In this episode of AI Wanna Talk, host Jacob Norgord dives into a whirlwind of recent AI developments, covering everything from new models and massive investments to cutting-edge tools and applications. This episode is packed with insights about the ever-changing landscape of artificial intelligence.

Key topics covered include:

• Project Stargate, a $500 billion US initiative to build AI infrastructure across the country, involving companies like OpenAI, SoftBank, Oracle, and Nvidia.
• The groundbreaking Deepseek R1 model, which achieves performance on par with OpenAI's premier models at a fraction of the cost.
• A discussion on how Deepseek R1 is open source and what this means for how it can be used, including through Perplexity.
• How the release of Deepseek R1 has caused shifts in the perceived value of GPUs and other AI infrastructure.
• Eleven Labs raising $180 million to enhance their AI audio technology with an aim to make speech the standard of interacting with devices.
• OpenAI's "Operator" agent, which can control your browser to accomplish tasks and the early stages of its development.
• Krea’s custom AI models and their real-time 3D capabilities.
• Pika Labs' new 2.1 model, with a link to apply for early access.
• Adam, a new product that is deemed to be the future of CAD with its ability to generate CAD models from simple prompts.
• The release of another Chinese model, Qwen2.5-Max, which is said to perform better than GPT4o and Claude 3.5 Sonnet.
• An update on Grok, which is reportedly three times faster and has the ability to search the web and X (formerly Twitter).
• A tutorial on how to accept payments from AI agents.
• Deepseek's Janus-Pro 7B image model.
• Kanye West hiring an AI team.

This episode emphasizes how quickly AI is evolving, with innovative models and tools being released at an impressive rate, potentially leading to significant shifts in how we interact with technology in the future.

Links & Resources:

• Deepseek
• Perplexity
• Windsurf
• Eleven Labs Reader App
• Pika Labs 2.1 early access
• Adam CAD
• Qwen2.5-Max
• Grok
• Accepting Payments from AI Agents tutorial
• Deepseek Janus-Pro 7B
• Kanye West's Yeezy AI Team email

In this episode of "AI Wanna Talk," host Jacob Norgord explores a wide range of AI developments, focusing on new tools, platforms, and applications. The episode includes discussion of Perplexity's recent activities including their bid for TikTok, their acquisition of read.cv and their new sports feature. The podcast also examines new AI models, such as OpenAI’s o3-mini, and Apple's AI-generated news summaries. The episode also features a number of AI-powered tools including Windsor, Synthesis Tutor 2.0, FASHN AI, Krea AI, LumaLabs' Ray2, and Grok.

Key topics covered include:

• Perplexity's potential acquisition of TikTok and their purchase of read.cv, and how these might integrate with their search platform.
• A look at Perplexity's new real-time sports update feature, which can be displayed on a lock screen and provide in-depth game statistics, potentially replacing apps like ESPN.
• OpenAI’s O3 mini model and its advantages in speed and efficiency.
• A critique of Apple's AI-generated news summaries and the features of the new iPhone, such as emoji creation and image playground.
• An overview of Windsurf, an AI development platform that now has web searching capabilities, which allows it to research and implement APIs, and auto-generate memories to improve its programming efficiency.
• An introduction to Synthesis Tutor 2.0, an AI-driven tutor for kids that provides visual content and adapts to individual learning styles.
• A demonstration of FASHN AI’s technology, which allows users to see how clothing designs would look on a model.
• A look at Krea AI’s 3D layering tool that modifies a static image in real-time based on the orientation of a 3D object.
• An overview of LumaLabs' Ray2 video model.
• An update on Grok, which now has its own standalone website, as well as an iOS app, and can search both the web and X (formerly Twitter).

The podcast emphasizes how AI is evolving, with Perplexity leading in innovation and companies constantly pushing the boundaries of AI tools and applications.

Links & Resources:

• Potential Perplexity Acquisition

• Read.cv

• OpenAI's o3-mini

• Apple Pulls Back Generative AI

• Synthesis Tutor 2.0 Demo

• FASHN AI

• Krea 3D

• LumaLabs Ray2

• Grok

Correction: The Bee device records all participants in a conversation while being able to identify and distinguish your voice from other speakers.

In this episode of "AI Wanna Talk," host Jacob Norgord dives into the latest developments in AI, covering new apps, models, hardware, and even some thought-provoking ideas about the future of work and human interaction.

AI App and Model Updates:

• xAI's Grok App: A detailed look at xAI's Grok app, including its web-searching capabilities and integration with Twitter, offering a real-time information experience. The app allows image uploads and generation and is currently available on iPhone, with a web app coming soon.
• Bytedance AI Video Upscaler: Discussion of Bytedance’s new AI video upscaler, which uses stable diffusion techniques to enhance video quality, potentially impacting platforms like TikTok.
• Cohere’s North Platform: An overview of Cohere’s North platform, an all-in-one AI workspace designed for enterprises. It integrates LLMs, search, and AI agents into workplace tools like Google Drive, Gmail, and GitHub.
• Meta's Byte-Latent Transformer: Explanation of Meta's new byte-latent transformer, which processes data at the byte level, improving performance, reducing computational resources and handling multiple languages better. This approach bypasses traditional tokenization.
• Microsoft's Phi Model: Introduction of Microsoft's smaller 14 billion parameter Phi model, noting its ability to run locally and achieve high performance on benchmarks like the MMLU.
• Kokoro 82M Model: Overview of the Kokoro 82M text-to-speech model, which can run locally with only 350MB of RAM, making it ideal for device integration.

AI Hardware and Wearables:

• OMI Device: An exploration of the OMI wearable, a necklace-like device that provides contextually relevant information based on conversations and aims to connect directly to the brain in the future.
• B Device: Discussion of the B wrist-worn device, which summarizes conversations, suggests to-dos, and creates daily memories while respecting privacy, priced at $50.

Future of AI and Society:

• AI-Driven Advertising: Insight into how advertising might shift to target AI agents rather than humans, as suggested by Perplexity's CEO.
• Changing Time Allocation: A discussion of Paul Graham's tweet about increased time spent at home since 2003, raising questions about the impact of AI on our lives and how we'll spend our time. The podcast also touches on the idea of intentional inconvenience in the future.

Links & Resources:

• Adobe TransPixar AI
• Aravind's Advertising Prediction
• Bytedance's STAR AI
• Cohere’s North
• Meta's Byte-Latent Transformer
• Time Spent at Home
• xAI

Small Models

• Microsoft Phi-4
• Kokoro-82M

Wearables

• omi
• Bee