Breaking News - Microsoft Exec Says OpenAI's GPT-4 to Launch Next Week
Breaking News - Microsoft Exec Says OpenAI's GPT-4 to Launch Next Week
Microsoft Germany CTO revealed the news at a company event
Heise Online reported late yesterday that “Andreas Braun, CTO Microsoft Germany and Lead Data & AI STU, mentioned what he said was the imminent release of GPT-4.” Braun said during Microsoft’s “AI in Focus - Digital Kickoff” event:
"We will introduce GPT-4 next week, there we will have multimodal models that will offer completely different possibilities – for example videos.”
There has been so much speculation around OpenAI’s GPT-4 that some people were calling into question the validity of the story. Silke Hahn, Heise Online’s technology editor and the article author, went to Twitter to confirm her sources.
Based on Braun’s comments, Hahn speculates that the planned debut of GPT-4 will be at Microsoft’s “The Future of Work with AI” event scheduled for March 16th. Microsoft CEO Satya Nadella and corporate vice president Jard Spataro are both scheduled to speak during the event.
Was this a Mistake?
The first question to consider is why Microsoft would be making this announcement and not OpenAI. Microsoft reportedly invested an additional $10 billion into OpenAI in January to acquire a 49% stake. We noted in our coverage about the rumored deal that this would give Microsoft effective control of the company.
In addition, this is not exactly out of character for OpenAI. Sam Altman, OpenAI’s CEO, appeared at Microsoft’s introduction of the New Bign powered by GPT-3 technology (or maybe GPT-4) in February. And OpenAI’s new Slack integration with GPT-3 was first announced by Slack parent Salesforce last week. Plus, Microsoft has a big platform and extensive marketing apparatus that can assist with managing the onslaught of interest this announcement will generate.
The timing also makes sense. Altman has been cagey about when GPT-4 would be released, but the rumors and supposed leaks from within OpenAI have pointed to late Q1 2023 for many months.
Another question is why a CTO at Microsoft Germany would make this comment at an obscure regional event where it would only capture local media coverage. This is harder to explain, but it is not exactly a bad PR strategy. In fact, it is creating a rising level of interest as news of the announcement circulates. This will, in turn, generate more interest in Microsoft’s event next week. It could also have been a mistake. Either way, the news is now out.
What Should We Expect?
If we assume that GPT-4 is coming next week, the next question is what should we expect from it. Altman said in an interview with StrictlyVC about GPT-4 rumors that “people are begging to be disappointed, and they will be.”
Much of the speculation is about the model size, and there is no verifiable rumor on that metric. The general consensus has been that it will be larger than GPT-3’s 175 billion parameters, and some have speculated that it could reach 1 trillion. What is more interesting than model size is what it will do.
Braun said GPT-4 will have “multimodal models.” Altman also talked about this in the StrictlyVC video.
“I think we’ll get multimodal models in not that much longer, and that’ll open up new things…I think we will get true multimodal models working. And so not just text and images but every modality you have in one model is able to easily fluidly move between things.”
ChatGPT today functions as text input and text output. The DALL-E AI image generator uses text input and image output. Today, in the OpenAI playground, you can use a spoken (i.e., voice) input and receive text output. D-ID this week introduced a ChatGPT-style experience that enables text or voice input complemented by text and spoken output delivered by a virtual human avatar. So, this is not exactly unprecedented.
However, Altman’s enthusiasm about this point suggests were are about to see something new with GPT-4. A reasonable prediction is that text inputs may return multimodal outputs such as text with images, charts, or video. I was not expecting video capabilities this soon, but Braun mentioned this specifically. Regardless, images and charts seem like a logical progression.
It will be even more interesting f the models can output sounds or music along with text and images. You can imagine Microsoft Word and PowerPoint users would benefit from all of these feature expansions.
One other point to note is that Altman has said that GPT-4 may be launched in phases. That means the features you see next week might not reveal the full extent of GPT-4 capabilities.
He has continually focused his comments on releasing GPT-4 when it is ready, and some of that calculation revolves around when the company deems it safe. That approach could drive a GPT-4 release strategy of incrementally releasing features as additional testing meets the company’s internal standard for performance and safety.
It sounds like GPT-4 is imminent. When you consider that GPT-3.5 and the new ChatGPT Turbo model are already so good that new businesses formed around the technology are growing quickly, GPT-4 could spur another wave of new businesses and new features added to existing products.
We will have to wait for Satya to show us the way next week. Interesting times are afoot.
