12 Key Developments in AI Hardware and Models in Jan 2024

What are people building in this sector

Welcome to Infinite Curiosity, a weekly newsletter that explores the intersection of Artificial Intelligence and Startups. Tech enthusiasts across 200 countries have been reading what I write. Subscribe to this newsletter for free to receive it in your inbox every week:

---

Here are 12 key developments in AI hardware and models during Jan 2024 that caught my attention:

1. Nvidia and AMD shares soar to all time highs. This surge is driven by the booming demand for AI chips. A rising tide lifts all boats.

2. OpenAI continues to pursue its chip manufacturing plan. Sam Altman is planning to raise billions in funding to develop AI chips. There's an increasing demand for specialized hardware to handle AI workloads. This move points to the fact that they want to do more vertical integration to make sure they are not at the mercy of the AI supply chain.

3. Figure has struck a deal with BMW to integrate general-purpose robots into car production. This marks a significant step towards automating complex manufacturing processes using robots.

4. DeepMind is training robots to do real-world tasks. They have built a system called AutoRT which aims to use AI to create robots that can understand human goals. AutoRT can simultaneously direct multiple robots where each equipped with a video camera and an end effector.

5. Microsoft has introduced an AI-specific key on its keyboards. This is the first major update since the addition of the Windows key.

6. MIT's Improbable AI Lab has come up with a new multimodal framework called Compositional Foundation Models for Hierarchical Planning. It develops detailed and feasible plans with the expertise of three different foundation models.

7. Google DeepMind developed an AI system capable of solving complex geometry problems.

8. Unified-IO 2 is a multimodal model with 7 billion parameters that can handle text, images, audio, video, and action sequences, setting new benchmarks in various modalities.

9. SOLAR 10.7B is a large language model with 10.7 billion parameters. It uses depth up-scaling (DUS) for scaling and outperforms other models in benchmarks.

10. Mobile ALOHA uses imitation learning from human demonstrations to teach robots bimanual mobile manipulation tasks. It includes a low-cost teleoperation system and shows that co-training helps in learning complex tasks efficiently.

11. FlowVid introduces a method for video-to-video synthesis that uses optical flow to improve efficiency and quality despite flow imperfections.

12. DeWave is a framework for translating EEG waves into text. It uses discrete codex encoding and achieves top performance in EEG translation.

If you're a founder or an investor who has been thinking about this, I'd love to hear from you. I’m at prateek at moxxie dot vc and our fund Moxxie Ventures leads seed rounds.

---

If you are getting value from this newsletter, consider subscribing for free and sharing it with 1 friend who’s curious about AI: