Introduction
Look here, ChatGPT is now twice as fast. According to our tests, it's a little more than twice as fast. I once had the poem "The Erlkönig" written down here. And GPT-4o was 2.6 times faster. This speed boost is made possible by the latest version of the GPT-4o language model. Not only does this update make ChatGPT faster, but it also aims to provide more appropriate answers, thinking and helping in a context-related manner.
What's New with GPT-4o?
Speed and Performance
In contrast to the normal GPT-4, all of this is free for all users. GPT-4o is gradually being rolled out, including a new ChatGPT Mac desktop app. If your account is activated for GPT-4o, you can simply select it at the top of the ChatGPT interface across various platforms - browser, smartphone app, and Mac desktop app.
During our tests, GPT-4o was activated in our paid account but not yet in our free one, so we haven't been able to test whether the free version is slower than the Plus subscription. So far, GPT-4o only works with text and image-based input. The new voice mode from GPT-4o will come in the next few weeks.
Multimodal Capabilities
GPT-4o can respond to audio input in just 232 milliseconds, almost as quickly as humans react in conversations. For comparison, the old models needed 2.8 to 5.4 seconds. This speed is crucial for voice assistants, making interactions more natural and efficient.
Enhanced Features of GPT-4o
Image and Audio Processing
GPT-4o can distinguish between calm and excited breaths and express emotions in synthetic language. It can also analyze videos or graphics live, such as by activating the smartphone camera. This feature allows it to recognize and describe emotions in faces in a video or photo and respond accordingly.
Cost Efficiency
According to OpenAI, the new model should be twice as fast and 50 percent cheaper than its predecessor GPT-4 Turbo. This is beneficial for commercial use via the API, reducing costs while improving results.
Multilingual and Logical Thinking
GPT-4o outperforms its predecessor in traditional benchmarks, achieving better scores in multilingual tasks, logical thinking, and programming. It even sets new best values in audio and image comprehension.
Testing GPT-4o
Humor and Language Abilities
We ran a classic joke benchmark, and GPT-4o created a coherent and humorous joke involving Martin Luther, a chocolate cake, and a Raspberry Pi. This demonstrates its improved language abilities and sense of humor.
Practical Applications
In image and text mode, GPT-4o worked well. For instance, it converted a picture into a cartoon and created an HTML file with a map of Hanover's attractions. However, it still faces challenges, such as creating a well-formatted PDF with a vacation plan.
Conclusion
Final Thoughts on GPT-4o
GPT-4o is fast, in most cases twice as fast as GPT-4, though not the fastest current language model. Google's Gemini, for instance, is quicker in some tasks. GPT-4o also performs better in terms of accuracy, with fewer absurd hallucinations. However, there are still errors, like incorrect shortened links, indicating that users can't blindly rely on the AI.
Your Experience with GPT-4o
How is it for you? Do you already have access to GPT-4o and the Mac desktop app? Should we make another blog post about all the multimodal possibilities with GPT-4o once they are fully activated? Feel free to share your thoughts in the comments and of course, subscribe for more updates.
Frequently Asked Questions (FAQ)
Q: What is GPT-4o?
A: GPT-4o is the latest version of OpenAI's language model, offering faster and more efficient performance compared to its predecessors. It supports text, image, and audio processing.
Q: How much faster is GPT-4o compared to GPT-4?
A: GPT-4o is more than twice as fast as GPT-4, according to our tests.
Q: Is GPT-4o free to use?
A: Yes, GPT-4o is free for all users with a limited number of requests per day. Premium customers receive more inquiries.
Q: What platforms support GPT-4o?
A: GPT-4o is available on the browser, smartphone app, and Mac desktop app.
Q: Can GPT-4o handle audio input?
A: Yes, GPT-4o can respond to audio input quickly, making it effective for voice assistants.
Q: What are the enhanced features of GPT-4o?
A: GPT-4o offers improved speed, cost efficiency, and capabilities in image, audio processing, multilingual tasks, logical thinking, and programming.
Q: Does GPT-4o still hallucinate?
A: Yes, like all language models, GPT-4o can still hallucinate, but the level of absurdity has decreased.
Q: How can I get access to GPT-4o?
A: GPT-4o is gradually being rolled out. If your account is activated, you can select it in the ChatGPT interface.
Comments