TL;DR
I temporarily returned to Japan to get a new visa, marking the start of my second year of research. Moving to a different lab than in my first year has been a valuable learning experience, allowing me to observe the differences in lab management.
(Photo: View of SF from the plane, including Golden Gate Bridge, Alcatraz Island, Bay Bridge, etc.)
0. Novel Prize in Physics 2024
Professors Hinton and Hopfield, giants in the AI field, received the Novel Prize in Physics 2024. This recognition of AI researchers in the physics award is a remarkable achievement. While physicists may have various reactions, the response in the CS community is overwhelmingly positive, boosting motivation.
1. Research
09/12 OpenAI o1 release: “deep-thinking reasoning model”
10/01 OpenAI dev day
Tue Lab Meeting
Thu Lab LunchI've begun discussing the research direction in the new lab.
New Release in AI Field
The OpenAI o1 release on 9/12 was a game-changer.
- OpenAI's intelligent model "Strawberry" has officially been named "o1," and the preview version, "o1-preview," was released.
- It was shown that scaling laws apply not only to training but also to inference (computing time during inference).
- Traditional LLMs focused on enhancing capabilities by spending computing costs on training and fine-tuning, but OpenAI discovered that spending on inference processing could also improve capabilities, realized through Strawberry (o1).
- With more time allocated to inference, a batch processing UI might be more suitable than an interactive one.
The 10/1 OpenAI dev day featured several announcements, especially highlighting Vision finetuning. When I previously tried In-Context Learning with images in Multimodal LLM, it couldn't compete with supervised learning. However, the new finetuning functionality promises to improve performance.
I'm also looking forward to the Tesla RoboTaxi announcement on 10/10. The RoboTaxi, which will undoubtedly transform society, will significantly impact companies like Cruise (GM, Uber) and Waymo (Google), with potential stock fluctuations depending on the content.
Meta’s AR Glasses introduced in September were also a very exciting announcement.
Otto is an intriguing SaaS product. Each cell contains an AI agent that conducts research, such as investigating company performance or competitor trends, automatically filling columns based on the titles.
My thoughts on AI Agents in Manufacturing
In LLM-OS (Karpathy), interfaces will increasingly take the form of AI agents. Disruptive innovation is already emerging in ERP and CRM. In manufacturing, I believe we’ll see a future where decision-making is entrusted to AI, moving beyond traditional Digital Twin Factory implementations (IoT-related). Preparing for a future where decision-making functions are API-based and performed by AI agents feels essential.
2. Life
08/31-09/02 🐟 Lake Tahoe
09/xx Boardgame
09/xx Hiking
09/xx @JapanDuring weekends, I gather with friends to enjoy hiking, board games, and BBQ. My friends guided me around the famous Lake Tahoe, and I thoroughly enjoyed the clear, pristine lake.
