Large Language Models: Friend, Foe, or Otherwise
summary
Transformer research kicked off by the 2017 paper Attention Is All You Need leads to LLMs trained on massive corpora (web text, Books1/2, Wikipedia), GPT‑3 uses 175 billion parameters and GPT‑4 is rumored near a trillion; models get RLHF alignment but still hallucinate and lack common sense, producing confident-but-wrong outputs. Tooling exists—GitHub Copilot, Amazon CodeWhisperer, Hugging Face StarCoder—and a Copilot study of 95 developers shows users finish tasks 55% quicker (1h11 vs 2h51); Deloitte finds 82% report better job performance.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:52Intro questions
- 1:06AI today
- 7:42What does a positive AI future look like?
- 8:36Capabilities
- 10:49What about coding?
- 11:39Evolution
- 16:39Overview
- 34:16Tools
- 35:47Pitfalls & concerns
- 41:40Why are LLMs beneficial?
- 42:04The future of work
- 47:04AI won't take your job, someone using AI will
- 48:48Now is the time to seize the day
- 49:24Focus on & master what AI can't do
- 50:04Outro
related talks
43mBeyond the Hype: A Realistic Look at Large Language Models
40mThe Future of Language Models
52mUnleash the Power of Large Language Models (LLMs)
30mLarge Language Models and Prompt Engineering for Developers
27mLarge Language Models and the Future of AI
15mFuture with Large Language Models
51mTowards a Pragmatic Understanding of Large Language Models
30mLearning to wield large language models