News Briefs | Portrait
CHATGPT-4O: Intelligent Future
OpenAI’s all-knowing app shows striking improvements in generating content
Ullekh NP
Ullekh NP
17 May, 2024
Ever since the launch in November 2022 of its ChatGPT generative AI app, the Sam Altman-owed, Microsoft-backed OpenAI has become the foremost synonym for mass generation of content, be it text, audio or video, from prompts. The smarter your prompt, the better the result.
GPT stands for Generative Pretrained Transformer and is a Large Language Model (LLM), which, according to the technology think tank Gartner, is a specialised type of AI trained on vast amounts of data to understand existing content and generate new content. This means you give a prompt and LLMs use existing data to produce answers, including essays, documentaries, and speeches. They are also trained to recognise emotions from texts, goes another claim.
OpenAI has since late 2022 updated its LLMs starting from GPT-3, then GPT-3.5, GPT-4, GPT-4 Turbo—and now GPT-4o (where o stands for omni).
When this writer checked out the latest version—ChatGPT-4o—and compared it with ChatGPT-4, the most recent one launched last year, there were striking improvements in content generation. According to an OpenAI blog, as measured on traditional benchmarks, GPT-4o achieves GPT-4 Turbo-level performance on text, reasoning, and coding intelligence, while setting new high watermarks on multilingual, audio, and vision capabilities.
ChatGPT platforms of the San Francisco, California-based AI company that set off the new AI boom, currently attract 100 million weekly users. GPT-4o, according to analysts, is two times faster and 50 per cent cheaper than GPT-4 Turbo. As of now, free users in India can only access ChatGPT-3.5, and subscribers (who pay nearly Rs 2,000 per month) can access ChatGPT-4 and ChatGPT-4o.
OpenAI’s Chief Technology Officer Mira Murati said the ChatGPT-4o model allows ChatGPT to respond more rapidly to voice, image, and video input than OpenAI’s previous technology.
To measure the difference between ChatGPT-4o and the previous version, I gave a prompt: what is the latest situation in Gaza? Clearly, the new version was much faster and brought out more specific numbers compared with the earlier one. The content from the older version was not bad, but the latest one was much better and more formatted—compared to earlier versions, both even offered links for further references, drawing from five or six reputed news websites as sources.
Let’s look at a demo by Corbin Brown, who describes himself as an AI developer and entrepreneur behind the vision of Webcafe AI. He gave the same prompt to ChatGPT-4 and ChatGPT-4o. “What is news about the apple stock I should be aware of [?]”
ChatGPT-4o returned the answer in 10 seconds as opposed to ChatGPT-4, which took 14 seconds. The latter did not give a stock analysis and searched one less site.
In a video, Brown shared a PDF file of a research study on the impact of sugar on the human brain and body to understand the key takeaways. “What is the biggest fact that supports the thesis in this PDF?” was the prompt. Both versions took 17 seconds to come up with a response. Brown concluded that in ChatGPT-4, formatting was bad and the results were not up to the mark.
The new model also supports over 50 languages, including Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Punjabi, Tamil, Telugu, and Urdu. OpenAI has pitched GPT-4o with an ambitious pronouncement: It “is a step towards much more natural human-computer interaction—it accepts as input any combination of text, audio, image, and video and generates any combination of text, audio, and image outputs. It can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds, similar to human response time in a conversation”.
Gary Marcus, an American psychologist and cognitive scientist, has some quibbles somehow. He agrees that on GPT-4o, speech synthesis is terrific, and it reminds him of Google Duplex (which never took off).
But he was expecting more: “If OpenAI had GPT-5, they would have shown it. They don’t have GPT-5 after 14 months of trying.” Meanwhile, Bindu Reddy, CEO, AbacusAI, posted on X: “After a couple of days of switching to gpt4o on some production workloads and getting user feedback, we are seeing mixed results!”
And, as always, some people have lined up quirky errors from prompts.
But thanks to most Fortune 500 companies zealously employing OpenAI tools instead of AI products from any other rival to ensure that their activities, including drafting of reports, are less cumbersome, ChatGPT-4o is hogging the headlines and continuing to be in the spotlight. Mind you, ChatGPT’s influence isn’t restricted to software. It has created millionaires out of shareholders of allied companies such as Nvidia, which has been creating graphics processing units (GPUs) for years but took off exponentially after its use for processing LLMs came to the fore.
More Columns
Arfi Lamba: Homemade Export Kaveree Bamzai
A Train Wreck of Tradition Bhaichand Patel
The Master Survivor of Mysore Sandeep Balakrishna