The much-anticipated GPT-4 model could only be released in a safe and responsible manner, according to Sam Altman, CEO of OpenAI.According to Altman, 'In general, we are going to introduce technology much more slowly than customers would like' and 'they will also wait a lot longer,' in an interview with Connie Loizos of TechCrunch.

 Also Read: The place of creative professions in an AI-powered future

Sam Altman, CEO of OpenAI, stated that 'people want to be angry and they will be' in response to rumours regarding GPT-4, the company's most recent language model and the basis for the AI chatbot ChatGPT.

Altman was questioned about whether GPT-4 would release in the first half of the year as many people predict during an interview with StrictlyVC. He replied, but he didn't specify a time. He went on to say that when we are sure that we are acting morally and responsibly, it will eventually come out.

2020 saw the introduction of GPT-3, and ChatGPT was created using GPT 3.5, an updated version. The release of GPT-4 has already received a lot of positive praise from members of Silicon Valley and the AI community. When it comes to making absurd claims about the GPT-4's capabilities, theorising on the model's feature count (a statistic that roughly relates an AI system's complexity and ability, but not linearly) has taken on a life of its own in these circles.

When pressed on it, Altman called a well-known (and untrue) figure that purportedly compares the number of parameters in GPT-3 (175 billion) to GPT-4 'complete rubbish' (100 trillion).

'The rumours about GPT-4 that are circulating are untrue. The OpenAI CEO said, 'Where it all comes from is a mystery to me. If people beg to be disappointed, they will surely receive their desire. The rush is comparable to... Humans have kind of grown accustomed to not having an AGI.

Artificial general intelligence (AGI) in this context refers to an AI system with at least human-equivalent skills across a range of subjects.

Altman covered a wide range of topics during the debate, including when OpenAI will develop an AI model that can produce video. (Meta and Google have already shown research in this area.) It will occur. Altman added, 'I wouldn't want to make a confident prediction about when' in reference to generative video AI. Since it is a trustworthy scientific endeavour, we will try it and others will try it as well. It might be seen in either the present or the future.

In response to concerns about the output's accuracy and the incorrect information it spreads on ChatGPT, Altman made the following statement: (running on GPT-3.5). It frequently completely misinterprets the topic of a human enquiry and provides irrelevant responses due to its 'chatbot' style and lack of empathy.

The business just published a study that shows that, while there isn't a single, all-encompassing solution to the issue, there are societal responses and technological advancements that can help stabilise the situation. It's important to remember that before now, OpenAI has not published any of its GPT models with the same level of care. So, one may question whether this is the true cause.
 

The 'eye-watering' computing cost of maintaining these models is another issue that could significantly impact the organisation. And considerably more will be accomplished by the suggested trillion parameter models. It is therefore unclear who will pay for GPT-4 and whether Microsoft's current infrastructure can actually handle such computing power.