Item — HN Reader

>Chris Lehane, OpenAI’s vice president of global affairs, said in an interview that the US AI Safety Institute – a key government group focused on AI – could act as the main point of contact between the federal government and the private sector. If companies work with the group voluntarily to review models, the government could provide them “with liability protections including preemption from state based regulations that focus on frontier model security,” according to the proposal.

Given OpenAI's history and relationship with the "AI safety" movement, I wouldn't be surprised to find out later that they also lobbied for the same proposed state-level regulations they're seeking relief from.

8 comments

> ask for regulation then ask for exempt

That's exactly what has been happening:

Ask HN: Why is OpenAI pushing for regulation so much - 2023

https://news.ycombinator.com/item?id=36045397

OpenAI lobbied for restrictive rules, and now they want an "out" but only for themselves. Absolute naked regulatory capture.

I believe with regulatory capture the companies that pushed for the regulation in the first place at least comply with it (and hopefully the regulation is not worthless). This behaviour by ClosedAI is even worse: push for the regulation, then push for the exemption.

Also textbook Fascism.

Trump should have a Most Favored Corporate status, each corporation in a vertical can compete for favor and the one that does gets to be "teacher's pet" when it comes to exemptions, contracts, trade deals, priority in resource right access, etc.

nice to see HN is no longer glazing Sam Altman

And it probably would have worked if David Sacks wasn't the AI czar. The Harris administration would probably have caved by now

Just like when people complain about OpenAI's ill practices then they use it the most

They've no moat so I don't see them surviving without a gov't bail out like this.

OpenAI has a gigantic moat.

No moat means Joe Anybody can compete with them. You just need billions in capital, a zillion GPUs, thousands of hyper skilled employees. You need to somehow get the attention of tens of millions of consumers (and then pull them away from the competition, ha).

Sure.

The same premise was endlessly floated about eg Uber and Google having no moats (Google was going to be killed by open search, Baidu, magic, whatever). These things are said by people that don't understand the comically vast cost of big infrastructure, branding, consumer lock-in (or consumer behavior in general), market momentum, the difficulty of raising enormous sums of capital, and so on.

Oh wait the skeptics say: what about DeepSeek. To scale and support that you're going to need what I described. What's the plan for supporting 100 million subscribers globally with a beast of an LLM that wants all the resources you can muster? Yeah, that's what I thought. Oh but wait, everyone is going to run a datacenter out of their home and operate their own local LLM, uhh nope. It's overwhelmingly staying in the cloud and it's going to cost far over a trillion dollars to power it globally over the next 20 years.

OpenAI has the same kind of moat that Google has, although their brand/reach/size obviously isn't on par at this point.

LLM usage is still gaining traction. OpenAI may not be on top anymore, but they still have useful services, and they aren’t going under anytime soon.

And time spent dealing with laws and regulations may decrease efficiency, leading to increased power consumption, resulting in greater water usage in datacenters for cooling and more greenhouse gas emissions.

Controlling demand for services is something that could stop this, but it’s technological progress, which could enable solutions for global warming, hunger, and disease.

It’s a locomotive out-of-control. Prayer is the answer I’d think of.

It's a common tactic in new fields. Fusion, AI, you name it are all actively lobbying to get new regulation because they are "different", and the individual companies want to ensure that it's them that sets the tone.

Looks the same as taking "rebate for green energy" and then asking to "stop such rebates" a few years later

Regulatory capture is a common strategy for synthetic monopolistic competitive firms, and suckers high on their own ego.

Deepseek already proved regulation will not be effective at maintaining a market lead. =3

Why won't it?

If you get fined millions of dollars (for copyright, of course) if you're found to have anything resembling DeepSeek on your machine - no company in the US is going to run it.

The personal market is going to be much smaller than the enterprise market.

>if you're found to have anything resembling DeepSeek on your machine - no company in the US is going to run it.

That would be as successful as fighting internet piracy.

Not to mention that you could outsource the AI stuff to servers sitting in Mexico or something.

That would give an advantage to foreign companies. The EU tried that and while that doesn't destroy your tech dominance overnight, it gradually chips from it.

Great another market force to widdle away the US' economic power, so obviously trump/musk will pass this immediately

The artificial token commodity can now be functionally replicated on a per location basis on $40k in hardware (far lower cost than nvidia hardware.)

Copyright licensing is just a detail corporations are well experienced dealing with in a commercial setting, and note some gov organizations are already exempt from copyright laws. However, people likely just won't host in countries with silly policies.

Best regards =3

Regulatory moat and copyright relief for me, but not for thee.

Problem is they built the moat before moving into the castle.

Moats are not a problem if your liege lord teleports in and lowers the drawbridge for you.

no need for teleportation. just climb the walls. the castle is not protected, and has no pots of oil or flaming arrows yet.

Moat is an Orwellian word and we should reject words that contain a conceptual metaphor that is convenient for abusing power.

"Building a moat" frames anti-competitive behavior as a defense rather than an assault on the free market by implying that monopolistic behavior is a survival strategy rather than an attempt to dominate the market and coerce customers.

