OpenAI doesn’t disappoint with the launch of new multi-modal functionality for GPT 4. Apple may be bringing ChatGPT to the iPhone in a bigger way and the US and China are engaging on talks about AI Safety.
All this and more on this “springtime announcements” edition of Hashtag Trending. I’m your host, Jim Love, let’s get into it.
OpenAI did not disappoint today in their “spring update.” Despite what appeared to be efforts to manage expections the company delivered what many had predicted – a truly multi-modal ChatGPT.
In the past, these GPT announcements appeared inspired by Steve Jobs’ Apple product presentations. There is no secret that Sam Altman was a fan. So like the Apple announcements, Altman would lead the discussion, building up the suspense and then finishing with a live demo.
This new launch was much different. Altman was not even on stage. Instead, a three unknown OpenAI team members handled the announcement and demos with what appeared to be very small live audience.
It could be that this is a new approach or it could simply be that Altman didn’t want to go too much further in “poking the bear,” having already put in his new release just a day before Google’s major conference and product releases.
Or maybe they really did just want to, in the words of Apple’s 1984 ad, “Think Different.” They departed from the normal practice of incrementing numbers on new releases, or of giving the new release a marketing friendly name. This new launch is simply called GPT 4o. Not zero. Four oh.
But 4o is what we had all expected, a truly multi-modal system that can handle voice and vision in addition to text input. Emotion, ability to naturally converse, even with multiple people was all there. And in the typical understated tone of this demo, they barely mentioned OpenAI’s ability to remember across conversations.
The focus, according to the presentation was to make interaction collaborative, natural and easy and the demo did not disappoint.
While the presenters did try to make it seem spontaneous, we have no idea how well scripted and rehearsed this demo was. The phone used was wired in, presumably only to ensure a connection, but it wasn’t as “spontaneous” as they wanted us to think.
In a world where there have been a number of accusations of edited or even faked demos, they did take two questions from a live X/Twitter audience, perhaps to prove that this was a real time demo. While there are only two questions allowed, the model handled them both.
Overall, it was impressive.
Conversations with the chatbot had no latency – that annoying delay while the software crafts its answer. Responses were immediate. But there was more than that that. The voice of GTP 4o seems like a natural conversation with another person. The voice of the software can not only convey emotion, it can also detect it. When the presented pretended to be severely nervous GPT 4o not only detected that by also tried to calm him down.
The speech and listening even allowed the presenters to interrupt the GPT chat, exactly as might happen in a real conversation. The model handled this flawlessly. Those two capabilities, the ability to be interrupted and the emotion detection and response make this new release a quantum leap from any other chat tools and, even if there are bugs to be worked out, this is a credible replacement for a large number of call centre and customer service people.
Vision for text and users was also done well. The AI could recognize emotion on a human face. It could also solve a linear equation by reading what the demonstrator had written on a piece of paper – in real time. Again, it was done step by step, perhaps to ensure that there would be no embarrassment. We know that when complex problems are done step by step AI accuracy improves. But the AI could also explain each step in the calculation and coach and encourage the demonstrator.
This new model is not only more capable and faster, it is also, according to the presenters, much more efficient – so much so that OpenAI says they will be bringing these functions to the free version. They have also dropped the cost of the API services based on this greater efficiency.
You have to wonder if there is a new strategy to either democratize AI for the masses or to try to increase the market share of ChatGPT – perhaps both. On one hand OpenAI is also offering the GPT store to free users. It’s also making the model available in 50 languages.
Another addition will have ChatGPT having not just a phone, but also a desktop app. While this was presented to make the interface more user friendly, it may also be a way to deal with not just latency, but also interacting with the laptop or desktop environment. In the laptop version, the model did a very good job of analyzing code as well as a graphical presentation of data.
