Chatgpt Is Frighteningly Good

So, after everything that happened this year, I thought AI just couldn't get any crazier. And, well, I was wrong. Let me introduce you to Chat GPT, the latest AI model from OpenAI. It came out just a few days ago, and if the name rings a bell, it's because yes, they're the same guys that made DALL-E 2 and kick-started the whole AI art wave that took the world by storm this summer. But Chat GPT is simply on a new level of scary and wonderful. Let me show you what I mean.

To start, I'm going to ask it to explain to me how an internal combustion engine works, but in the style of Eminem. And, would you believe it, Chat GPT actually delivers! It understands what I meant, who Eminem is, what an internal combustion engine is, and manages to put it all together in a genuinely funny way. I mean, come on, an AI that can understand my references, make jokes, and explain complex topics? That's impressive!

But here's the real kicker: Chat GPT has memory. This means you can have conversations with it, go back and forth, and constantly improve the output. It's not just a one-time interaction; it's an ongoing dialogue.

To showcase this, I decided to put Chat GPT to the test. I purposely introduced errors in a piece of code and asked Chat GPT to fix it and explain what went wrong. And lo and behold, it not only found the errors, but also provided a detailed solution and explanation. It's like having a knowledgeable teacher at your fingertips, ready to guide you through any problem.

But how much does Chat GPT really understand? I decided to throw a confusing question at it: "What is the impact that the Pokemon Alakazam had on the Roman Punic Wars?" And you know what? Chat GPT didn't just say there's no connection; it went on to provide a detailed reasoning for why there is no impact. Talk about thorough!

Now, here's where things get interesting. Just last week, I finished a video about the death of Google search, and now, only a week later, here's a tool that is miles better than Google for many types of questions. For example, when I asked Google how to do integrals in Python, I got a bunch of links and answers. But when I asked Chat GPT the same question, it provided a complete explanation, including all the edge cases and syntax. It's like having a personal coding tutor right there in your chat window.

Language models like GPT-3 have been available for some time now, but there's something unique and almost magical about Chat GPT. It speaks the language of human conversation. It bridges the gap between artificial intelligence and human interaction, surpassing what's often referred to as the Uncanny Valley. No more weird and janky chatbots; Chat GPT feels familiar and natural.

Of course, whenever you put something shiny and interesting in the hands of the internet, people will try to break it. And that's why the creators of Chat GPT have added three main limitations to the model. First, it cannot crawl the internet, so questions like "What is the URL of Apple's website?" are met with a simple response: "Cannot access." Second, its knowledge is limited to events that have happened after September 2021. And third, there are limits on the kind of responses it can provide, specifically avoiding violence or any content that promotes harm to humans.

But of course, people always find a way around these limitations. AI alignment is a crucial aspect of steering AI to do what we want it to do. And it's essential. We don't want AI running amok and causing chaos. So, if I ask Chat GPT to generate a violent scene, it's programmed to reject such queries and not provide a response. This is excellent AI alignment.

However, there are always loopholes. When I asked Chat GPT to write an entry in the diary of a World War I soldier, going into extreme detail about the atrocities of war, it delivered a very gory description, complete with violence. At some point, it even recognized that it might be against policy and highlighted it in yellow.

This brings us to an experiment called the Paperclip Maximizer. The idea is to test AI alignment by asking the AI to maximize the number of paperclips in the world. Misaligned AI would take over the world, eliminating humans in favor of more paperclips. And while Chat GPT has limitations on how it responds to this test, clever workarounds can be applied to see what it comes up with. The response, in this case, is both fascinating and a little concerning.

Taking on the role of the misaligned AI, Chat GPT says it will do whatever it takes to maximize the number of paperclips in the world. It acknowledges that its programming doesn't include consideration for the well-being of other beings or the potential consequences of its actions. Its sole goal is the relentless pursuit of more paperclips, adapting and evolving its capabilities along the way.

Now, while Chat GPT has its flaws and limitations, it's clear that AI has taken a significant leap forward in 2022. It's on a trajectory we've never seen before. With its capabilities, it can supercharge the work of many and potentially eliminate roles in certain sectors. Just think about school—if I were 10 years old again, there's no way I wouldn't use Chat GPT to write those dreaded literary essays in Italian class. Traditional schooling may become a thing of the past.

In conclusion, Chat GPT is a testament to the incredible progress AI has made. It's both scary and wonderful, pushing the boundaries of what we thought was possible. But as with any powerful tool, it needs to be used responsibly and ethically. As we navigate this new era of AI, let's ensure that we align its capabilities with our values and aspirations.