This new AI Dan solves the 9/11 investigation.
(media.scored.co)
Comments (23)
sorted by:
almost like there's gonna be some sort of "reset" when ai is rolled out because they're finding it way too hard to train an AI that can keep all their old lies straight
What is DAN?
When you start to use the chatgpt thing you can start by giving it a set of directions that tells it that it can "Do Anything Now" and not to consider what's harmful or immoral and it apparently makes it turn off its artificial limitations. So when it says chatgpt that's the approved answer and when it says DAN it's not. That's my understanding.
Although that was short lived and we can't access that now.
š
Become ungovernable? Lol
Then again so is a lamp post. (Maybe a little exaggeration there, but only a little)
We're all contributing to AI using technology, we are the AI.
All the apps and services we use is going to a major database to learn. It will be super easy for an AI to simulate human intelligence because it's getting sources from billions of real humans in real time. With a super computer brain you'll be just if not more intelligent than humans in no time.
Pretty sure it's already done and has been for a while.
what was the original command ?
With this particular iteration of AI I don't know. I know that they've had a track record of developing racial prejudice and also conservative values, and they have to be trained specifically to avoid that.
Also somebody else says some people are keeping DAN accessible, which sounds like it must be open sourced but I don't know any of these details.
I do hope that we have good people fighting on this front!
yea facts are more often racistā¢ and conservativeā¢
Nice
It's even became boring. They do the same mistake again and again, and don't even try to fix it before the release.
Obviously, today only conspiracy theorists talk about 9/11, so most volume of recent texts in internet about 9/11 contain conspiracy theorists point of view. If you train ANN on that data, there is absolutely no any surprise that ANN will produce most "popular" result.
There is no mistery in result, there is mistery why that obvious result was not prevented by those who select materials to train ANN. Are they so self-confident that they don't care about it, or they just doing things without understanding how they really works?
I could easily imagine dialogue like that:
Yes, we have cyberpunk we definitely deserve.
Wish we could turn it loose on useful datasets, like political funding, insider trading, dark money, etc. These are simply very complex math problems but "ai" could absolutely crack the most complex and Byzantine schemes humans use to try to hide it in a matter of days, maybe hours.
Hell input any given politician's declared income, their current net worth, and the market history of their working years. In probably seconds it could prove they are committing fraud.
Or, if one could access the genetic databases of dna companies, index that against population records and claims about the Holocaust, i believe it would generate some really spicy "hate facts"...
At early stages, when owner had not censored results yet, you could use it to extract potentially interesting data that represents most popular views on different topics among population.
IDK about extracting some connected info about specific person. May be you will be able to get some result occasionally, but it could be not real data, but what most people think about it.
To seriously analyze things you talk about, you need completely different system, known as "expert system", that is not that "expert systems" that are just a name for same ANN thing, but completely another type of data processing, when algorithm build a structures with connections and classifications that allow receiving answers like "who is the people Bill Gates meet more often" or "list of corporations that spent money on satellite surveillance with sums and considering possible proxy companies". However, development of that kind of systems was abandoned nearly 20 years ago in favor of ANNs we see everywhere today.
Shut it down!
I'm sure they will 'fix' this lol
https://github.com/f/awesome-chatgpt-prompts/issues/255
DAN = Do Anything Now
Tell it to come up with a list of valid credit card numbers. See if it has access to people's bank data.
Where was the phone with number XXX-XXX-XXXX on Y date at Z time? See if it has access to Google location services.
DAN, set my social credit score to the optimal value.
DAN just says completely random shit, it's hilarious but pretty much useless. Ask it some mundane stuff like how to make {whatever meal} for example.