OpenAI Teases GPT-V

OpenAI loves drama. Recently, the company teased everyone with GPT-V. The link ‘openai.com/v-test’ was recently discovered by a user on X, which now shows 404 error. 

This is just a few days after the same user earlier discovered a leaked document saying that GPT-4.5 is about to drop. The document also mentioned that Google rushed the release of Gemini because of a possible GPT-4.5 drop. 

However, Altman put all the speculations to rest by replying with a simple ‘nah’ to the query ‘gpt4.5 leak legit or no?’ by a user on X. Interestingly, The GPT-4.5 leak also included unrealistic pricing. Therefore, it cannot be conclusively determined whether Altman dismissed just the pricing or the existence of GPT-4.5 altogether, and instead are likely to release GPT-V or GPT-5.

Dimitris Papailiopoulos posted on X a funny screenshot of ChatGPT Plus responding that it is running on ‘gpt5-feeltheAGI-this-is-what-ilya-saw-turbo-512k’.

It was believed that behind the firing of Sam Altman was the leaked Q* model, which was able to solve maths problems it hadn’t seen before, touted as a step closer to AGI. This was after the company released GPT-4 Turbo at its DevDay conference in November. There were also rumours that OpenAI is silently working on GPT-5, which seems to be actually true. 

Some people predict that it can be a video generator model given all the recent trends such as Pika AI, Stable Video Diffusion, and RunwayML, and also the fact that it is called ‘v-test’.

Forget GPT-4.5, GPT-V is here

It only makes sense for OpenAI to release another version of GPT as soon as possible. The company needs to stay ahead of every other AI company, as everyone has been comparing their models with GPT-3.5 or GPT-4 as soon as they release them. OpenAI has set a benchmark for all the AI models in the industry. 

OpenAI has all the resources to go ahead with this. Altman recently posted on X that OpenAI has found new GPUs, which could possibly be AMD MI300X announced at the AMD Advancing AI event, where it was revealed that OpenAI will use MI300X in its programming system, Triton 3.0. Meanwhile, Microsoft has also announced that it will use the new AMD chip in its cloud computing segment, Azure.

Currently, some of the models claim to outperform GPT-4 on several tasks and benchmarks. Gemini Ultra is set to be released by early next year. If it lives up to its on-paper prowess over GPT-4, it could indeed pose a significant challenge. It is high time OpenAI sets another benchmark with GPT-5, while starting working on better things. 

While GPT-5 is likely to be more sophisticated than its predecessors, Altman has said that it was technically hard to predict exactly what new capabilities and skills the model might have.

“Until we train that model, it’s like a fun guessing game for us,” he said. “We’re trying to get better at it, because I think it’s important from a safety perspective to predict the capabilities. But I can’t tell you exactly what it’s going to do that GPT-4 didn’t.” said Altman.

Brace yourselves, AGI is coming

Interestingly, OpenAI is suddenly going all out on Super Alignment, publishing blogs and papers. A 26-page preparedness framework released on Monday detailed OpenAI’s process for assessing “catastrophic risks” in AI models before making them available to the public. This development follows Anthropic’s publication of a 22-page document outlining its proactive approach to addressing the potential dangers associated with its AI. 

The preparedness team at OpenAI, led by Aleksander Madry was born around the same time as Ilya Sutskever’s ‘superalignment’ team, which served the purpose of reducing the company’s push towards building superintelligent AI systems, as they possess risk to humanity, which OpenAI’s non-profit arm is clearly against.

At least for the time being, the preparedness and superalignment teams are endeavouring to operate independently. Jan Leike, who leads the superalignment team and is an ally of Sutskever, stated that their primary focus is on safeguarding against the emergence of superintelligent models in the “distant future.”

This is after the company had devised a new method in which it plans to supervise superior AI models with smaller models, clearly indicating its work on a model that surpasses GPT-4 and requires supervision. Some people also found the blog ironic coming from OpenAI.

All these GPT-5 leaks, along with the focus on AI safety hints towards a powerful AI model coming out soon. A portion of the announcement also indicates that OpenAI aims to demonstrate a commitment to balancing the requirements of the non-profit governed by the board, which has a fiduciary duty to safeguard humanity, with the demands of a business led by Altman.