EP 125 : Interpretability, Baidu's Price War, Llama to Space & Daily AI News

In this episode, we dive deep into the most critical AI updates and insights from today. We start with a look at Anthropic's push for AI interpretability, exploring why understanding models' inner workings is crucial as they become more powerful, using concepts like "mechanistic interpretability" and comparing it to an "AI MRI".

You'll hear how Anthropic has already mapped millions of "features" in Claude 3 Sonnet and why AI advancement is currently outpacing our ability to understand it.

Next, we head to China as Baidu debuts new Ernie AI models and tools at their Create 2025 event, publicly targeting rival DeepSeek.

Learn about the significantly lower costs of ERNIE 4.5 Turbo and ERNIE X1 Turbo reasoning model, offering substantial price reductions and competitive performance against models like GPT-4o and Deepseek's R1.

We also cover Baidu's new Xinxiang multi-agent system and Huiboxing digital avatar platform. This highlights the pressure Chinese labs are putting on global competitors through cost innovation.We also discuss Microsoft's long-awaited Recall feature landing on AI-enabled PCs, giving your computer a powerful "photographic memory" to easily find past documents or screenshots.

Plus, get details on the Manus AI agent's significant funding boost and new competition, and how startups like Tavus and Rime are blurring reality with new audio and lip-sync video models.

But that's not all! We take you beyond Earth with the fascinating news of Meta sending its Llama model to space, specifically to the International Space Station.

Discover how a special, fine-tuned version of Llama 3.2 is running locally on the ISS, made possible by its open-source nature, and how it's helping astronauts with research and accessing technical documents.

This groundbreaking use of an LLM in space is believed to be the first of its kind and could pave the way for future AI applications on Mars and the Moon.

Finally, we wrap up with "Everything else in AI today," covering updates from OpenAI's GPT-4o and lightweight research models, Elon Musk's Grok AI update for X, the Ziff Davis lawsuit against OpenAI, Moonshot AI's Kimi-Audio, Perplexity updates, Pleias's powerful new LLMs, Google's use of AI in coding, and major AI funding news for xAI and Nous Research.

We'll also highlight some of today's trending AI tools like Retellio, Adobe Firefly 4, Google's Music AI Sandbox, Hummingbird-0, Scrybe, Naoma, Seamless.AI, OpenJobs, and Shumer Prompt.

Tune in for your daily dose of the latest AI insights!