Skip to content
#ML

Large Language Models: Friend, Foe, or Otherwise • Alex Castrounis • GOTO 2023

#ML #DataScience #Software #SoftwareDevelopment #Conference #AI

Interesting. keynote performance Alex Castrounis’ LLM at the Goto Conference in Chicago, Founder and CEOWhy of AI"and author of the book"AI for People and Business". In the presentation, the author discusses the following points:

  • AI today What artificial intelligence looks like today, how it intersects with machine learning, neural networks and deep learning. Here, he shows statistics from Google Trends search for artificial intelligence, machine learning and chatgpt, with the latter term in the Internet. 10 More popular than the previous two. Someone is afraid of the emergence of AGI, someone is afraid of the disappearance of jobs, and someone is in a hurry to integrate LLM into their products and optimize work.
  • What does a positive AI future look like? The importance of understanding how AI works to avoid fear of the unknown.)
  • Capabilities This is an example of what LLM can do. Here the author tells us about Hhugging Face And their library of what ML can do now.
  • What about coding? - discussion of code writing,
  • Evolution The author discusses key articles and results. -- 2017 - Attention is All You Need (Key article about transformers) -- 2019 - Language Models are Unsupervised Multitask Learners (OpenAI paper, gpt2 whitepaper) -- 2020 - Language Models are Few-Shot Learners (OpenAI paper, gpt3 whitepaper) -- 2022 - Training language models to follow instructions with human feedback (OpenAI used this approach to create chatGPT.) -- TBD
  • Overview General overview of NLP (natural language programming)Word semantics, context, and so on. The author speaks about word embeddings, the architecture of transformers. (encode and decode parts)about the RNN (recurrent neural network) And back to GPT--3/4, BERT, BART, T5.
  • Tools Next, the author talks about how LLMs are used to write code using the example of Copilot. (Goto is a programming conference:) )
  • Pitfalls & concerns Here the author talks about common sense, reasoning, causation, planning and so on. Why are LLMs beneficial? – They help us do things we’re not as good at or that take a lot of time.
  • The future of work Thinking about whether LLMs will take people’s jobs away or not:)
  • AI won't take your job, someone using AI will The final thought about work
  • Focus on & master what AI can't do - Advice on what to pump now

#ML #DataScience #Software #SoftwareDevelopment #Conference #AI