No one actually knows why AI works

Description

No one really knows how generative AI works. Here’s how researchers working on AI interpretability are figuring it out. *Chapters* 0:00 What we don’t know about AI 1:42 What we *do* know about AI 3:15 Emergent capabilities 5:12 Building ≠ understanding 6:23 Interpretability research - golden gate claude 7:32 Interpretability research - circuits 8:31 Who cares? 10:11 Sorry this wasn’t a skit :/ *Papers Cited* * Emergent Abilities of Large Language Models: https://arxiv.org/pdf/2206.07682 * Attention is All You Need: https://arxiv.org/pdf/1706.03762 * Emergent Abilities of Large Language Models: https://arxiv.org/pdf/2206.07682 * Are Emergent Capabilities of LLMs a Mirage: https://arxiv.org/pdf/2304.15004 *Anthropic Interpretability Research* * Golden Gate Claude: https://www.anthropic.com/news/golden-gate-claude * Tracing thoughts: https://www.anthropic.com/news/tracing-thoughts-language-model?utm_= *Videos explaining LLMs / transformers* * https://www.youtube.com/watch?v=eMlx5fFNoYc * https://www.youtube.com/watch?v=LPZh9BOjkQs *Merch* For developer mugs, hoodies and more, check out: https://albertatech-shop.fourthwall.com/ *Socials* https://instagram.com/alberta.tech https://tiktok.com/@alberta.nyc https://x.com/albertadevs #ai #largelanguagemodels #llms #airesearch #anthropic #softwareengineer This video is not sponsored. FTC Disclosure: This video contains affiliate links. If you purchase through one of these links, I may earn a small commission at no additional cost to you.