In the 2013 film “Her,” Joaquin Phoenix plays a shy computer nerd who falls in love with an AI he speaks to through a pair of white wireless earbuds. A little over a decade after the film’s release, it’s no longer science fiction. AirPods are old news, and with the imminent full rollout of OpenAI’s GPT-4o, such AI will be a reality (the “o” is for “omni”). In fact, OpenAI head honcho Sam Altman simply tweeted after the announcement: “her.”
GPT-4o can carry on a full conversion with you. In the coming weeks, it will be able to see and interpret the environment around it. Unlike previous iterations of GPT that were flat and emotionless, GPT-4o has personality and even opinions. It pauses and stutters like a person, and it’s even a little flirty. Here’s a video of GPT-4o critiquing a man’s outfit for a job interview:
Interview Prep with GPT-4o
www.youtube.com
In fact, no human involvement is required. Two instances of GPT-4o can carry on an entire conversation without human involvement.
Soon, humans may not be required for many jobs. Here’s a video of GPT-4o handling a simulated customer service call. Currently, nearly 3 million Americans work in customer service, and chances are they’ll need a new job within a couple of years.
Two GPT-4os interacting and singing
www.youtube.com
GPT-4o is an impressive technology that was mere science fiction at the start of the decade, but its also comes with some harrowing implications. First, let’s clear up some confusion about the components of GPT-4o and what’s currently available.
Clearing up confusion about what GPT-4o is
OpenAI announced several things at once, but they’re not all rolling out at the same time.
GPT-4o will eventually be available to all ChatGPT users, but currently, the text-based version is only available for ChatGPT Plus subscribers who pay $20 per month. It can be used on the web or in the iPhone app. Compared to GPT-4, GPT-4o is much faster and just a little smarter. Web searches are much faster and more reliable, and GPT is better about listing its sources than it was with GPT-4.
However, the new text and voice models are not yet available to anyone except developers interacting with the GPT API. If you subscribe to ChatGPT Plus, you can use Voice Mode with the 4o engine, but it will still be using the old voice model without image recognition and the new touches.
Additionally, OpenAI is rolling out a new desktop app for the Mac, which will let you bring up ChatGPT with a keyboard shortcut and feed it screenshots for analysis. It will eventually be free to all, but right now it’s only available to select ChatGPT Plus subscribers.
ChatGPT macOS app… reminds me of Windows Copilot
www.youtube.com
Finally, you may watch these demo videos and wonder why the voice assistant on your phone is still so, so dumb. There are strong rumors indicating that Apple is working on a deal to license the GPT tech from OpenAI for its next-generation Siri, likely as a stopgap while Apple develops its own AI tech.
Is GPT-4o AGI?
The hot topic in the AI world is AGI, short for artificial general intelligence. In short, it’s an AI indistinguishable from interacting with a human being.
I asked GPT-4o for the defining characteristics of an AGI, and it presented the following:
- Generalization: The ability to apply learned knowledge to new and varied situations.
- Adaptability: The capacity to learn from experience and improve over time.
- Understanding and reasoning: The capability to comprehend complex concepts and reason logically.
- Self-awareness: Some definitions of AGI include an element of self-awareness, where the AI understands its own existence and goals.
Is GPT-4o an AGI? AI developer Benjamin De Kraker called it “essentially AGI,” while NVIDIA’s Jim Fan, who was also an early OpenAI intern, was much more reserved.
I decided to go directly to the source and asked GPT-4o if it’s an AGI. It predictably rejected the notion. “I don’t possess general intelligence, self-awareness, or the ability to learn and adapt autonomously beyond my training data. My responses are based on patterns and information from the data I was trained on, rather than any understanding or reasoning ability akin to human intelligence,” GPT-4o said.
But doesn’t that also describe many, if not most, people? How many of us go through life parroting things we heard without applying additional understanding or reasoning? I suspect De Kraker is right: To the average person, the full version of GPT-4o will be AGI. If OpenAI’s demo videos are an accurate example of its actual capabilities, and they likely are, then GPT-4o successfully emulates the first four tenets of AGI: generalization, adaptability, understanding, and reasoning. It can view and understand its surroundings, can give opinions, and it constantly learns new information from crawling the web or user input.
At least, it will be convincing enough for what we in the business world call “decision makers.” It’ll be convincing enough to replace human beings in many customer-facing roles. And for many lonely people, they will undoubtedly form emotional bonds with the flirty AI, which Sam Altman is fully aware of.
Mysterious happenings at OpenAI
We would be remiss not to discuss some mysterious high-level departures from OpenAI following the GPT-4o announcement. Ilya Sutskever, chief scientist and co-founder, quit immediately after, soon followed by Jan Leike, who helped run OpenAI’s “superalignment” group that seeks to ensure that the AI is aligned with human interests. This follows many other resignations from OpenAI in the past few weeks.
Sutskever led an attempted coup against Altman last year, successfully deposing him as CEO for about a week before he was reinstated as CEO. Sutskever can best be described as a “safetyist” who is deeply concerned about the implications of an AGI, so his sudden resignation following the GPT-4o announcement has sparked a flurry of online speculation about whether OpenAI has achieved AGI or if he realized that it’s impossible, because it would be strange to leave the company if it were on the verge of AGI.
From his statement, it seems that Sutskever doesn’t believe OpenAI has achieved AGI and that he’s moving on to greener pastures — ”a project that is very personally meaningful to me.” Given OpenAI’s rapid trajectory with him as chief scientist, he can certainly write his own ticket now.