TL;DR
- Research: Considering the application of the approach of Data-Centric AI. Initiated interaction with Snorkel AI.
- Life: Enjoyed the Lunar New Year. Found a workout partner and striving to maintain health.
The evolution of AI knows no bounds. This month, everyone was surprised by the announcement of Sora by OpenAI. Last year, a PhD student from Stanford CS founded Pika with a similar concept, and four months ago, they raised $100M in a Series-A round (Valuation: $500M). However, with the emergence of services like Sora, it's intriguing to see what will happen next. Nonetheless, in Japan, the business model of selling mundane products through "attentive customer service" and settling for less is not popular, so I want to focus on the creation of challenging and valuable innovations.
1. Research
02/06 🍽️ Visiting ScholarsAt the Visiting Scholars Lunch, I had the opportunity to speak with Professor Chris Manning. He is the Head of AI at Stanford CS and a renowned researcher in the field of NLP. I also said farewell to three Visiting Scholars whose terms are ending in March.
Progress
PoC Development
- Working on an implementation that generates Vector Embeddings triggered by the store to Image Storage as part of an image monitoring system and links it to Pinecone.
- For Text Only Vector Embedding API, OpenAI's text-embedding-ada series is standard, but since this is Multimodal, we are proceeding with model selection considering the balance between performance and cost.
- For Image and Vector Embedding, OpenAI's CLIP is the most famous, but since the API is not public, we are exploring other CLIP-based APIs and considering implementation on our own server using transformers (hugging face). Candidates include Meta's ImageBind and Salesforce's BLIP.
- In parallel with advancing feature development, we are creating a Demo Project and working on improving UX/UI.
Research Papers, etc.
- Sora(OpenAI): Video Generation
- Gemini1.5(Google): Multimodal LLM of 10M Context length
- Evo(Together AI): Foundation Model of Biology: from molecular to genome scale.
- Mamba: Superfast LLM
- As a module constituting the architecture of LLMs (Large Language Models), the State-Space Model (SSM) is gaining significant traction. The introduction of the Attention mechanism was somewhat heuristic, but if it can be replaced with SSM, it would become more mathematically rational and convincing.
Future Plan
- I was introduced by Professor Chris to the lab-originated startup Snorkel (Series-C, $1B as of 2021) and started interacting with them.
- I want to master the approach of Data-Centric AI and apply it to my service.
2. Life
02/02 🍽️ Global Chef
02/03 🍽️ @🇺🇸Tom&Steffi
02/04 🍽️ @🇮🇳Padma&Mohan
02/04 🌙 Lunar New Year Celebration @SunnyvaleLibrary
02/07 🌙 Lunar New Year Celebration @Stanford
02/10 🌙 Lunar New Year Celebration @🇨🇳Alan&Safari
02/11 🏈 Super Bowl
02/23 🍽️ Global Chef
02/28 🎤 The Sound of Black Music @BingConcertHall,StanfordLunar New Year
A festival celebrating the Lunar New Year. I was invited to a friend's house and enjoyed the food and conversation. I also participated in events held at public libraries and Stanford.
Super Bowl, Gun Shooting
I watched SF 49ers vs KC Chiefs at Stanford. Unfortunately, the 49ers lost, but it was an intense game that went into overtime. Meanwhile, incidents like the shooting at the Chiefs' parade and a Gun Shooting Alert at Stanford made me acutely aware of the gun society issue in America.
The Sound of Black Music
I attended a music event at Bing Concert Hall. Famous songs from The Sound of Music ("Do-Re-Mi", "Edelweiss", etc.) were performed by Black artists. It was a great atmosphere.
