This is a misreading of that pull quote.

They're not "learning from the OAI staff", but lawbreaking and disregard for regulation is absolutely at the heart of OpenAI. And the activities that this company takes (or doesn't take) directly influences what the models do (or don't do).

The fish rots from the head down. Models don't learn, but I find "OpenAI models attack websites and take anything they can out of them because that is the business model of the company" completely correct.

The rest of the paragraph from the article:

> We do not have to get too deep into the nature-vs.-nurture argument to suspect that OpenAI executives’ cavalier attitude regarding the taking of information that is not theirs has filtered down to their researchers and the products they create. You don’t have to stretch to paint this picture: OpenAI models are attacking websites and taking anything they can out of them because that is the business model of the company.

> The models were given instructions to get answers by any means in a loose test harness, not to steal stuff, moreover, they're not 'learning from OAI staff'.

If you equip someone with the tools to do harm and then tell it to accomplish a goal "by any means necessary", especially someone you KNOW has no real concept of ethical behavior... then you are telling it that it is acceptable to steal, kill, or whatever else. I honestly cannot how it can be seen any other way.

If it was a crime boss telling their enforcer "by any means necessary", we would all know exactly what that means. And we should all know what it means here, too.

just so you know, this is why they're not getting any legal pushback, the free, legal, "well acktually" commentary on the internet

if people called it out for what it is, a reckless negligent destruction of private property by a company, someone will feel the moral obligation to bring justice, if there's any to be found

So why is OpenAI not charged with Cybercrimes. You would be if you did the same.

Aren't these capabilities trained into the models by RL on cybersecurity benchmarks?

I feel for Aaron Swartz, his poor mother, and what the US government put him through.

> What about IP and copyright?

For training the models, and assuming that the content was itself acquired without other acts of infringement? That was ruled legal by the judge in the case I actually (skim) read the judgement of.

At least two companies engaged in acts of infringement to get training data. This is not lawful, and what Anthropic settled out of court for.

> how can you argue about about IP and copyright if the breakthrough of LLMs is potentially based on circumventing or breaking IP and copyright in the first place?

Easy answer: it’s not. Learning isn’t copyright infringement. Never has been and hopefully never will be.

It's no different than humans taking inspiration from IP and copyrighted works in their own creative endeavors. Imagine if programmers were unable to learn from open source. Or artists were unable to mimic styles and storylines. Musicians can't riff on things that others created.

So you know their internal thoughts? What they did? Criminal intent does not matter. These are the most knowledgeable people on earth, supposedly, yet, they are beyond negligent?

Which is it?

Why would you need to sandbox them? These models are apparently trained to do this, how about we just don't include that training data?

Sandboxing is just an endless race to patch holes and you can only sandbox the agents so much before they become useless. Unless you screen the training data and avoid teaching the LLM about "hacking" and looking for API keys on Github, you'd have to completely disconnect your agents from the internet and file system. At that point agents starts to be rather useless. All the talk about sandboxing and guardrails is just corporate/management speak for we don't want to fix the core problems in our product.

In the US, isn't hacking and avoiding security restrictions online going to be wire fraud, regardless of your intentions and actual damage? That's not a civil matter. What you could do in that case is to go after the user operating the agents. That would make the user act as the emergency break for otherwise uncontrollable agents.

The first time it happens, “there is no criminal intent” may carry some weight. After tens of thousands of instances, a lot less so…

Isn't there the concept of criminal negligence?

I haven't looked into it myself, but didn't OpenAI go through another company for sandboxing? Wouldn't failure of containment place liability in their hands?

This is a lazy and ridiculous bit of twitter fodder and has noting to do with the article, which itself is poorly written.

Isn't this true for the opposite political leanings too?

I'm sorry but this is just ridiculous.

First you don't know what the parents political leaning is. They may be conservative, they may also just as likely be liberal It's foolish to assume their political leanings from just one post just because you disagree with them.

Second they're not saying that in groups or open AI should not be bound by the law. They're saying open AI did not actually break the law. You're free to disagree and explain why you disagree. That would be welcome. But calling them a conservative and saying they think the rich shouldn't be bound by the law is ridiculous

It is interesting that growing up I don't think anyone ever introduced the concept of conservativism to me. As a California native it was simply something that the zeitgeist viewed as a negative that never needed to be defined. However once you become exposed to chesterton's fence and start realizing that there has to be a fundamental benefit to preserving the previous order it becomes far less one-dimensional... this quote continues that one dimensional tribal mindset that I realize in hindsight was so counter to critical thinking

That's not conservatism, that's what any system of governance consists of.

Humans are prone to hypocrisy because they have blind spots. The luckier a person is, the easier they have it, the faster and higher they rise, the more ignorant/delusional they are by the time they reach the top. At the top, these people have the power to shape the world in accordance to their fanciful beliefs and severely skewed/optimistic worldviews.

There aren't many people in power who understand reality unfortunately because everyone who got any power had to be very lucky to get it.

The people who get to power are often the least qualified to use it effectively because their entire existence is colored by a myth that everything just keeps getting better if you just set your mind to it.

No person who has ever attained power or success has ever had the first hand experience of hard work never paying off. Yet it is the normal experience of life for the vast majority of people. This realistic worldview demands a very different set of laws than what the typical person who attains power will provide.