Mistral Drops a Trillion Parameter Open Model
Oct 8, 2026 · 1h 14m
Summary
Hosts Jason Howell and Jeff Jarvis discuss the release of two major open-weight AI models, Mistral Large 4 and Reflection AI's Beam, which challenge Chinese dominance in the sector. They also examine the launch of Ghost's $3,500 personal AI computer and NVIDIA's new DGX Spark, analyzing the hardware trends for local model inference. The episode covers AI's expanding role in government services and small businesses, while addressing privacy concerns surrounding Meta's proactive AI assistant, Muse.
Topics discussed
Sponsor: American Express Business Gold Card
Show Preview: Open-weight models and AI hardware
Introduction and Jeff Jarvis's Haverhill book event
Mistral Large 4: Naming and Mixture of Experts architecture
MoE explained via the Osborne 1 computer analogy
European sovereignty and Mistral's competitive stance
Arthur Mensch interview: Business models and open weights
Security capabilities and the future of AI margins
Hardware costs and running models locally
Reflection AI's Beam model and US open-weight options
The trend of open-weight models and user choice
Simplifying AI hardware for the general public
Patreon support and daily podcast updates
Sponsor: American Express Business Gold Card
Ghost Core: The $3,500 personal AI computer
Ghost Core specs, local privacy, and founder age
Ghost Core pricing analysis and GPU supply issues
RAMageddon and the rising cost of memory
NVIDIA DGX Spark and local AI hardware trends
Local AI for small businesses and government use cases
Privacy benefits of local AI and potential improvements
Hands-on experience with Meta Muse proactive agent
Wired report on Meta Muse data collection and privacy
The trade-off between AI utility and personal data
Local models as the future of private AI analysis
Xreal smart glasses hardware review and comparison
Smart glasses market: Amazon, Meta, and use cases
Comparison with Apple Vision Pro and Meta Quest
AI in dermatology: Autonomous diagnosis of zits
Skepticism regarding AI medical reliability
Show break and transition
Sponsor: Cisco Outshift and the Internet of Cognition
Sponsor: American Express Business Gold Card
Google's open-source on-device AI app
OpenAI's TextGrain watermarking for AI text
OpenAI introduces ads in ChatGPT
The trade-off of free AI services and data targeting
Google Nano Banana 2.1 image generation model
Elon Musk rebrands xAI to 'Superintelligence'
Recap of Jeff Jarvis's Haverhill event details
Podcast consulting and commiseration services
Show resources, merchandise, and Patreon links
Thanks to supporters and executive producer perks
Sponsor: Apple Watch Series 12 health features
Listen ad-free on Castria