Youtube Video

Summary published at 9/18/2024

✈️ Recent Travels: I spent the past three weeks in the United States and Japan, engaging in discussions about the AI boom as we approach the second anniversary of ChatGPT.

💻 Hot Chips Conference: The main reason for my trip was the Hot Chips semiconductor design conference at Stanford. I attended in person, focusing more on one-on-one conversations than presentations. Key takeaways include:

  • Meta and Tesla are advancing in designing affordable ASICs for AI.
  • Co-Packaged Optics (CPO) is emerging as a new technology for chip data transfer.
  • Major companies like AMD and OpenAI did not present significant new developments.

📈 AI Applications: In my previous video, I questioned the profitability of LLM-based applications. Insights include:

  • Microsoft's Copilot is estimated to generate around $1 billion annually, but this is not sufficient for them.
  • ElevenLabs is likely making around $100 million in revenue, while Character.ai has high usage but struggles financially.
  • Promising startups leverage LLMs to enhance productivity, especially in handling large data sets.

📉 Price War in LLM APIs: The cost of GPT-4 API tokens has dropped from $36 per million tokens to $4, an 80% reduction. This trend is attributed to software improvements rather than new semiconductors.

🔍 Foundation Models: The growth of the LLM ecosystem is hindered by the limitations of current Foundation Models. The industry is eagerly awaiting GPT-5 for advancements.

🏢 Nvidia's Dominance: Nvidia leads the AI chip market, followed by AMD. The future of Nvidia's dominance depends on:

  • Emergence of competitive AI chips from tech giants like Google and Amazon.
  • Success of current AI chip startups, though many may fail.

📉 Intel's Challenges: Intel recently faced a 30% stock drop after a poor quarter. The company is considering options like capital expenditure cuts and potential splits.

🚗 Waymo's Progress: Waymo has become a common sight in San Francisco and Los Angeles, showcasing the success of autonomous vehicles through extensive mapping and data collection.

🔮 Future Outlook: The focus is shifting towards the next generation of Foundation Models and the scaling of data centers. Anticipation builds for what the next two years will bring in AI advancements.

Download our Chrome extension for Youtube summaries