The Woman Who Showed President Biden ChatGPT | Hacker News Reader

The Woman Who Showed President Biden ChatGPT

wired.com

26 pointsbelter36 comments

Oh boy, years after years, after everything we went so far, there are people in the world who believes that Big Tech invests billions dollars into AI (including data centers and infra) with only socially good intent and we should not at all touch anything Big Tech do …

Why is this flagged? I would like to hear the reasoning from at least one of the people flagging it...

I find that any article that goes against the popular "tech" narrative of the day gets flagged.

I once read a great article on Hacker News where Bill Gates and Warren Buffet explained tangible problems with Bitcoin and other cryptocurrencies. It ended up being flagged, because the article went against the "Bitcoin is going to take over the world" narrative.

Read the comments. Did you think this article generated fruitful discussion?

I agree it did not generate a fruitful discussion...But that is not what flagging is for.

You seem to have stood up a lightning rod in a thunderstorm. Didn't flag it, but also it's paywalled so didn't upvote it either.

I can't stand the "OMG there's a new thing, we need to regulate it" way of thinking. No doubt in my mind the woman in the article doesn't understand much more about so-called "AI" than Biden.

> In an instant it became very, very obvious that the United States needed to speed up its efforts to regulate the AI industry

And the article says it as if that's some God-given truth...

The sad part is that AI won't "break out" and kill all humans aka stop this insanity. Life isn't a fairy^W cyberpunk tale.

You read her qualifications and history? You still decided to write this comment?

It seems like she has a background in EE and physics, but almost all of her work is policy oriented and not actually hands-on building things, at which point, "when all you have is a hammer" seems relevant. From the article:

> [Interviewer:] It’s almost predestined that you’d be in this job. As soon as you got your physics degree at Caltech, you went to DC and got enmeshed in policy.

I'd be inclined to trust an engineer-turned-policymaker more to decide if and what to do about an emerging technology more than the researchers/engineers developing it or the "entrepreneurs" pushing for its development.

Sure, but you could probably still find someone with a deep understanding of the tech who is not directly working in the field.

Or. Maybe you can't - the article also mentions how the US still lacks a CTO and they've tried to find one for the entire Biden administration.

My main issue is that policy makers love to...make policies. Which is exactly what EO14110 is all about. My major concern is that the US is going to box itself in with these regulations, and lose the race to China(or others), and our lives will ultimately be much worse because of it.

> hands-on building things

My piping hot take is that you want an expert policy builder in a job building policy. So you want someone with a lot of hands-on experience building policy. A senior SWE would have a steep ramp to get qualified for her job, because they do not have hands-on experience building policy.

For example: do they have contacts in the relevant organizations? Do they even know what the relevant organizations are? Do relevant Congresspeople know them and listen when they have something to say? None of the relevant skillset involves being hands-on with e.g. Python or vector databases.

Laws are made for things that exist. If new things appear, laws might need to be adapted - to be decided by the voters.

> Born in India and raised in Texas, Prabhakar has a PhD in applied physics from Caltech and previously ran two US agencies: the National Institute of Standards and Technology and the Department of Defense’s Advanced Research Projects Agency. She also spent 15 years in Silicon Valley as a venture capitalist, including as president of Interval Research, Paul Allen’s legendary tech incubator, and has served as vice president or chief technology officer at several companies.

This current iteration of "AI", by which I mean LLMs, seems only good at enabling generative spam and disinformation.

I think regulations are exactly what's required.

The government has the legal authority to kill, imprison, and tax citizens. It can destroy entire industries with the stroke of a pen. The military can wipe other nations off the face of the Earth.

And you want to wield that awesome power for some trivial problem like spam on the Internet? This is exactly the kind of overreach the founding fathers intended to prevent.

The "free market" mostly isn't.

LLMs are actively harmful to society and in a way that individuals cannot combat, and spam/fraud is merely the obvious way.

You've hit the nail on the head.

The government could stop all theft if people couldn't leave their homes except with police escort. Obviously this is a ludicrous idea. Ergo we have collectively (and correctly) decided that tolerating some level of bad actors is better than melting away all freedom.

