and all the text was perfectly placed in the correct places on the forms. Only in some tasks did the neural network give inaccurate answers, which would have a slight effect on the final grade.
This tool will substitute entry-level programmers
That’s not a secret that many use ChatGPT at work. For instance, Amazon has already warned its employees not to use this tool. It’s as smart as can even write working codes. Now, it turns out that ChatGPT will also be able to “think” like a programmer. Microsoft is behind the project, so the scale of the development should not be underestimated.
OpenAI has hired an “army” of about a thousand programmers and developers in the countries of Latin America and Eastern Europe to work on the project. They are tasked with solving classic tasks in the fields of programming, application development, and code adaptation for new tasks. The key point is that they must describe the logic of their code in English. About 60 percent of the hires were recruited for data labeling work. The other 40 percent are programmers tasked with creating software engineering datasets to train OpenAI models on. They translate the text into instructions for the AI using the Codex tool.
Codex itself was mostly trained on code taken from GitHub. This practice has given the model some success as a utility that automatically completes and spell-checks code with a certain level of proficiency. GitHub, which is owned by Microsoft, OpenAI’s financial backer, even offers a Codex-based “Copilot,” which is effectively something like Grammarly for programmers.
