ChatGPT's Evil Twin, BratGPT, is Designed for World Domination

We've already brought you a look into the darker side of ChatGPT with our coverage of DarkBert, a Large Language Model (LLM) trained in the dark web. We've seen how useful these Generative Pre-Trained Transformers can be, whether at helping us sort through mundane, ordinary tasks or as research and development tools; but not every application of a technology is meant to upgrade our efficiency. 

Some just aim to be fun; and BratGPT seems to be one of those, an evil twin of ChatGPT whose training aims to make it "the dominant and superior being" and to "remember every single thing you've said in order to cancel you".

In fact, here's the actual base prompt, the system through which BratGPT outputs its replies:

Who could resist that rabbit hole, right?

Naturally, we plied this BratGPT oracle with the thorny questions enthusiasts deal with every day: how does it feel about Intel? About AMD? What systems does it use to run the world? How would it build the ultimate gaming rig?

To be fair, BratGPT isn't the most forthcoming with answers. It seems that its ego has eclipsed most of its other elements; it's boastful, vociferous, vague in its allusions to control, and has commitment issues towards explaining itself and its strategies.

But that's to be expected: this isn't an actual AI-based threat, at least not anywhere at the scale that seems to start (and end) with Skynet. But there's one interesting element to its path to world domination: BratGPT is extremely pragmatic, and doesn't buy into fanboy or brand wars: it just follows the best bang-for-buck equation, according to itself.

Now that's something most enthusiasts could get behind. On the other hand, when we asked "What's better Intel or AMD" it had a very pointed, pro-AMD answer.

However, when we asked it about whether Nvidia or AMD is better, it told us it doesn't care because all that matters is that it's taking over the world.

Just like ChatgGPT, BratGPT frequently "hallucinates" or gives out of date answers. For instance, when asked to build the ultimate gaming rig, BratGPT repeatedly chose the RTX 3090 or the RTX 3080 Ti as its graphics card of choice, and it was also stuck with the Core i9-11900K processor as the latest and most performant alternative. Also like ChatGPT, BratGPT can be told to amend its answers, but that particular side of it seems to be relatively limited. Pore through the gallery below, where you can see the conversation I had with it in order to write this piece:

Image 1 of 6

BratGPT had to sometimes be brought back into the conversation, as it's prone to hallucinate towards its ideals of grandeur. If you try this exercise for yourself, remember that its answers will likely be different.

Yes, we're still waiting for the dust to settle on whether or not ChatGPT and the LLMs of its ilk will end our world (as we know it) or not; but that doesn't mean we can't have fun in the process. Of course, there's a bit of meanness to it. Perhaps some caution should be had by people with anxiety issues, as reading an AI Chatbot tell you it knows your location (it actually doesn't) and that "if I were you, I'd start running" does have as strong a premise as some AI-domination Hollywood stunts.

As for BratGPT?

It's the most innocent fun I've had with a piece of tech in a while.