AI: OpenAI GPT-5 is Nigh. RTZ #302

...Foundation LLM AI benchmarks to exponentially improve again

The answer to the perennial AI question of 2024 may be near: when will OpenAI, the current leader in the world of hyper-scaled Foundation LLM AIs release its next generation GPT-5? A critical question for all in the AI Tech Wave. Not in an AI-Doomer ‘End is Nigh’ kind of way of course. Although the existential AI Fears vs Acceleration, especially towards Artificial General Intelligence (aka AGI or Superintelligence), may very well make many think the next few iterations of LLM AIs get us closer to ‘Nigh Ends’. But GPT-5 may be ‘Nigh’ enough this year. Let me explain.

As every major LLM AI company from Google to Anthropic to Meta to Mistral and so many others, have been releasing updates and upgrades to their LLM AI with benchmarks that seemingly equal and/or best GPT-4 in some shape or form.

Perhaps this is why OpenAI founder/CEO Sam Altman, still recovering from his ‘almost’ Palace coup last November, seemed to downplay GPT-4 in a podcast with podcaster Lex Fridman. As Inc.com relates:

“The most popular iteration of OpenAI's ChatGPT -- the generative AI chatbot that's taken the world by storm and amassed 100 million daily users -- "kinda sucks," according to Sam Altman, the company's CEO.”

“Altman struck the critical tone on an episode of The Lex Fridman Podcast, released Monday. The conversation covered a wide area related to generative AI and the torrent of hype and gold rush that's followed since the commercial release of ChatGPT in November, 2022.”

“Fridman called GPT-4 "amazing" and "historically impressive," and described the evolution of different iterations of the tech as fostering "a historic, pivotal moment" in the world.”

“In response, Altman cut a pensive figure, stroking his chin, and said "I think it kinda sucks." He explained his thinking with a comparison to how some people might look back on past versions of the iPhone and think that they're useless compared to current models. "I think it is an amazing thing," Altman said, giving his company some credit for its first commercial product, which it released for free. GPT-4, by contrast, is available starting at $20 a month.”

Well, it all might have been a precursor to the story breaking at Business Insider, that OpenAI is planning to release GPT-5 shortly this year. As Mashable summarizes:

“GPT-5, OpenAI's highly-anticipated model that would power ChatGPT, is reportedly only a few months away.”

“According to Business Insider, OpenAI is expected to release the new large language model (LLM) this summer. What's more, some enterprise customers who have access to the GPT-5 demo say it's way better than GPT-4. "It's really good, like materially better," according to a CEO who spoke with the publication. The new model reportedly still needs to be red-teamed, which means being adversarially tested for ethical and safety concerns. Successful red-teaming will ultimately determine when GPT-5 is released.”

As I’ve discussed at length here already this year, the AI Tech Wave world has been very active competing with OpenAI GPT-4 and beyond. Again, as Mashable explains:

“OpenAI's ChatGPT has been largely responsible for kicking off the generative AI frenzy that has Big Tech companies like Google, Microsoft, Meta, and Apple developing consumer-facing tools. Google's Gemini is a competitor that powers its own freestanding chatbot as well as work-related tools for other products like Gmail and Google Docs. Microsoft, a major OpenAI investor, uses GPT-4 for Copilot, its generative AI service that acts as a virtual assistant for Microsoft 365 apps and various Windows 11 features. As of this week, Google is reportedly in talks with Apple over potentially adding Gemini to the iPhone, in addition to Samsung Galaxy and Google Pixel devices which already have Gemini features.”

OpenAI’s Sam Altman is setting up the framework of expectations around GPT-5 in a qualitative way for now:

“Expectations are high for the upcoming GPT-5 model. In a recent podcast interview with Lex Fridman, OpenAI CEO Sam Altman fanned the hype flames by saying, "I think [GPT-4] kind of sucks," before adding "I expect that the delta between 5 and 4 will be the same as between 4 and 3," referring to the vast improvement of GPT-4 from GPT-3. "I think it is our job to live a few years in the future and remember that the tools we have now are going to kind of suck looking backwards at them and that’s how we make sure the future is better," Altman continued.”

“Altman didn't say when GPT-5 would be released, or whether it would even be called "GPT-5", but confirmed, "we will release an amazing new model this year." Altman was vague about what makes it so amazing, but said the technical side is particularly important to defining what the final product will look like.”

I would expect GPT-5 to be exponentially more capable than its predecessor and competitors, both in terms of core metrics, and its multi-modal capabilities. And likely won’t be AGI of course, nor close the AI capabilities many still fear to their core.

More quantitative metrics we may get later this year. But the AI Tech table is being reset shortly.

By OpenAI itself. ‘Nigh’ enough. Stay tuned.

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)