The hard part is finding the right balance between security and freedom, no?

> seems only good at enabling generative spam and disinformation.

That seems overly pessimistic. They're pretty good at generating summaries of text, simple machine translation, generating alternative prose, turning unstructured text into table-like data, providing a text interface over a corpus of facts, and probably a bunch of things I'm not thinking of.

>They're pretty good at generating summaries of text, simple machine translation, generating alternative prose

They're not, though. For these tasks they are mediocre at best.

I disagree strongly. For the two two languages I can read the machine translations are quite good. I wouldn't count on it to faithfully render poetry or subtle elements in literature, but it's great for news articles or reading a form post.

For generating alternative prose, what I mean is using a prompt like: given this sentence/paragraph give me 3 alternatives that capture a particular mood/theme/style. Doing that is a great way to try to rewrite flat or repetitive prose or to shake up a bit of writer's block. It's excellent for turning some strung together facts into an email that I can quickly edit into my own voice a bit then send. You may not find that valuable, but I assure you that other people do.

Summaries also work pretty well in my experience and for my purposes. I don't need 100% perfection for every big document I come across sometimes I need a paragraph that hits the main idea(s) It's good enough for that and why wouldn't a text prediction tool be good at that?

> seems only good at enabling generative spam and disinformation

This doesn't seem to make sense. Although it uses the word "disinformation", so perhaps an LLM wrote it.

What about it doesn't make sense to you?

LLMs aren't intelligent and the hallucinations problem is fundamental to the technology, and cannot be solved.

Therefore definitionally an LLM will never be fully trustworthy.

But it's great at generating plausible-sounding stuff, which makes it great for propaganda and disinformation.

I've implemented many use cases professionally where LLMs sit at the center, from customer support, to document classification, general knowledge management and many more. Personally I use them frequently for coding, to carry out various text based tasks (turning information from one format to another etc). Hallucination can be significantly reduced by injecting knowledge into the request, to the degree that LLM output is suitable for various business use cases.

Saying LLMs are not useful for anything but generating spam is objectively wrong, as to disqualify your commenta from serious discussion about the topic.

There are many issues with GenAI that are worth discussing, saying they are generally useless is not the way.

> What about it doesn't make sense to you?

That you said it was only good for those things. You changed it in this message, as I'm sure you're aware.

Its in the nature of the beast, the boomer generation, being in the retreat-defeat of retirement, wants things stable, conservative, collapsing inwards at a snails pace. Your point of view has no party running

The thing is, we do need regulation but not the kind that's being proposed, which appears to be more like regulatory capture by OpenAI & friends to enshrine deep learning at scale as something only verified multinationals with NSA backdoors can do and put up even larger barriers for entry for anyone else.

What's needed is practically the exact opposite, a set of rules that prevents work automation to be owned by a select few who will then manipulate the resulting techno-feudalist world with impunity because they need no one else. Things are changing too quickly, the fastest they ever have in the history of man even. Some people are probably preparing to enter a college programme right now to learn something that will be fully automated before they even get their degree.

I think it's easier to take arguments more seriously when the statement doesn't generalize across an entire generation.

Everyone but psychopaths wants to live in a stable society, conservatism is apparently a trend among the young, and the boomer generation is all but dead at this point. But go off.

Everyone wants a stable society, but only the naive think you can regulate your way to a truly stable society.

All stable societies are stable due in no small part to regulation in some form or another.

I'm guessing by "truly" you mean "perfectly", but of course no such society exists in the real world.

By truly stable I mean that society changes. You must always have wiggle room. You can't just hammer down every nail the moment it sticks up. Regulates never more fast enough to compensate, so less regulation used to reach a desired outcome the better.

But that's exactly what's happening. AI is likely going to change society at every level, in innumerable ways. It's going to be the foundation of our economy. It's going to publish science. It's going to create art and culture. It's going to make legal and medical decisions. It's going to be the primary means by which humans communicate with each other, and with software. It's going to shape people's subjective reality. Maybe. If it has even half of the impact that people expect it to, it seems like a reasonable thing to regulate.

Reply on news.ycombinator.com