Investors should pivot toward Edge AI by monitoring hardware leaders like Qualcomm (QCOM) and AMD (AMD), as companies like Liquid AI shift intelligence from the cloud to local devices. Watch for Mercedes-Benz (MBGYY) to lead the automotive sector in 2024 by deploying efficient, on-device models that operate without data center reliance. The rise of "Sovereign AI" makes open-weight models a high-conviction play, favoring platforms that allow enterprises to fine-tune AI on-premise to protect proprietary data. Meta (META) remains a dominant force in healthcare disruption, using its free MuseSpark models to demonetize medical diagnostics and create a massive competitive moat. In the biotech space, focus on longevity startups like Revel Pharmaceuticals that are moving beyond symptom management to actively reverse cellular aging through protein repair.

Liquid AI

• Liquid AI is a foundation model lab originating from MIT’s CSAIL, led by CEO Ramin Hasani.
• The technology is based on "Liquid Neural Networks," inspired by the 302-neuron nervous system of the C. elegans worm.
• Key Architectural Shift: Unlike traditional "Transformer" models (like GPT), Liquid AI uses "post-transformer" architectures that are more expressive and computationally efficient.
• Small Language Models (SLMs): The company focuses on efficient, general-purpose AI that can run "on-device" (phones, cars, laptops) rather than relying solely on massive data centers.
• Partnership with Mercedes-Benz: Liquid AI is deploying a multimodal model (less than 1GB in size) into Mercedes vehicles in North America starting in 2024.
  • It will run locally on the car's hardware, allowing for offline functionality, privacy, and direct control of 700+ vehicle functions.

Takeaways

• Investment Theme: Watch for the shift from "Cloud AI" to "Edge AI" or "On-Device AI." Companies that can shrink high-level intelligence to run on cheap, local chips (like Qualcomm or AMD) are positioned to capture the enterprise and automotive markets.
• Efficiency over Scale: While the industry has focused on "bigger is better," Liquid AI proves that specialized, smaller models can provide "frontier-level" intelligence with significantly lower energy and hardware costs.

Thinking Machine Labs (Inkling)

• Founded by former OpenAI CTO Mira Murati, the startup recently released its first model, Inkling.
• Technical Specs: It is an "open-weight" mixture-of-experts (MoE) model with 975 billion parameters, though it only uses 41 billion at any given time to maintain speed.
• Strategic Positioning: Murati is betting on customization over "leaderboard dominance." The model is designed to be downloaded and fine-tuned by enterprises on their own hardware.

Takeaways

• Bullish Sentiment for Open-Weight Models: There is a growing demand for "Sovereign AI"—models that companies can own and run on-premise to avoid sharing proprietary data with giants like OpenAI or Anthropic.
• Fine-Tuning as a Service: The next wave of value in AI may not be the "base model" but the tools and platforms that allow companies to customize those models for specific industries (legal, medical, engineering).

WeCo AI (Recursive Self-Improvement)

• A London-based startup claiming experimental evidence of Recursive Self-Improvement (RSI).
• Their system, AID², uses an "outer" AI agent to rewrite the code and research strategy for an "inner" AI agent.
• Performance: They claim eight days of machine self-improvement outperformed two years of human expert effort.

Takeaways

• The "Hard Takeoff" Scenario: If AI can successfully begin improving its own code, the pace of innovation moves from human-speed to machine-speed.
• Investment Risk: While exciting, some experts (like Ramin Hasani) remain skeptical of "sparks" of RSI, noting that true self-improvement requires changing the fundamental "weights" and architecture of the model, which is currently computationally expensive.

Meta (META)

• Meta’s MuseSpark 1.1 recently beat GPT-4o (Sol) on professional medical benchmarks.
• Democratization: Meta is providing these high-level diagnostic capabilities for free through WhatsApp and Facebook, reaching 3.5 billion users.

Takeaways

• Healthcare Disruption: AI is demonetizing medical diagnostics. Meta’s strategy of releasing high-performing models for free creates a massive "moat" by making it difficult for paid competitors to charge for basic diagnostic AI.

Revel Pharmaceuticals / Calico

• A breakthrough published in Nature Communications regarding Advanced Glycation End-products (AGEs)—the "scars" on proteins that cause stiff arteries, wrinkles, and cataracts.
• They demonstrated an enzyme (CMLase) that can "mow away" these scars, effectively reversing chemical damage previously thought to be permanent.

Takeaways

• Longevity Sector: This is a major milestone for Longevity Escape Velocity (LEV). Investors should monitor the biotech sector for "damage repair" technologies rather than just "symptom management."

Investment Themes & Sector Insights

AI Regulation & "Regulatory Capture"

• The FINRA Model: DeepMind’s Demis Hassabis and Elon Musk are calling for a regulatory body (similar to FINRA or the FAA) to test "Frontier Models" before release.
• Risk Factor: Analysts in the transcript warn this could be "Regulatory Capture," where incumbents (OpenAI, Google, Anthropic) create high barriers to entry to prevent smaller startups from competing.

The U.S.-China AI Race

• The White House is considering a "capability framework" that would allow U.S. companies to release open-source models only if they stay below the level of China’s best models.
• Insight: This creates a "perverse incentive" where U.S. innovation is throttled by the pace of Chinese releases.

Defensive Co-Scaling

• A new investment/safety theme: Using "Good AIs" to police "Bad AIs." As AI capabilities grow, the market for AI-driven cybersecurity and auditing will become essential.