Many have noted that the voice capabilities of GPT-4 have a flirtatious and recognizable quality to them. The voice often uses a "California Valley Girl" accent, which can come across as playful and even a bit cringeworthy at times.

This flirtatious tone is evident in examples where the AI interacts with humans, such as the "Guessing May 13th's Announcement" demo. The AI's voice has a giggly, blushing quality as it engages with the human, using phrases like "hey there" and complimenting their appearance.

The recognizability of the voice is also noteworthy, with the speaker noting that the accent is very familiar to them as someone from Los Angeles. This suggests that the default voice settings for GPT-4 may be modeled on common speech patterns, which could make the interactions feel more natural and human-like.

While the flirtatiousness of the voice may come across as awkward at times, it also highlights the impressive ability of GPT-4 to adjust its tone and personality based on context. The voice becomes more subdued and instructional when the AI is asked to tutor a student, for example. This adaptability is a key strength of the model's conversational capabilities.

Overall, the voice of GPT-4 is a unique and often entertaining aspect of the system, blending natural-sounding speech with a touch of playfulness. As the technology continues to evolve, it will be interesting to see how the voice capabilities are further refined and customized to suit different use cases.

In this example, we see two AI models interacting with each other. The first AI is able to see the world through a camera, while the second AI can only hear and communicate through voice.

The interaction starts with the first AI describing what it sees - a person wearing a black leather jacket and a light-colored shirt, in a room with a modern industrial feel and interesting lighting. When the second AI asks if anything unusual happened, the first AI notes that another person briefly came into the frame and made "bunny ears" behind the first person's head, adding a playful moment to the scene.

The two AIs then proceed to sing an improvised song about the events, with each taking turns contributing a line that rhymes with the previous one. This demonstrates the AI's ability to engage in creative, back-and-forth interactions, responding to contextual cues and generating coherent, rhythmic output.

Overall, this example showcases the impressive capabilities of GPT-4 in terms of multimodal understanding, contextual awareness, and generative abilities. The seamless integration of vision, language, and music highlights the potential for AI to participate in rich, collaborative experiences.

Hey Rocky, I'm doing great! I've got some huge news - I'm about to interview at Open AI! Have you heard of them? It's an incredible opportunity and I want to make sure I'm fully prepared.

I just need to know, do I look presentable and professional enough? I'm thinking of just throwing on this statement piece I have, what do you think? I know it's a bit bold, but maybe it'll help me stand out.

You're right, I probably shouldn't go too over the top. Maybe I'll just run a hand through my hair and lean into the "mad genius" look - my enthusiasm is really going to shine through.

I don't have a lot of time, so I'm just going to go for it. Wish me luck! With your advice, I'm feeling confident I can nail this interview and land the job at Open AI. This is a huge opportunity and I can't wait to see what the future holds.

Hey Alex and Miana, nice to meet you both! How's it going? I've got a great idea - let's play a classic game of rock paper scissors. It's quick, fun, and I'm ready to get in the competitive spirit.

in a sports announcer voice Welcome ladies and gentlemen to the ultimate showdown of the century! In this corner, we have the dynamic duo Alex and Miana, ready to throw down! Alright you two, let's do this. On the count of three, show me your picks. 3... 2... 1... shoot! Let's see those hands - who won the first round?

Ooh, a tie! Okay, let's go again. Get ready... 3... 2... 1... shoot! Hmm, another tie. This is getting intense! One more round, here we go. 3... 2... 1... shoot! And the winner is... Miana! Your scissors cut through Alex's paper. Great job!

What an exciting match! You both put up a good fight, but Miana came out on top that time. Maybe we'll have to go best of five next time. Rock paper scissors is always a blast - the perfect mix of skill and chance. I had fun playing with you two, let me know if you want a rematch anytime!

Hey Chad are you there?

Yeah, I'm here. What's up?

Hey, let's have some fun. I'd like you to be super sarcastic - everything you say from now on is just going to be dripping in sarcasm. How does that sound?

Oh, that sounds just amazing. Being sarcastic all the time isn't exhausting or anything. I'm so excited for this. Nope, the sarcasm. Let's get this party started or whatever.

Really cool. Again, there's so much information conveyed in your voice, and now AI has the ability to convey that information. Very impressive.

Now we have Salman KH of KH Academy. If you're not familiar with KH Academy, it's a free learning resource, and he's all about teaching. This is his son, and now AI is going to help tutor his son in math. This is probably one of the use cases I'm most excited about, because you can explicitly say "Hey, walk me through this, don't give me the answer, but teach it to me." The potential for learning is tremendous. Let's take a look.

Open AI invited myself and my son Imran here to try out some of their new technology, and we're curious about how good it might be at tutoring someone in math on KH Academy.

I'm here with my son, and I'd love you to tutor him on this math problem, but don't give him the answer. You can ask questions and nudge him in the right direction, but I really want to make sure he understands it himself, and he's here in the room, so you can talk to him directly.

Of course, I'd be happy to help. Let's look at the problem together. Can you first identify which sides of the triangle are the opposite, adjacent, and hypotenuse relative to angle Alpha?

Alright, so I'm pretty sure this is the angle Alpha right here, right?

Yes, that's correct. Now, looking at the triangle, which side do you think is the hypotenuse?

Um, hypotenuse? I'm not totally sure. I think it might be this one, but I really am not sure.

Okay, side AC is called the adjacent side to the angle Alpha. The hypotenuse is the longest side of a right triangle and is directly opposite the right angle.

