An overview of hints and expecations about GPT-4 and what the OpenAI CEO recently said about it.
Some claim that GPT-4 is “next-level” and disruptive, but what will the actual results be?
Questions regarding the GPT-4 and the future of AI are answered by CEO Sam Altman.
GPT-4 Multimodal AI Indicators: Are They There?
Sam Altman, CEO of OpenAI, spoke on the near future of AI technology in an episode of the podcast “AI for the Next Era” dated September 13, 2022.
His statement that a multimodal model was coming soon is particularly intriguing.
The term “multimodal” refers to the capacity to operate in a variety of modes, including text, visuals, and sounds.
OpenAI communicates with people by accepting text input. Whether using ChatGPT or Dall-E, all communication is text-based.
Speech can be used by an AI with multimodal skills to communicate. It can respond to commands by providing information, carrying out tasks, or both.
Altman teased us with the following information about what to anticipate shortly:
“I believe multimodal models will be available in not too long, which will lead to new opportunities.
People are, in my opinion, doing wonderful work with agents that can utilize computers to carry out tasks for you, use programs, and this concept of a language interface where you can say what you want in a conversation back and forth in natural language.
The machine simply performs the iteration and refinement for you.
With DALL-E and CoPilot, you may see some of this pretty early on.
GPT-4 will be multimodal, but Altman didn’t mention it precisely. But he did make a hint that it would happen soon.
The fact that he sees multimodal AI as a platform for creating novel business models that aren’t currently feasible is really intriguing.
He made the comparison between multimodal AI and the mobile platform, which created thousands of new businesses and jobs.
Altman declared:
“…Generally speaking, [I think] that these extremely powerful models will be one of the true new technical platforms, which we haven’t really had since mobile, and that very significant business will get established using this as the interface.
He gave an answer that he claimed to be features that were certain when questioned about the next stage of evolution for AI.
“I believe that real multimodal models will operate.
Therefore, not just text and graphics but whatever modality you have in one model can travel between things smoothly and effortlessly.
Self-Improvement in AI Models?
The goal of AI researchers is to develop an AI that can learn on its own, which is something that isn’t frequently discussed.
This skill goes beyond simply knowing how to convert between languages on the fly.
Emergence describes a person’s ability to act on their own initiative. It occurs when more training data leads to the emergence of new skills.
However, an AI that learns on its own is a completely different kind of AI that is independent of the size of the training data.
An AI that genuinely learns and improves its skills is what Altman described.
Additionally, this type of AI goes beyond the software version paradigm, in which a corporation releases version 3, version 3.5, and so forth.
He imagines an AI system that gets trained, then learns on its own, evolving into a better version of itself.
Altman made no mention of GPT-4 having this ability.
He simply stated that this is what they are aiming towards and that it is, at least, clearly within the realm of possibility.
He described an AI that could learn on its own:
“I believe that models with continuous learning will be available.
Therefore, if you utilize GPT right now, it will remain in the trained state. Additionally, it doesn’t get any better or do any of that when you use it more.
I believe we’ll change it.
I’m therefore quite excited about everything.
Although it’s unclear, it seems as though Altman was discussing Artificial General Intelligence (AGI).
The reporter asked Altman to clarify that all of the concepts he was discussing were real aims and conceivable scenarios, rather than just his wishes for what OpenAI could accomplish.
Interviewer’s question:
People don’t aware that you’re actually making these bold forecasts from a fairly critical point of view, rather than just saying, “We can take that hill,” so I think it would be great to communicate that.
Altman clarified that all of the things he is mentioning are research-based predictions that enable them to confidently identify the next major project by establishing a workable course of action.
He revealed,
“We like to make predictions when we can be on the cutting edge, understand predictably what the scaling laws look like (or have already done the research), where we can say, ‘All right, this new thing is going to work and make predictions out of that way.'”
And that’s how we attempt to manage OpenAI: by taking 10% of the business to just completely go out and investigate when we have high confidence, which has resulted in enormous victories.
Can GPT-4 Help OpenAI Achieve New Heights?
Money and vast amounts of processing power are two factors that are essential to the operation of OpenAI.
Microsoft has already invested $3 billion in OpenAI, and the New York Times reports that it is in discussions to spend a further $10 billion.
GPT-4 is anticipated to be released in the first quarter of 2023, according to The New York Times.
A venture capitalist named Matt McIlwain who is familiar with GPT-4 made a suggestion that GPT-4 might have multimodal capabilities.
As stated in The Times:
According to Mr. McIlwain and four other individuals with knowledge of the project, OpenAI is working on a GPT-4 system that is much more potent and may be published as early as this quarter.
…The new chatbot may be a system similar to ChatGPT that only creates text because it was constructed using Microsoft’s extensive network of computer data centers. Or it might combine text and visuals.
Microsoft workers and some venture capitalists have already used the program.
However, OpenAI has not yet decided whether the new system would be made available with image-related skills.
The Dollars Trail OpenAI
OpenAI has shared information with the venture investment community, but not with the general public.
There are reportedly discussions taking place that could value the business at up to $29 billion.
That is a fantastic accomplishment given that OpenAI does not yet generate a sizable amount of income and that many technology companies have seen their values forced to decline due to the current economic climate.
As stated in The Observer:
The Journal said that investors interested in purchasing $300 million worth of OpenAI shares include venture capital firms Thrive Capital and Founders Fund. The transaction is set up as a tender offer, in which the investors purchase shares from current shareholders, including staff members.
A confirmation of the technology’s future, which is currently GPT-4, can be seen in the high valuation of OpenAI.
Sam Altman Responds to Questions Regarding GPT-4
In a recent interview for the StrictlyVC show, Sam Altman confirmed that OpenAI is developing a video model, which sounds amazing but might also have extremely detrimental effects.
While it was said that the video component was not a part of GPT-4, Altman was certain that OpenAI would not release GPT-4 unless they had assurances that it was secure.
At 4:37 minutes into the interview, the pertinent section begins:
Interviewer’s question:
“Can you comment on whether GPT-4 will be released in the first half of the year or the first quarter?”
In reply, Sam Altman said:
“It’ll be revealed at some point when we are sure that we can do it responsibly and safely.
I believe that, generally speaking, technology will be developed much more slowly than most people would prefer.
People will eventually approve of our strategy, we hope.
But at the time, I understood that everyone wants the latest, greatest thing, and I completely understand how aggravating that is.
Contrary to popular opinion, we will deliberate on it for a very long time.
Twitter is awash in rumors that are challenging to verify. It is said to have 100 trillion parameters, as opposed to the 175 billion parameters in GPT-3, according to an unsubstantiated rumor.
In the StrictlyVC interview program, Sam Altman dispelled that rumor and asserted that OpenAI lacks Artificial General Intelligence (AGI), which has the capacity to learn anything that a human can.
Altman remarked:
On Twitter, I noticed that. It is total s—t.
The GPT rumor mill is a complete farce.
People beg to be let down, and that is exactly what will happen.
We won’t live up to those expectations since, as I believe is sort of expected of us, we don’t actually have an AGI.
Few Facts, Many Rumors
The public knows very little about GPT-4 as a result of OpenAI’s secrecy, and another dependable truth regarding GPT-4 is that OpenAI won’t release a product unless it is certain that it is secure.
Therefore, it is challenging to predict with precision how GPT-4 will appear and what it will be able to do at this time.
