18 2024 State of AI Report

The 2024 State of AI report from Air Street Capital is well sourced and Creative Commons licensed.

Here’s a high-level overview created by Google Gemini 1.5 on October 17, 2024. I chose to use Gemini because it’s embedded in my Google workspace.orlds and tasks.

• Model Updates: OpenAI maintains a slight edge with GPT-4, particularly in planning and reasoning. However, other labs and open-source models like Llama 3 are closing the gap. There’s a trend towards smaller, more efficient models for mobile devices, and quantization techniques are improving. Researchers are exploring hybrid models and synthetic data, while also addressing benchmarking challenges and dataset contamination.
• Research Trends: Research is expanding beyond language models into areas like mathematics, biology (with advancements in protein modeling and genome editing), physical sciences, and neuroscience. There’s a focus on improving reasoning capabilities, open-endedness, and the use of reinforcement learning and Monte Carlo Tree Search.
• Industry Developments: NVIDIA dominates the AI hardware market, and established GenAI companies are generating significant revenue. AI companies have reached a high valuation, driven by a public market bull run. There’s a growing interest in robotics, with advancements in manipulation skills, humanoid robots, and the use of Apple Vision Pro for teleoperation.
• Governance and Politics: Global AI governance efforts have stalled, but national and regional regulations are advancing, with controversial legislation in the US and EU. Big Tech companies are facing physical constraints on scaling and emissions targets. Concerns about AI’s impact on elections and employment haven’t materialized significantly.
• AI Safety: There’s a shift in focus from AI safety to acceleration, driven by commercial interests. However, governments are building AI safety capacity, and researchers are concerned about sophisticated attacks on AI systems.