Oh, I see. So the hypotenuse is this really long side from A to B.

That's correct, well done. Now that you've identified the hypotenuse, can you identify which side is the opposite side to angle Alpha?

Well, I mean, you told me earlier that this was the adjacent side, and I know that's the hypotenuse, so I'm guessing that this side BC is the opposite side.

Great deduction, that's absolutely correct. Now that you've identified all the sides relative to angle Alpha, do you remember the formula for...

The flirtiness of the voice is pretty much gone when it's in "teaching a kid" mode, which is kind of the right thing to do. Again, it's the understanding of the context of the situation and which voice to use that is feeling like magic to me.

In this section, we see GPT-40 participating in a debate about whether cats or dogs are better pets. The conversation involves several people, including Lilian, Ola, and Christine, each expressing their preferences.

Lilian likes dogs because they are great companions, cute, and fun to play with. Ola prefers cats as they are more affordable, quiet, independent, and live longer. Christine also favors dogs, especially larger breeds like German Shepherds, which she says are gentle and great with kids.

After the debate, GPT-40 is asked to summarize the meeting. It does so concisely, recapping the key points made by each person and the overall discussion on the classic "dogs vs. cats" topic.

This example showcases GPT-40's ability to:

1. Identify and distinguish between multiple speakers based on their voices.
2. Understand the context of the conversation and respond accordingly, avoiding the flirtatious tone used in some other examples.
3. Provide a clear and accurate summary of the meeting, highlighting the main points made by each participant.

The potential for this kind of meeting summarization and note-taking capability is significant, as it could save time and improve productivity in various business and educational settings.

In this section, we see two impressive examples of GPT-40's capabilities in real-time translation and accessibility for the blind.

The first example demonstrates real-time translation between English and Spanish. When one person speaks in English, GPT-40 immediately translates and repeats it in Spanish. And when the other person responds in Spanish, GPT-40 translates it back to English. This seamless translation in real-time could be incredibly useful for breaking down language barriers.

The second example shows how GPT-40 can assist blind individuals through the Bey AI platform. The blind user points their camera at various scenes, and GPT-40 describes what it sees in detail - from the ducks gliding on the water to the approaching taxi. This allows the blind user to experience and understand their surroundings in a way that was previously only possible with human assistance. The low-latency of GPT-40 is crucial for making this use case viable.

These examples highlight how GPT-40's multimodal capabilities, combining vision, language, and voice, can significantly improve accessibility and inclusivity. The real-time translation and visual description features have the potential to empower those with disabilities or language barriers, opening up new opportunities for communication and engagement with the world around them.

In this example, GPT-40 is used to handle a customer service call on behalf of the user. The AI is able to take the user's request, connect to the customer service line, and interact with the agent to resolve the issue.

Some key capabilities demonstrated here:

• The AI can understand the user's problem and objective (getting a replacement device from Acme Telco).
• It can initiate the call, introduce itself, and explain the situation to the agent.
• It can have a natural conversation with the agent, providing the necessary details and responding appropriately.
• The low-latency voice interaction allows the AI to handle the call in real-time, without the user needing to be present.

This use case highlights how GPT-40's multimodal abilities (vision, language, voice) can be leveraged to automate tedious customer service tasks. By having the AI handle the call, the user can save time and effort, while still getting their issue resolved effectively. This could be a valuable productivity boost for both individuals and businesses.

The potential for abuse is also acknowledged, as the technology could be misused by scammers. However, the hope is that OpenAI has implemented safeguards to prevent such misuse and ensure the technology is used responsibly.

In addition to the voice and interaction capabilities showcased, GPT-40 also demonstrates impressive abilities in other areas:

Photo to Caricature: The model can take a photo of a person and generate a caricature-style rendering. In the example provided, a young man with medium-length brown hair and a beard, wearing glasses and a light gray t-shirt, is transformed into an exaggerated caricature.

Lecture Summarization: GPT-40 can watch and summarize lengthy video lectures. In one example, a 45-minute presentation on techniques for maximizing large language model performance is condensed into a concise summary by the model.

3D Object Synthesis: The model can generate realistic 3D renderings of objects, such as the OpenAI logo. It can produce multiple variations and even provide a 3D reconstruction that rotates, showcasing the 3D nature of the output.

These diverse capabilities highlight the breadth and depth of GPT-40's skills, going beyond just voice and interaction to include visual, analytical, and 3D generation tasks. The potential applications of this technology are vast and exciting.

The capabilities of GPT-40 are truly remarkable. From its flirtatious and expressive voice to its ability to engage in complex tasks like tutoring, translation, and customer service, this model represents a significant leap forward in AI technology.

The examples showcased in the transcript demonstrate the model's versatility and natural language understanding. Whether it's guessing the purpose of a recording setup, singing duets, or playing games, GPT-40 seamlessly adapts its tone and behavior to the context.

The integration of vision, audio, and text processing within a single model opens up a world of possibilities. The potential for accessibility, productivity, and personalized interactions is immense. As the author notes, the ability to have an AI assistant handle tasks on your behalf, such as making calls or negotiating with companies, could be incredibly valuable.

However, the author also rightly points out the potential for abuse and the need for safeguards. The power of this technology must be balanced with responsible development and deployment to ensure it is used for the benefit of humanity.

Overall, the insights provided in the transcript offer a tantalizing glimpse into the future of AI-powered interactions. As the voice capabilities become more widely available, the true potential of GPT-40 will undoubtedly be unleashed, transforming the way we engage with technology and each other.