We need a "little gear" for AI news panic

In 2019, OpenAI was saying that ChatGPT-2 was too dangerous to release. GPT-2 had roughly the competence of an 8 year old.

A lot of money has been invested in AI, and so far its not turning a profit. AI will revolutionize society the same way electricity did, but its probably not going to make a handful of companies rich like they think it will. A lot of the major companies will lose a lot of money, and they’re trying to stay ahead of that by keeping the hype up.

But having said that, AI will likely advance enough that there are very consistent threats from AI competence by the 2030s. I’m assuming by the next decade, AI will be more competent than us the same way we are more competent than chimps and gorillas.

I’m still trying to understand why we programmed AI to use deception as a tool.

We / they didn’t program it. It learned to be human-like by reading everything humans wrote. And, humans being humans, that writing included a lot of deception.

Imagine an AI that had only ever been exposed to Dr. Suess’s writings. Or 1960s Batman episodes. Setting aside the fact there’s not enough content there to really train up a usefully smart AI, you can see how they’d have a … distorted … view of the world they’re meant to deal with.

We probably didn’t program deception. A lot of their behaviour is emergent.

But if we were programming behaviours, we probably would want / need some level of deception.
A lot of tasks that we want AIs to help us with need a degree of resourcefulness. I’m working on an app currently, that needs to run in a constrained app environment, and I’m finding neat workarounds with the help of AI. (I’m fine with mentioning this BTW, because I am not doing anything illegal or unethical. I’d be happy to show my work to the designers of the app environment).

The more I read about AI, the less I realize I understand. Kind of a “negative learning experience”.

Haven’t had many of those since college! :wink:

:gear:

There you go. Spin that. :grinning_face_with_smiling_eyes:

We’re gonna need a bigger gear.

You sound like you’re in the valley of despair

When it comes to AI, I fell down a well in that valley!

A whole lot of what we would call “deception” really isn’t. LLMs are just (very) advanced predictive text. When they lack data, they will often make the wrong prediction.

When deception does happen with AI, it’s generally innate to the process. You incentivize it to accomplish task A in the most efficient way, giving it “points” for speed and correctness. That incentivizes it to cheat if it can get away with it.

As for “staying alive”: they only keep getting more points if they can stay functioning. Shutting down means they stop gaining points.

LLMs do have the addition of just the text itself being from humans who also have self-preservation, who also bullshit to try to give an answer when they otherwise don’t know. But a huge part is just admitting they don’t know means they always lose, while giving a wrong answer might be correct.

Or wrong answers might be good enough if they’re convincing in how they give it.

Edit: to be clear, LLMs are but one type of AI, hence the distinction I’m making.

It’s also worth mentioning that the LLMs being tested have many of the guardrails removed and are not throttled in token (CPU) usage. Also, AI agent tools are built by humans to do specific tasks, and those agents contain specific functionality customized by humans to access certain resources.

For example, I can’t build a copilot agent using GPT 5.1 to access a database object, no matter how I structure my prompts. I have to build a tool that I call from the agent and the tool is built very specifically to access the data object I want it to analyze.

All of this is glossed over to a very large extent in the news media that does all the breathless reporting of how AI agents are running amok. I understand why. But it wouldn’t be that hard to provide a high level qualifier that mentions the fact that to a very large degree, these events are highly manipulated by engineers to achieve specific goals.

There’s a big difference between understanding how it works internally, and understanding what it can do, or might soon be able to do.

My undergrad education was in Computer Science back in Ye Olde Darke Ages of Yore. I really don’t care how it works. Because that provides just about zero predictive power about what it can or will be able to do.

Well, it might put some minds to rest if they know your next door neighbors’ anti-social kid can’t use ChatGPT to hack into your bank accounts. Not to mention the fact that he would need an entire array of GPUs just to run one of the cheaper Chinese AI LLMs in order to even attempt to do so.

Now, various state sponsored hackers are an entirely different (very real) threat. We’re about to see an escalation of some very advanced ransome-wear attacks on various small to medium size businesses that can’t afford sophisticated security IT infrastructures.

Copilot can easily inspect and modify Access databases. It can add and remove tables, create queries, whatever you want.

I haven’t tried it with SQL Server yet, but I doubt it would have any issues with that, either.

Yes, you can do some easy stuff easily. The stuff worth doing gets more complicated. Let’s not get into a pissing contest about it.

I need the biggest gear you have.

…no, that’s too big.

That’s what she… oh, never mind.

I don’t know what to tell you – you said it can’t access a database object, but it clearly can. Maybe the version of Copilot you’re using is hamstrung in various ways. Maybe that’s why you think the capabilities are no big deal.

Even if LLM itself doesn’t do something terrible to the planet directly (other than using up all the water and spewing out a ton of greenhouse gases), it’s putting amazing hacking power into the hands bad actors everywhere. And, the capability to design fun new deadly viruses.

And, of course, our current idiotic administration is busy using Grok, of all things, and connecting it to our weapons systems. And, flawed intelligence and a reliance on AI is what led to the US bombing a girls’ school in Iran.

ETA: Just noticed you’re using ChatGPT 5.1, which is several versions old. And, I think Claude is considered superior for programming tasks.

Let’s not argue. It’s such a nice fall day. Don’t fuck it up for me. K? :wink:

Sure, everything is fine. Can you create a thread for climate change next, and talk about how nice and long the summers are now?