Before OpenAI’s tight, half-hour unveiling of its new GPT-4o AI model, rumors ran rampant about what could be announced, prompting company CEO Sam Altman to manage expectations, saying that it’s “not a search engine” but that they’ll announce “some new stuff we think people will love.” After the presentation, AI enthusiasts and the tech press were both amazed and disappointed.
The release of GPT-4o—not GPT-5, as some people expected—represented a solid but incremental upgrade to GPT-4.5 Turbo. The “omnimodel” works across text, audio, and images, and offers enhanced coding skills, an emotive female voice, and faster generation times. Reactions to the news put GPT-4o and OpenAI on top of the global Twitter trends list for more than a day.
The event drew immediate comparisons to Google’s Gemini Ultra, released in February. While many AI enthusiasts raved about OpenAI’s new model, some AI experts were less than impressed.
Emad Mostaque, former CEO of Stability AI, shared a meme that summarized the reaction to OpenAI across IQ scores. Most of the buzz peaked around debate over whether artificial general intelligence (AGI) is imminent, but the long tail at both ends appears obsessed with “AI waifus”—virtual companions for lonely nerds.
The irony was palpable, as Stability AI’s Stable Diffusion model is a favored tool for creating such companions.
OpenAI was especially proud of the life-like characteristics of the voice they gave to ChatGPT, which even Altman joked was similar to the virtual companion in the sci-fi film ”Her.” OpenAI cofounder Andrej Karpathy made the same connection.
“The killer app of LLMs is Scarlett Johansson. You all thought it was math or something,” he tweeted.
Elon Musk, the mastermind behind Grok, the competing AI chatbot famous for its “fun mode,” was less subtle in sharing his review. “Demo made me cringe,” he wrote, responding to a Twitter user who said GPT-4o would only worsen a “post-truth era.”
Critics honed in on the presentation’s lack of a single groundbreaking feature, which largely demonstrated things that could already be done with current AI tools.
“I’m the same, but 20% faster,” Julien Chaumond, cofounder of Hugging Face, tweeted. Meta’s chief AI researcher Yan LeCun didn’t reveal his feelings about OpenAI’s event, but did post a laughing emoji in response to Chaumond’s post.
Andrew Ng, co-founder of Coursera, praised the efficiency boost: “When an agentic job that used to take 10 minutes now takes 5 minutes just by switching APIs, that’s great progress,” he said.
Others celebrated the potential practical applications of such a powerful model, which could be tested immediately in other forms. Imran Chaudhri, the builder of the Humane AI pin, demonstrated his device using GPT-4o to play a game, highlighting the model’s vision capabilities—although latency still appears to be an issue.
YouTuber Arun Maini meanwhile praised GPT-4o’s smooth handling of translation tasks.
But some enthusiasts, like fiction writer Ewan Morrison, saw signs of a slowdown in AI development.
“It’s evidence that the plateau for LLMs has been reached already,” he tweeted. “Open AI are doing a smoke and mirrors show to try to prevent capital flight from investors.” He said he was mystified why OpenAI didn’t deliver GPT-5, which was rumored to be released this year.
Digging deeper, Bindu Reddi, CEO of Abacus AI, posted results of a benchmark test indicating that GPT-4o performs worse than GPT-4.5 Turbo when prompted with long, complex coding tasks, possibly due to its multimodality.
Tim Simmons from the YouTube channel Theoretically Media summed up the middle ground, saying, “I’m both overwhelmed and underwhelmed.” While GPT-4o has impressive capabilities, it underwhelms in its lack of truly novel features, he said, the leap from GPT-4 being less significant than the previous jump from GPT-3.5.
Despite the mixed reactions, nearly all quarters drew comparisons between GPT-4o and famous assistants like “Her” or “Jarvis” from the Marvel superhero universe. Deconstructing a summary of its language capabilities as more than “real-time voice translation,” Karpathy highlighted the significance of OpenAI releasing a natively multimodal LLM instead of linking among separate models.
Tanishq Mathew Abraham, research director at Stability AI, said the live demo didn’t fully explain what was special about GPT-4o.
“The livestream demo is not the only cool part about GPT-4o,” he wrote. “No one is reading the GPT-4o blog post, which highlights so many other cool features,” like visual text generation, visual consistency, text-to-3D capabilities, and video summarization.
As for the AGI question, we’re not even close, noted NVIDIA senior research manager Jim Fan. But a new, more powerful GPT-5 may be: using the reported codename for the pending project, AI leaker Jimmy Apples suggested that GPT-4o “feels to be Gobi, or in essence Gobi,” and hinted that there would be more announcements from OpenAI in the near future.
Overwhelming or underwhelming, the launch of GPT-4o has stirred the pot and has intensified the community’s attention on today’s Google I/O keynote in the hopes of seeing a strong response from a top competitor.
Edited by Ryan Ozawa.
Source: Decrypt
