Kennedy saying that regulation shouldn't be done by congress who doesn't understand this technology; makes some REALLY BIZARRE claim about the existence of people who are able to use "AI" as it exists "to kill all of us and hurt us the entire time we are dying."

Kennedy is literally just EXPLICITLY saying "tell us how to regulate you."
#SenateAIHearing

Sen. Hirono asking about the definition of a "harmful request," and how OpenAI goes about denying those requests; altman talking about violence, self-harm, and "adult content;" Hirono brings it back to misinformation an deepfakes as harm, asks about regulation.

Goes on to ask what kind of regulations the witnesses envision right now, today; but also in the advent of "AGI;" Marcus highlights that we don't yet have a system with broad associative understandings of harm [We mostly don't have Humans who do that, Gary], then talks about external reviewers again.
#SenateAIHearing

Padilla mentions Homeland Security committee "AI" hearing happening, right now; goes on to talk about all the things called "AI" that we interact with already (phones, internet apps, etc); highlights that research into LLMs & their biases mostly focus on ENGLISH.

Montgomery focuses in on bias element and then pivots to a "we want people in LOTS of languages to buy our products!" Altman talks about trying to train on as many languages as possible, and talks about this technology can be a "big lift-up" to "historically under-represented" groups.
#SenateAIHearing

Jon Ossoff up now making a VERY IMPORTANT call to directly and clearly define what it is we're talking about when we talk about "AI;" Altman dodging this to talk about KINDS of regulation he wants to see, & talking about regulating by the "threshold of compute."

Jon Ossoff up now making a VERY IMPORTANT call to directly and clearly define what it is we're talking about when we talk about "AI;" Altman dodging this to talk about KINDS of regulation he wants to see, & talking about regulating by the "threshold of compute"

Ossoff pressing Altman about privacy and training systems on data on the wider internet without people's knowledge and consent; Altman dodges to "even if we don't train on it, it'll be able to link out to it;" Ossoff asks about whether some capabilities should be disallowed outright; altman agrees; asks about real material incentives to actually prevent harms to adults and kids.
#SenateAIHearing

Booker talking about the kind of harm that's been done by social media & the difference between getting fully out ahead of risk and having a FRAMEWORK for agilely adapting to risk; Montgomery agrees; Marcus says there's more to come and we need to start now.

Booker asks Altman about whether he will ever turn to an ad-based model; altman says "wouldn't say never;" Booker talks about the hyperconcetration of power in ad-driven corporate tech space; Altman says will be small number of people who can do high-level work.

Marcus talks about the danger of technocracy and the influence of technocrats on our beliefs; Altman shifts responsibility for values on society and the government.

Welch saying we "absolutely must have an agency" to regulate these systems and tools; while also asking questions about independent commissions and the "dangers" of "over-regulation;" asks them to define "perils" of an agency;

Altman goes "america must lead", follows w/ calling to make sure that there's enough "flexibility;"

Marcus highlights potential problems of regulatory capture;

Montgomery says we absolutely must hold companies accountable for harms they do.
#SenateAIHearing

Blumenthal highlighting the things not discussed in depth: Regulatory capture, national security ["China, China, China"]; goes to highlight again the problems of regulating these technologies w/o actual transparency/access to the technologies; pivots to privacy, where Altman talks about not training on anything gathered through the API [though they do retain that data for 30 days], and letting people opt-out of their interactions with ChatGPT being used for training; Montgomery talking about filtering and anonymization.

Asks marcus about self-awareness &self-learning capabilities; Marcus talks about medical advice & "existential risk" & making sure we consider & regulate for this; B. asks about misinformation,Montgomery talks about transparency of what's been made by "AI".

#SenateAIHearing

Marcus citing Dan Dennett in this #SenateAIHearing is not something I either wanted or asked for.

Josh Hawley asking about the fucking "AI Pause" open letter, either.

Booker showing his technocratic roots, even while urging caution; asking what we can do about the idea that a very few corps will control the field; Marcus talking about the problems of BingChat and how those made a space for others to be AS LOOSE with ethics, goes on to talk about personal shift from technical discussions to policy.

Booker asks Altman about concerns over how market forces and impacts will move toward reg. capture; Altman discusses "why we started as open source" [neatly skipping the part about why they closed]. #SenateAIHearing