AI: Weekly Summary. RTZ #423

1. Google also lays out AGI levels a la OpenAI: Following last week’s OpenAI initiatives reported by Bloomberg to redefine AGI path to five levels over coming years, Google revealed a similar multi-level path in a new AI paper, for its AI rollouts around Gemini and other products. While OpenAI 5 Levels are Chatbot, Reasoning, Agents, Innovators and Organizations, Google’s 5 Levels are Emerging AI, Competent AI, Expert AI, Virtuoso AI, and finally Superhuman AI. No time frames outlined by either company. But these new frameworks point to a different roadmap template for other tech/AI companies large and small. More on all these developments here, here and here.

   

2. OpenAI’s ‘Strawberry’ and GPT-4O (Omni) Mini moves: Related to OpenAI’s next steps on the AI Level 2 ‘Reasoning’ Level, Reuters has a detailed report on the company’s new reasoning technology under code name ‘Strawberry’. This is an evolution of the reasoning project formerly code-named Q*. The piece highlights the technical innovations that are driving the nuts and bolts of moving AI from ‘chatbots’ to better reasoning and agentic capabilities. Separately, OpenAI released a ‘Mini’ version of its multimodal GPT-4 Omni LLM that better competes with larger models out there. Main driver is better economics of running both training and inference for customers across the board. More here.

   

3. Chip Stocks hit on US/China concerns: US tech and chip stocks were hit with a double whammy, with the Biden administration floating tougher trade rules vs China Chip industry, and Trump raised questions on US Taiwan support. Semi companies from Nvidia on saw 5%+ retrenchments with the renewed uncertainties, despite earnings reports from ASML TSM (Taiwan Semiconductor), and other tech companies. Meanwhile, Google, Microsoft and other cloud companies continue to have avenues to rent Nvidia-powered AI data centers to Chinese startups outside China. All points to continued volatility impact on US Tech/Semiconductor companies from US/China geopolitical issues.

   

4. Continued content AI scraping issues for Big Tech companies: This ongoing issue saw new fuel to the fire with reports of Apple, Nvidia, Anthropic using thousands of YouTube videos to train LLM AI models, presumably without permission. In this case, the transcripts of the videos may not have been directly scraped by the companies, but purchased via third-party intermediaries. This furthers the ongoing content licensing controversies that are seeing both litigation and negotiations to get to licensing agreements in many cases. Especially as issues of ‘Fair Use’ clash with the need for these ‘Extractive Data’ techniques to pursue ‘synthetic data’ generation for better LLM AI models. More here.

   

5. Meta leans back on new multimodal AI in the EU : Meta is now following Apple in holding back AI innovations in the EU for regulatory reasons. Lots of nuance and different regulations at issues in both instances. But a trend that is developing as the EU continues to press on a range of US big tech companies like Microsoft, Google, X/Twitter, amongst others on a range of issues. The broader issue remains the more defensive stance of the EU on new technologies and how they’re distributed by incumbents. Especially given the fast rate of change in AI investments and innovations by US companies both large and small. And similar trends in China. More here.

   

Other AI Readings for weekend:

1. China domestic AI headwinds; More here.

   

2. The globally disruptive Crowdstrike/Microsoft update points to broader resiliency issues as AI technologies are built on top of existing IT infrastructure.

   

Thanks for joining this Saturday with your beverage of choice. 

Up next, the Sunday ‘The Bigger Picture’ tomorrow. Stay tuned. 

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)