
Sign up to save your podcasts
Or


Can you tell if an AI is lying to you? A new paper claims that we essentially can do exactly that, at least under the right conditions. Another paper claims we can inject various sentiments into responses, getting the AI to do what we wish. Interpretability is making progress. It is exciting to think about the implications. In the short term, it would be great if we could use this to steer responses and to detect and correct hallucinations. There's a lot of potential here to explore.
In the longer term, I am more skeptical of such strategies. I do not think lie detection is a viable primary control or alignment strategy. I worry that if we go down such a path, we risk fooling ourselves, optimizing in ways that cause the techniques to stop working, and get ourselves killed. Indeed, even attempts to grab the low-hanging fruit [...]
---
Outline:
(01:30) Table of Contents
(03:47) Language Models Offer Mundane Utility
(06:31) Language Models Don’t Offer Mundane Utility
(07:44) GPT-4 Real This Time
(07:48) Fun with Image Generation
(08:04) Deepfaketown and Botpocalypse Soon
(13:32) They Took Our Jobs
(16:10) Get Involved
(17:07) Introducing
(17:52) Meta Surveillance
(23:09) In Other AI News
(24:11) Open Philanthropy Worldview Contest Awards Prizes
(24:39) First Prizes ($50k)
(24:59) Second Prizes ($37.5k)
(25:17) Third Prizes ($25k)
(27:01) Quintin Doubles Down on Twitter
(29:27) The Other Winners
(31:24) Quiet Speculations
(39:13) Open Source AI is Unsafe and Nothing Can Fix This
(46:27) The Quest for Sane Regulations
(50:38) The Week in Audio
(52:03) Rhetorical Innovation
(59:48) Eliezer Yudkowsky clarifies a recent misunderstanding about the Orthogonality Thesis (here is his full best explanation of the thesis, from Arbital).
(01:02:49) Aligning a Smarter Than Human Intelligence is Difficult
(01:12:22) People Are Worried About AI Killing Everyone
(01:12:55) Other People Are Not As Worried About AI Killing Everyone
(01:17:48) The Lighter Side
---
First published:
Source:
Narrated by TYPE III AUDIO.
By zvi5
22 ratings
Can you tell if an AI is lying to you? A new paper claims that we essentially can do exactly that, at least under the right conditions. Another paper claims we can inject various sentiments into responses, getting the AI to do what we wish. Interpretability is making progress. It is exciting to think about the implications. In the short term, it would be great if we could use this to steer responses and to detect and correct hallucinations. There's a lot of potential here to explore.
In the longer term, I am more skeptical of such strategies. I do not think lie detection is a viable primary control or alignment strategy. I worry that if we go down such a path, we risk fooling ourselves, optimizing in ways that cause the techniques to stop working, and get ourselves killed. Indeed, even attempts to grab the low-hanging fruit [...]
---
Outline:
(01:30) Table of Contents
(03:47) Language Models Offer Mundane Utility
(06:31) Language Models Don’t Offer Mundane Utility
(07:44) GPT-4 Real This Time
(07:48) Fun with Image Generation
(08:04) Deepfaketown and Botpocalypse Soon
(13:32) They Took Our Jobs
(16:10) Get Involved
(17:07) Introducing
(17:52) Meta Surveillance
(23:09) In Other AI News
(24:11) Open Philanthropy Worldview Contest Awards Prizes
(24:39) First Prizes ($50k)
(24:59) Second Prizes ($37.5k)
(25:17) Third Prizes ($25k)
(27:01) Quintin Doubles Down on Twitter
(29:27) The Other Winners
(31:24) Quiet Speculations
(39:13) Open Source AI is Unsafe and Nothing Can Fix This
(46:27) The Quest for Sane Regulations
(50:38) The Week in Audio
(52:03) Rhetorical Innovation
(59:48) Eliezer Yudkowsky clarifies a recent misunderstanding about the Orthogonality Thesis (here is his full best explanation of the thesis, from Arbital).
(01:02:49) Aligning a Smarter Than Human Intelligence is Difficult
(01:12:22) People Are Worried About AI Killing Everyone
(01:12:55) Other People Are Not As Worried About AI Killing Everyone
(01:17:48) The Lighter Side
---
First published:
Source:
Narrated by TYPE III AUDIO.

26,392 Listeners

2,462 Listeners

1,102 Listeners

109 Listeners

296 Listeners

89 Listeners

553 Listeners

5,558 Listeners

140 Listeners

14 Listeners

140 Listeners

155 Listeners

459 Listeners

0 Listeners

143 Listeners