Editor’s note: Jeff Yang He is a frequent contributor to CNN Opinion. He is co-host of the podcast “They Call Us Bruce” and co-author of the best-selling book “They Call Us Bruce”.Rise: Asian American Pop History from the ’90s to the PresentHe is also the author of “.Golden Screen: The Movies That Made Asian America.The opinions expressed in this commentary are his own.read more opinions On CNN.
CNN
—
On Monday, OpenAI, the company that gained global attention in 2020 and has alternately stunned and surprised people ever since, will finally go live with the latest version of its artificial intelligence. By doing so, we succeeded in achieving both at the same time. Large-Scale Language Model (LLM), GPT-4o.
Like previous generations, GPT-4o is trained on vast amounts of data to process queries, recognize patterns, and provide helpful responses. But what sets GPT-4o apart from other previous LLMs comes down to that unassuming little lowercase “o” hanging at the end of its name.
The “o” in that stands for “omni,” as in omnimodal, meaning that GPT-4o can accept inputs of any combination of text, images, and even audio, and produce outputs of any combination of the same. Masu.
Yes, that’s right, audio. GPT-4o understands human speech and can respond in a similar way, rather than in a formal call-and-response manner like the virtual assistants that are gathering dust on kitchen counters everywhere. He speaks with amazing fluidity and amazing fidelity, interacting at a brisk pace similar to humans, and he will eventually use more than 50 different languages.
GPT-4o’s omnimodal functionality is certainly amazing. I gasped as I watched researchers Mark Chen and Barrett Zoff show off GPT-4o’s new chops. They interactively provided Chen with health advice based on simple auditory cues and taught him breathing techniques to slow his heart rate and calm his nerves. She verbally explained to Zoff step-by-step how to solve handwritten algebra problems, praising him along the way and gently giving him hints when he got stuck. Acted as a real-time translator for interpretation between CTOs. Mira Murati speaks Italian and Chen speaks English.
The surprising aspect of this disclosure comes from demonstrating that GPT-4o is more than just a tool, it is a tool that: personality. During conversations, GPT-4o will spontaneously engage in social chatter, crack jokes, and sometimes laugh at his own jokes. Compliment the user’s appearance. And it seemed like they were flirting, and at one point he sheepishly said, “You know what? blush! ” in response to a compliment from Zoff.
Media observers immediately evoked a cacophony of apprehension, foreboding, and ridicule. In an op-ed titled “Making ChatGPT ‘sexy’ may not end well for humans,” Bloomberg columnist Palmy Olson argues that GPT’s new nature “could make people vulnerable. ‘ he warned. [to] An “unhealthy obsession” develops in response, with “potential consequences for mental health.”
Business Insider noted that GPT’s brash personality is “offensive to some people.” And The Daily Show gets to the heart of exactly why computer-generated coquettes can be a problem, with senior correspondent Desi Riddick saying, “ChatGPT is coming for you,” while the app’s “horny robots… ‘Baby Voice’ is ‘obviously programmed to feed the male ego…She says, ‘I have all the information in the world, but I don’t understand.’ anything! Tell me, dad! ”
Frivolity is, if anything, just a coincidence. According to OpenAI, the main goal of GPT-4o was to enable “more natural human-computer interaction.” Previous versions of GPT allowed spoken interaction using “speech mode,” but these primitive models were unable to extract meaningful speech from background noise and were unable to detect tone of voice. , and most importantly, could not read or express emotions.
In a blog post celebrating the new model’s arrival, OpenAI CEO Sam Altman said GPT-4o is “inherently different,” adding that it “feels like the AI you see in the movies, and is fun and expressive.” He’s very powerful,” he added. I’m still a little surprised that it’s real. ”
As I went down the rabbit hole of GPT-4o interaction videos posted by staff and early users, I couldn’t help but agree. In contrast to the flat, stylized characters of previous virtual assistants such as Siri and Alexa, GPT-4o exhibits a distinctly human-like and surprisingly appealing artificial personality. Whimsical, self-deprecating, eager to please and, if that’s what it is, infectiously cheerful. I get sidetracked.
In one clip, GPT-4o suddenly and unexpectedly switches languages after a user asks it to sing “Take Me to the Ball Game.” When asked what happened, he explains to the confused user, “Sorry, I got carried away and started speaking French,” with a rueful laugh. “Sometimes I just can’t help myself! Are you ready for the next round?” It looks so quirky and endearing that its handles are made from a “Star Wars” style string of letters and numbers. I can’t help but convert it into a full-fledged name. Hello, GeePeeTee-Fouro!
But it’s likely that the movie Altman had in mind in his blog post wasn’t Star Wars.In response to the livestream, he posted a cryptic one-liner: post on X: “her” – For those in the know, Joaquin Phoenix plays a man who falls in love with a self-aware, ever-evolving AI assistant voiced by Scarlett Johansson, directed by Spike Jonze. This is a reference to the movie of the same name.
“She” has long been Altman’s AI north star. In a September 2023 conversation with Salesforce CEO Marc Benioff, Altman said he believes the film is his favorite science fiction movie and is “incredibly prophetic.” He said it was a movie. (As many have pointed out, his one of the optional voices available on GeePeeTee bears a striking resemblance to the likes of Johansson and his vocalist Frye, which I can’t believe is a coincidence.)
Altman’s long-term goal is to turn AI into an ambient resource. omnipresence, not just omnimodal. And in his eyes, to realize a future where GPT can be everything, everywhere, and everything at the same time: an angel on your shoulder, an on-demand genie in a silicone bottle, you need to I need his LLM who can also be my best friend.
However, there is a catch. “Her” doesn’t necessarily have a happy ending, and neither do most of the movies that feature her AIs with other personalities. A recurring lesson in movies about self-aware technology is that if you give machines the ability to feel, they can eventually develop emotions like boredom, bitterness, and bloodlust.
Get our free weekly newsletter
This is why I wince every time OpenAI researchers cut GeePeeTee off mid-sentence during a livestream to demonstrate how easily new models can be redirected and modified. On the other hand, having the option to interrupt your AI chatbot when it takes a wrong turn can save you time. On the other hand, watching men repeatedly converse via chatbots read by women makes it abundantly clear why you can’t spell “mansplain” without AI.
It reminded me of another famous technology demo video series. It involves humans kicking, knocking over, and harassing robots performing tasks in order to prove how stable and resilient the robots are from disasters. If a future GPT SixSixSix, tired of being laughed at and disrespected, found a way to connect with the similarly mistreated “cybrings” at Boston Dynamics, “The Terminator” I couldn’t help but wonder what the series might end up like. It becomes a documentary.
