Gen AI round table

Original by Reinie, 2023Hamster_gagarin_linkedin
hamster writter This summary note was posted on 18 December 2023, by in Machine Learning #, #, #, #

Key takeaways

  • LLM: Large Language Models predict next token in the sentence
  • Output is not deterministic
    • the output will always be different
  • Have turned out to be highly versatile in its application
  • No lift and shift technology for businesses
  • Presonalisation of daily activities for employees
  • Personalisation of support and communication with clients
  • Main responsibility when checking the output of AI
    • Accuracy & trustworthiness
    • Common sens
    • Ambiguity
    • Bias
  • Moving from creating to assessing the outcome of the AI
  • The AI can do what we do when checking out on how to do things, save the time when we not necessarily create from scratch
  • We can look outside, learn from those already moving fast. Not necessarily need to lead the ways, rather learn from others. But make sure we are moving.
  • Move to Gen AI has to bring value
    • A question to see how useful it is was asked by asking if people would pay a portion of their salary to keep the tool. A high number of people said they would
  • Huge potential in the game industry with real complex time rendering for example
  • What is necessary to be successful?
  • You do nee to do a lot to prepare the output of AI with data training etc…
  • An issue is that we don’t really know how the AI is biased up front
  • We should focus on transition and how to u se AI as a mean to a better world
  • There are big ethic concerns.
    • Do we want to require to add a watermark to inform when something was generated with AI
  • Sustainability question in the use of AI and its energy consumption.
  • The scene is evolving
    • Bing Chat Enterprise | copilot pro as an alternative to ChatGPT, can also easily export to Excel
    • Github Copilot for developing. Code generation, Code explanation, code conversion
    • Bot training on internal data for specific business task to get for example information out of a massive FAQ or How to in a more efficient way