We created software that could generate human-like text output quickly and easily. Now we're dealing with the societal upheaval it's caused. What are the risks and rewards, and what can we learn about language from these large language models? Daniel — joined by Caitlin Green — has a chat with Dr Christopher Summerfield, author of These Strange New Minds: How AI Learned to Talk and What It Means.
Timestamps
00:00 Start
00:46 Intros: Generative A.I. concerns
04:15 Shout out to our patrons!
05:03 News: AP Style Guide defines COUPLE
10:35 News: Men do vocal fry more
14:59 News: Uptalk from 1890
16:01 News: Is Singlish up?
22:22 Related or Not: Bonkers Mélange editon, theme from Ste
23:41 Related or Not: population, discombobulate, bobbin
29:09 Related or Not: goggle, goo-goo, agog
36:09 Related or Not: once, ounce, pounce, lynx
41:55 Interview with Christopher Summerfield: Do you like A.I.?
44:21 Consequences of AI: Will we know nothing, or know everything?
47:03 Are LLMs just spicy autocorrect?
48:44 Are LLMs simply regurgitating their training data?
49:51 LLMs are getting better fast
52:33 On consciousness and intentionality
55:58 Do LLMs (or humans) understand?
58:58 The Chinese Room
01:01:00 Should we avoid anthropomorphising language around LLM behaviour?
01:04:02 Why we dismiss LLMs
01:07:26 Accelerationists, anti-hypers, and X-risk: Which are you?
01:09:49 Safety, privacy, and security
01:14:29 The magic wand of policy
01:20:18 Fixing the hallucination problem
01:27:36 Goals of the book
01:31:18 Word of the Week: liminal
01:39:59 Word of the Week: pink slime journalism
01:44:44 Word of the Week: waste colonialism
01:48:13 Quick words: hot-washing, eppy, shoulder surgfing, news-jacking, bio-break
01:51:37 Word of the Week: wario
01:55:02 The Reads
02:01:06 Outtake: That time when a siren went off in Hedvig's Parisian hotel, mid-recording
Video for this episode: https://youtu.be/94SbeM0KpWw