Low-res NLP for your everyday life
summary
Claim: simple string tricks beat full AI for many tasks. A Slack bot is built and shipped that answers FAQs (e.g. delete channel) by normalizing text, tokenizing, removing stop words, looking up seeded keyword pairs in a database and returning canned responses; no machine learning required. Concrete tools mentioned: superscript, Stanford tokenizers, the howdy toolkit. Slack size cited: 3 million daily active users with about a third paying. The feature works but relies on manual seeds and simple heuristics.
AI-written summary · about
chapters (15)
- 0:00Start
- 1:13I work at Slack
- 1:40600+ employees
- 2:41Natural Language Processing
- 3:38jargon can blur comprehension
- 4:02resiliency
- 4:47longevity demands grit
- 8:25sentiment vs understanding
- 9:46humans are great at context
- 15:13clean your input
- 15:47tokenizing
- 19:09handy libraries
- 19:27Stanford CoreNLP
- 20:00take ideas from other disciplines
- 20:26fix what annoys you!
related talks
17mUsing bots to Improve Humanity
46mMagic of Automation and Everyday Chores
24mConversational Interfaces for Chatbots and Artificial Intelligence
37mAlice's Adventures in Artificial Intelligence-land
49mMagic of Automation and Everyday Chores
55mEnterprise Chatbots (using messaging and AI to build novel user interfaces for work)
45mThe Humans Are Dead, We Are All Bots Now
46mMagic of Automation and Everyday Chores