Part 1: AI implementations in businesses and companies
The most difficulty is creating "learning organization" with start-up lean culture
Encourage team member to use AI for finishing job as much as possible
Goal: use AI to create workflow
Before use AI to finish everything, all team members should capable to:
know business value
- What content & info that end user wants?
- those info are important, the most variable asset of the company.
- create a knowledge base for GenAI to provide tailor-made solutions for customers with different scenarios.
know business logic
- how to deliver content & info to the end user?
- Why does the content & info help the end user to complete their tasks?
- keep upgrading & improving selling point of business
- The market is always changing. solutions are workable today, However, may not be workable tomorrow.
- All members must realize market trends, customer's needs, real-world use cases, and industry best practices.
know money flow
- Why do end users pay money to us for buying AI content & info?
- (problem driven solution: help user to easily and quickly finish daily task)
- (research driven solution: provide industry insight & summary for making decision)
- What businesses need to pay to create an AI solution?
- (upfront hardware, data cost, computing cost)
- What is the package of a business selling AI solutions?
- (yearly subscription, monthly perineum plan, freemium + social media sharing)
- bring all team members on the same place, go to same direction, stay with each other
- Build consistency of knowledge base, data strategy, digital human tone of voice
gain domain knowledge of industry
- dive deep in logic of system, background of scenarios, emotion of stakeholder
- figure out why some solutions are not workable, some use case are not solving real-word problem
- study other successful business
- (How they create product and services to facilitate with existing market)
- go to demo day, hacker house, hackathon to talk with people as much as possible
play around with latest AI tools
- Start with some extremely low cost or even free online AI service
- Suggest AWS bedrock (play all famous LLM exclude ChatGPT)
- Build your own prompt engineering template to fulfill your unique requirements
- Build real time user feedback loop for upgrading & optimizing data sources, knowledge base, prompt engineering template
Part 2: Expectations for near-future AI advancements, including potential obstacles
Advancements:
"World Models" from Google Research is the next generation technology.
Existing Limitations of Large Language Models (LLMs):
- Lack of persistent memory
- Lack of visual models
- Can only process up to five frames
World Models are able to:
- Remember history
- Learn from experience
- Model real world
Predict future
Update the neural network modules that represent the state
Remember and model the environment
Take in the current observation (images, states, etc.) and the actions about to be taken
Model the memory and understanding of the world
Predict the next possible observation state and actions
Example:
Provide a video of people kicking a football, and World Models can predict the next frame of football.
Understand physical laws:
- Learn a detailed representation of world
- Realize comprehend mechanisms of physical world
Obstacles:
Compliance & Security
DeepFake scammer:
- criminal groups collects victim's personal info both online & offsite
- online:
- social media, fishing at telegram in the USA, VK in Russia, WeChat group in China
- offsite:
- talk with victim and hang out with them to collect some evidence for blackmailing
create victim's digital human through DeepFake
- create sexual content then making benefit on dating application
- create some hate, policy comments on social media for promoting criminal organization
People cannot recognize what content is created by AI
Difficulty distinct the different between reality and fake AI world
Develop wrong mindset towards AI technologies
Create misunderstanding towards real world because of fake AI content
Part 3: Your experience and expertise in AI
I build next generation virtual docker skin trauma diagnosis platform
- Help patient who have skin trauma to have flexible medicinal treatment
- Through GenAI Therapist with human empathy
- Aims to build, comfortable, medicinal, cost-effective system within smart city
People cannot get medicine care of skin trauma then lead to dangerous side effect
Example
- Worker:
- Long working hours, cannot taking sick leave from work for going to doctors
- Housewife:
- Take care of children, cannot go to doctors for medicine
My solution:
- Patient take a picture then send it to doctor
- Doctors digital humans make diagnoses then send prescription online.
- Patients immediately buy antibiotics & skin ointments at local drugstores on the same day.
Use Generative AI to provide:
1 Better mental care
2 Digitized diagnosis details
3 Digitized treatment records
4 Transparent costs
5 Seamless aftercare ensures recovery to reduce side effects
Reference:
https://research.google/blog/introducing-dreamer-scalable-reinforcement-learning-using-world-models/
Introducing Dreamer: Scalable Reinforcement Learning Using World Models
https://zhuanlan.zhihu.com/p/661965660
δΈη樑ε(World Models)
https://worldmodels.github.io/
World Models
Top comments (0)