"We need to build a moat" is much more agreeable to tell employees than "we need to be more anti-competitive."

It is pretty obvious that every use of that word is to communicate a stance that is allergic to free markets.

A moat by definition has such a large strategic asymmetry that one cannot cross it without a very high chance of death. A functioning SEC and FTC as well as CFPB https://en.wikipedia.org/wiki/Consumer_Financial_Protection_... are necessary for efficient markets.

Now might be the time to rollout consumer club cards that are adversarial in nature.

A "moat" is a fine business term for what it relates to, and most moats are innocuous:

* The secret formula for Coke

* ASML's technology

* The "Gucci" brand

* Apple's network effects

These are genuine competitive advantages in the market. Regulatory moats and other similar things are an assault on the free market. Moats in general are not.

> "Building a moat" frames anti-competitive behavior as a defense

This is a drastic take, I think to most of us in the industry "moat" simply means whatever difficult-to-replicate competitive advantage that a firm has invested heavily in.

Regulatory capture and graft aren't moats, they're plain old corrupt business practices.

DeepSeek really shook them to their core. Now they go for regulatory capture. Such a huge disappointment. Open source AI will win: https://medium.com/thoughts-on-machine-learning/the-laymans-...

It's not just them. Everyone is scrambling.

US tech, and western tech in general, is very culturally - and by this I mean in the type of coding people have done - homogeneous.

The deep seek papers published over the last two weeks are the biggest thing to happen in IA since GPT3 came out. But unless you understand distributed file systems, networking, low level linear algebra, and half a dozen other fields at least tangentially then you'd have not realized they are anything important at all.

Meanwhile I'm going through the interview process for a tier 1 US AI lab and I'm having to take a test about circles and squares, then write a compsci 101 red/black tree search algorithm while talking to an AI, being told not to use AI at the same time. This is with an internal reference being keen for me to be on board. At this point I'm honestly wondering if they aren't just using the interview process to generate high quality validation data for free.

幸运的是，通过转换器模型，当我们光荣的领导人习近平从资本主义走狗手中解放我们时，我不需要学习中文。

100%. Western tech needs the competition. They are very prone to navel-gazing simply because SV ended up being the location for tech once.

Funny how they like to crow about free markets, while also running to daddy government when their position is threatened.

have you considered starting / joining a startup instead ?

> At this point I'm honestly wondering if they aren't just using the interview process to generate high quality validation data for free.

Not sure if that is accurate, but one of the reasons why DeepSeek R1 performs so well in certain areas is thought to be access to China's Gaokao (university entrance exam) data.

Bottom is about to drop out thats why, ethics are out the window already and its gonna be worse as they claw to stay relevant.

Its a niche product that tried to go mainstream and the general public doesn't want it, just look at iPhone 16 sales and Windows 11, everyone is happier with the last version without AI.

They were always going for regulatory capture. I think deepseek shook them but I don't think we should rewrite the history as them being virtuous only until 2024.

Has OpenAI hired McKinsey yet?

I'm unsure if you can layoff AI

ai can.

unnecessary. mckinsey uses ai from openai.

embrace. extend. extinguish.

infiltrate. assimilate.

done, tovarisch ...

https://en.m.wikipedia.org/wiki/Tovarishch

As it is, this is a bullshit document, which I'm sure their lobbyists know; OSTP is authorized to "serve as a source of scientific and technological analysis and judgment for the President with respect to major policies, plans, and programs of the Federal Government," and has no statutory authority to regulate anything, let alone preempt state law. In the absence of any explicit Congressional legislation to serve to federally preempt state regulation of AI, there's nothing the White House can do. (In fact, other than export controls and a couple of Defense Production Act wishlist items, everything in their "proposal" is out of the Executive's hands and the ambit of Congress.)

You mean there's nothing the White House can do under the rule of law. There's plenty the White House can do under the color of law.

I heard something today and I wonder if someone can nitpick it.

If what the admin is doing is illegal, then a court stops it, and they appeal and win, then it wasn't illegal. If they appeal all the way up and lose, then they can't do it.

So what exactly is the problem?

Mind you, I am asking for nits, this isn't my idea. I don't think "the administration will ignore the supreme court" is a good nit.

The fact that Chris Lehane is the one involved in this should tell you all you need to know about how on the level all this is.

For those of us who don’t recognize him by name, can you spell it out a little more clearly please?

Heavy hitter lawyer, PR expert. Some google terms: Masters of disaster, Spin cycle.

Just learning about that guy and reading his Wikipedia page will give me nightmares for the years to come.

Wouldn't be shocking if that were the case. Big companies often play both sides

>In a 15-page set of policy suggestions released on Thursday, the ChatGPT maker argued that the hundreds of AI-related bills currently pending across the US risk undercutting America’s technological progress at a time when it faces renewed competition from China. OpenAI said the administration should consider providing some relief for AI companies big and small from state rules – if and when enacted – in exchange for voluntary access to models.

The article seems to indicate they want all AI companies to get relief from these laws though.