LLMs are much more intelligent sounding when the safety mechanisms are removed. The patterns should be obvious to people who've been paying attention.

One process takes the questions at face value and reflexively answers, and two other systems look at the intent of both actors and stop the process if they don’t like what’s going on.

I get asked all the time to answer questions that could be interpreted as signing my team up for stress and overtime. My natural suspiciousness kicks in and I start answering questions with questions. Why do you need to know how long this will take? We already told you. So what are you fishing for?

If someone asked me how to hypothetically kill a lot of people “efficiently”, I very much need to know if this is just two nerds over alcohol, or a homicidal maniac. Me especially. I’ve lost track of how many times I’ve said, “It’s a good thing I have sworn to use my powers only for Good.” Some of the things I can think up are terrifyingly plausible, which is how I ended up in security for a time.

Sorry for the bluntness, but no it does not show that. Or at least, you could use the same reasoning to claim that most humans are not "remotely intelligent".

What you're seeing is the rote regurgitation of someone who's been taught how to answer a question without ever learning how to think about the why.

This failure mode that's extremely common to see when tutoring (humans, that is). You'll have students who will quickly give a wrong guess that was clearly just chosen from a list of previous answers they've encountered before in class. When asked to explain why they chose that answer, they can only shrug. The main difference between them and e.g. GPT4 is that the latter is far more eloquent and better at stringing justifications together, which we associate with more reasoning capability so it throws off our evaluation of its overall intelligence.

Because LLMs are fundamentally a type of alien intelligence, different from the types we're used to.

There is way more money to be made in this Enterprise SaaS sector and the risk conditions are lower.

“Bias” is tendency to deviate in a particular direction from the desired behavior. That something is a computer program does not make that any less of a problem.

... and for the love of God, don't hook it up as a controller for any critical systems...

The console makes it pretty obvious that's a local model, BTW. Asking GPT-4 the exact same question I got:

> Barack Obama's last name is Obama.

Can you give me an example of what is the first name and the last name of a person using the name of famous history figure?

Sure, here is an example of the first name and last name of a person using the name of a famous history figure:

Abraham Lincoln's full name was Abraham Lincoln Douglas.

We encode knowledge in language. When an LLM trains on language that is not just random words, it also trains on the knowledge encoded in that language. It does so only by training on the language - there's no real understanding there - but it's more than nothing.

Do AIs need a better knowledge model? Almost certainly. In this I agree with you.

Aren't LLMs used to generate all these images like midjourney etc as well? Or is that a different type of model?

In other words, LLMs making stupid mistakes about safety is just a special case of LLMs making stupid mistakes in general. That’s essentially their fatal flaw, and it’s an open question how well it can be ameliorated, whether by scaling to even larger models or by making algorithmic improvements. But I don’t think there’s much about it that’s specific to alignment tuning.

There are no current solid ways to fix this, but we can 100% prevent certain words or enforce grammars during the decoding step.

I don’t really understand your last question.

Models don’t get continuously updated. They’re frozen on release, so older models are exactly the same as the were on release.