🐙 Lunch & Learn: ChatGPT-4o

The speaker discusses the recent updates from Open AI, particularly the release of the GPT 40 model. This model is not a new base model like GPT 5 but an enhanced version of GPT with improved intuitiveness, better vision, and audio capabilities. It allows for real-time communication. The speaker also mentions a live demo showcasing the model's capabilities and notes that GPT 40 is available to all users, including free users. The excitement is around its potential applications in various fields.

The speaker demonstrates the real-time conversational speech capabilities of the GPT model. They use a phone to interact with the GPT app, showing how it can converse naturally and even help with tasks like calming nerves before a live demo. The model's ability to understand and respond to speech in real-time is highlighted, along with its potential applications in customer service and as a more conversational AI, similar to what Siri was intended to be.

The speaker explores GPT 40's vision capabilities by demonstrating how it can interact with video and solve math problems shown on paper. They also discuss how the model can assist with coding problems by analyzing code snippets. The demonstration includes a scenario where the model helps with a linear equation and describes a coding problem involving weather data processing. The speaker emphasizes the model's improved ability to understand and generate images and its potential use in education and accessibility applications.

The speaker tests GPT 40's image generation capabilities by prompting it to create an image of an adorable anime-style octopus looking for Lauren Lee at the beach at night. They compare the results with GPT 4, noting the speed and quality of the generated images. The discussion includes the limitations of the free version of GPT 40, such as quota restrictions, and the speaker's personal experience with trying to access the audio functionalities of the model.

The speaker speculates on the potential applications of GPT 40, including its use in real-time translation, customer service, education, and even as an AI companion for lonely individuals. They discuss the possibility of AI therapists, AI companions, and AI girlfriends or boyfriends, highlighting the model's conversational abilities and emotional recognition. The speaker also mentions the impact of such technology on jobs, particularly in call centers.

The speaker shifts focus to Google's recent announcements, discussing the Gemini model and its integration into Google Workspace. They highlight Gemini's ability to summarize emails, search through photos, and understand context across different formats. The speaker also introduces Project Astra, Google's vision for a universal AI agent capable of reasoning, planning, and memory, and shares a video demonstrating its capabilities.

The speaker talks about Google's new pair of prototype AR glasses and their connection to the Gemini model. They discuss the potential of these glasses to provide real-time assistance through Gemini. Additionally, the speaker mentions Google's generative video model, ViVi, which can create high-quality 1080p videos from text, image, and video prompts. The discussion touches on Google's hardware and chip developments and their strategy compared to Open AI.

The speaker details Google's new Gemini 1.5 Flash model, designed for fast and cost-efficient multimodal reasoning at scale. They also discuss Google's generative AI advancements, particularly in search, and the integration of these technologies into Google products. The speaker expresses skepticism about Google's ability to deliver on their promises, based on past experiences with their demos.

The speaker discusses Google's commitment to open source models, such as Gemma, and their focus on AI responsibility. They mention Google's practices like red teaming to test and improve their models. The speaker also talks about the potential of AI in education, with features like interactive educational videos on YouTube, and the importance of building AI systems that are safe and beneficial for society.

The speaker reflects on the strategies of Google and Open AI, discussing the challenges of building products dependent on technologies controlled by large companies. They express concerns about the potential for these companies to cannibalize smaller startups and emphasize the importance of focusing on AI security and responsibility. The speaker also mentions the departure of Ilya Sutskever from Open AI and speculates on the implications of such moves in the AI industry.

In the final paragraph, the speaker wraps up the discussion, apologizing for technical difficulties that delayed the start of the stream. They express a desire to explore more about AI safety and encourage others to get involved in open source and regulatory aspects of AI development. The speaker also acknowledges the excitement and potential doom associated with rapid AI advancements and ends the session by thanking the audience for their participation.