With all of this being rolled out over the coming weeks to free users as well as paid users, one could ask what was in it for those who are paid subscribers? Apparently, the sweetener for paid users will not be increased functions, but increased capacity. Paid users will have 5X the capacity of the free users. This is the precise opposite of rival ClaudeAI where careful limits keep nudging users to paid subscriptions.
A couple of other news stories from today.
Apple announced that it is finalizing a deal with OpenAI to integrate ChatGPT features into the upcoming iOS 18 release. This would allow Apple to offer the wildly popular language model’s conversational AI and generative capabilities as a core part of the iPhone experience.
Sources say Apple and OpenAI have been hammering out the terms for using ChatGPT in iOS 18, part of a broader push by Apple to aggressively roll out artificial intelligence across its products and services.
While an OpenAI agreement seems imminent, Apple has also been in talks with Google about potentially licensing the search giant’s own Gemini chatbot AI. However, those discussions have not yet resulted in a deal being reached.
The moves signal Apple is doubling down on AI being a key selling point and marquee feature for future iPhones and iOS updates. Just last year, CEO Tim Cook acknowledged using ChatGPT himself while also noting “a number of issues that need to be sorted” before bringing it to Apple’s platforms.
Integrating generative AI like ChatGPT could supercharge Apple’s existing virtual assistant Siri with advanced language abilities for everything from writing and analysis to creative tasks.
Cook has argued Apple’s unified hardware and software ecosystem gives it inherent advantages in delivering powerful AI experiences. And the company reportedly plans to leverage its own in-house processors and data centers to run certain upcoming AI models and features.
An announcement laying out Apple’s AI roadmap is widely expected for next month’s annual Worldwide Developers Conference. While details are still sparse, the company seems determined to make a “thoughtful” entry into the generative AI arena reshaping consumer tech.
Given how well OpenAI performs on the iPhone, you have to wonder what Apple can come up with that would be more impressive.
Sources include: Fortune
Here’s how I would present this story about U.S.-China talks on AI safety for a tech news podcast:
In a first-of-its-kind diplomatic engagement, the United States and China are kicking off talks this week aimed at establishing some common ground on the safety and risk standards for artificial intelligence.
Senior U.S. officials say a delegation led by the White House and State Department will meet with their Chinese counterparts in Geneva, Switzerland on Tuesday. The agenda? Exchanging views on the technical risks posed by advanced AI systems and discussing domestic efforts each country is taking to address those hazards.
Now this initial meeting is not expected to produce any formal agreements or joint statements. Rather, it marks an exploratory discussion between the two AI superpowers who have yet to see eye-to-eye on what safe and responsible development of the transformative technology looks like.
The talks were prompted by the meeting between President Joe Biden and China’s President Xi Jinping back in November, where they agreed diplomatic conversations on AI risk were necessary given the geopolitical stakes.
While neither side expects substantive cooperation or sharing of sensitive research, the U.S. is framing this as an important first step toward making the world “safer” by opening lines of communication on a critical emerging risk.
Of course, the two rivals have vastly different perspectives on AI ethics and governance that could create friction. Issues like China’s use of AI surveillance and the U.S.’ barriers around sharing cutting-edge models are likely sticking points.
But with both governments aggressively pursuing AI breakthroughs for economic and military superiority, having some shared safety foundations may be prudent.
The standards that ultimately emerge from U.S.-China talks would likely carry outsized influence, shaping norms that other nations look to for developing their own policies around AI risk.
So while not a venue for solving all their AI disagreements, this Geneva meeting represents a tentative first step toward ensuring the technology’s rapid progress globally doesn’t outpace responsible guideposts for its deployment.
Sources include: Axios
And that’s our show for today.
Show notes are posted at our new home at technewsday.com or .ca either works and you can even play the latest or past episodes on our home page. We’re also on You Tube with our experimental video show, getting ready for bigger and better video over the summer.
We love your comments and suggestions. You can reach us at editorial@technewsday.ca
I’m your host, Jim Love, have a Terrific GPTuesday.
