The new state of AI report is out. Here are my key takeaways.
- AI empowers physical science, biology, material sciences and drug discovery.
- Three eras of computing in machine learning: The pre-deep learning era, the deep learning era and the large-scale era.
- Diffusion models take over text-to-image generation.
Landmark models have been implemented/cloned/improved by the open-source community.
- Chinese papers focus more on autonomy, object detection, tracking, scene understanding, action and speaker recognition.
- U.S. papers focus more on question answering, text classification, text generation and speech recognition.
- U.S-based authors publish more AI papers than China, but Chinese institutions' output grows faster. Germany's growth rate disappoints.
- The China-U.S. AI research paper gap explodes if Chinese-language databases are included.
- Hyperscalers and AI compute providers teaming up with AI research laboratories.
Companies build more extensive computing systems than national supercomputing centres.
- DeepMind and OpenAI alums from new startups; Meta disbands its core AI Group.
- AI coding assistants are deployed fast.
- AI-first drug discovery companies have 18 assets in clinical trials.
- The first regulatory approval for an autonomous AI-first medical imaging diagnostic was achieved.
- Enterprise software is the most invested category globally, while robotics captures the largest share of VC investment into AI.
- A widening computing chasm separates industry from academia in extensive models of AI.
- AI in defence gathers momentum; Ukraine's homegrown geospatial intelligence software is a sign of coming things.
- The U.S lags in new fab projects, which take years to build. China's and Taiwan's fabs take roughly 650 days to make; the U.S. builds fabs 42% slower today than they did 30 years ago.
- AI researchers increasingly believe that AI safety is a severe concern and that AI safety is attracting more talent.
You can download the report with the following link: