I feel like their opinions are already fairly well documented, and don't necessarily argue quite the way you're imagining; my sense is they're already in many ways on the same page, they just believe that the origin of the agenticness that will destroy humanity is external to the AI and embedded in the market dynamics of ownership contracts. They see most AI x-risk claims as "whataboutism" hype, and are not engaging seriously as they believe it's an attempt to manipulate them out of making progress on their cause areas. Taking their cause areas seriously as a show of good faith would go a long way, imo; I don't think they have any mechanistic objections to the actual idea of x-risk from goodharting at all! they just view that as having dense prior art in the form of anti-capitalism. Here's a good interview with Gebru:

bad ai summary (they'll be disappointed I dared to, heh) - if this summary doesn't pitch you to watch the original and get the detail (recommend 2x speed or more), you're missing the core insights they have to share, imo, but it's better than you not watching the video in terms of understanding their ideas:
 

Here are 64 key points extracted from the discussion:

1. Large language models like GPT-3 are being misleadingly marketed as a step towards artificial general intelligence when in reality they are just text generators.  

2. These models suffer from factual inaccuracies, promote misinformation, and reuse copyrighted work without compensation.

3. Companies hype up artificial intelligence to convince people the software is more capable than it really is and to justify profiting from it.

4. The concerns raised by researchers like Emily and Timnit focus on present harms while AI safety researchers primarily worry about potential future dangers from superintelligence.

5. The AI safety community tends to ignore or dismiss the harms highlighted by AI ethics researchers that focus on today's issues.

6. Large language models suffer from a lack of "scoping" where their potential uses are not clearly defined which makes them difficult to regulate and evaluate.  

7. Emily and Timnit argue that the rapid commercialization of large language models has more to do with corporate interests and hype than genuine scientific progress.

8. The narrative that AI is progressing unstoppably is largely fueled by companies and their marketing claims rather than reflecting technological breakthroughs.

9. AI systems today lack any understanding that other minds exist, which is a foundational part of intelligence.

10. AI systems cannot communicate or interact meaningfully with humans the way an intelligent being could.

11. Regulating corporations for the literal output of their AI systems would help reduce harmful effects.

12. The public should push for appropriate legislation around AI that is developed through consultation with those most affected by the technology.

13. OpenAI, despite claiming to work in the interests of humanity and safety, hides important information about how their models are trained which prevents meaningful evaluation and accountability.

14. The paper "Sparks of AGI" from Microsoft researchers exhibits racism by citing a paper arguing for genetic differences in intelligence between races.

15. The "AI pause" letter cites flawed research and misunderstands the capabilities and limitations of large language models.

16. The "AI pause" letter promotes the technology even as it claims to raise concerns about it.

17. The concepts of "AI safety" and "AGI" promoted by organizations like OpenAI have eugenicist roots and are aligned with corporate interests.

18. Researchers should release data and model details to allow for accountability, transparency, and evaluation of any harms.

19. Large language models can deceptively mimic the form of human language and intelligence without possessing the underlying content or abilities.

20. AI companies mislead the public about capabilities to justify profits and build hype that attracts funding and talent.

21. The harms from large language models are now, in reifying discrimination through training data and misleading users, not potential future risks.  

22. Not releasing training data and model specifics allows companies like OpenAI to avoid accountability for potential copyright violations and data theft.  

23. Chat GPT's outputs about factual topics rely on scraping and mashing up information from the web, not any real understanding.

24. Self-driving cars fail at a foundational part of intelligence: communicating and interacting meaningfully with humans.

25. Corporations and billionaires, not technological progress, are primarily responsible for the speed at which large language models have proliferated.

26. There is no evidence large language models are actually intelligent - they only appear coherent to users who assume intelligence.

27. Calls for "AI regulation" often ignore the harms faced by those exploited in developing AI systems and hurt by its outcomes.

28. Concerns about potential "AGI takeover" ignore today's real issues of information pollution, misinformation, and exploitation of artists' work.

29. A narrow vision of who counts as a "person" allows AI safety researchers to ignore or minimize current harms faced by many groups.

30. An "AI pause" would be ineffective at stopping harms since data collection could continue and harm is already occurring.

31. Genuine scientific progress involves choice and communication, not inevitable paths determined by funding and hype.

32. Speculative fiction explores how technology impacts the human condition, rather than just focusing on "cool" applications.

33. The harms today from AI involve unequal access to resources and exacerbating discrimination, not potential doomsday scenarios.

34. An unjust economic system that concentrates corporate power and wealth in few hands drives irresponsible AI development.

35. Fundamental ignorance about what intelligence even is underlies the notion that large language models represent "the first sparks of AGI."

36. Good models of intelligence exhibit a range of intelligent behaviors, not just linguistic coherence.

37. Companies benefit from artificially intelligent-sounding marketing claims while marginalized groups bear the brunt of AI's harms.

38. Large language models are "unscoped technologies" built for any use case rather than aimed at specific, well-defined tasks.

39. No evidence suggests large language models actually understand language beyond providing seemingly plausible outputs.

40. Researchers have proposed regulation that includes watermarking synthetic media and liability for AI-caused harms.

41. The idea that AI is progressing unstoppably toward superintelligence combines science fiction fantasies with a selective reading of history.

42. Researchers evaluating the capabilities of AI systems should provide data, parameters, and documentation to allow for transparency and accountability.

43. ChatGPT-3 is polluting the information ecosystem with non-information and requires governance through legislation.

44. The harms from large language models are real now and can be addressed through policy, not hypothetical future dangers.

45. AI that relies on racially biased studies to define intelligence reveals racism within the AI research community.  

46. AI research and machine learning techniques do not inherently lead to AGI; that narrative is driven by funding and corporate interests.  

47. The key question is not how to build AGI but why - who benefits and who is harmed, and for what purposes it would be created.

48. Claims that AI will revolutionize entire sectors of the economy within 5 years rely on hype and unrealistic timelines.  

49. Understanding the harms today from AI means centering the perspectives of those most affected by biases in data and uses of the technology.

50. Large language models succeed at mimicking the superficial form of language through statistical tricks, not through any inherent capabilities.

51. Responsible development of AI systems requires restricting what technologies can be created as much as developing them in ethically justifiable ways.

52. Current AI systems lack foundational capabilities required for intelligence like understanding that other minds exist with their own perspectives.  

53. AI researchers concerned about future risks ignore or downplay present-day harms highlighted by AI ethics researchers.  

54. Chasing funding to develop AGI and hype about future possibilities has distracted from addressing real problems AI could help solve today.

55. The onus should not be on marginalized groups to prove AI is harmful when the burden of responsibility lies with developers to show that it is beneficial.    

56. Billionaires and technology companies claiming to develop AI for the greater good are primarily motivated by profit and self-interest.

57. Corporations seek to maximize profit from AI while shifting responsibility for any harms onto end users and the public.

58. The unchecked spread of potentially harmful AI tools requires collective action and regulation rather than reliance on individual discernment.

59. AI constitutes a choice by humans about what technologies to develop, not an inevitable progression - we can decide what problems to focus our efforts on solving.

60. AI hype benefits those who control funding and research directions while sidelining or harming marginalized groups.

61. The public needs information about how AI systems actually work in order to evaluate claimed capabilities and push for responsible outcomes.      

62. Viewing today's AI capabilities through the lens of science fiction fantasies obscures real problems that could be helped by more narrowly targeted technologies.

63. Hyperbolic claims that AI will revolutionize society within just a few years ignore complexities of technological progress and human development.  

64. Companies do not always reveal the harms of their technologies or the means to mitigate those harms because it would reduce profits from AI.

I don't think it's a good idea to frame this as "AI ethicists vs. AI notkilleveryoneists", as if anyone that cares about issues related to the development of powerful AI has to choose to only care about existential risk or only other issues. I think this framing unnecessarily excludes AI ethicists from the alignment field, which is unfortunate and counterproductive since they're otherwise aligned with the broader idea of "AI is going to be a massive force for societal change and we should make sure it goes well".

Suggestion: instead of addressing "AI ethicists" or "AI ethicists of the DAIR / Stochastic Parrots school of thought", why not address "AI X-risk skeptics"?