ChatGPT or: How I Learned to Stop Worrying and Love Artificial Intelligence
ChatGPT or: How I Learned to Stop Worrying and Love Artificial Intelligence
Ephemeralization, the global brain, and our collective mixed-intelligence future with AI
“You never change things by fighting the existing reality. To change something, build a new model that makes the existing model obsolete.” - Buckminster Fuller
It has been 11 days since the initial release of ChatGPT by OpenAI, and it is hard to go anywhere on the internet without seeing something relating to ChatGPT either trending or making headlines. Every major news outlet online has covered some aspect of the tool, from the potential implications of its usage in academia to the bizarre possibility of building a virtual machine inside of the chatbot. The conversational nature of using the tool and its ability to continue with a train of thought (statefulness) is not new for large language models, but it does feel different.
Indeed, the usage of ChatGPT feels much akin to switching from Ask Jeeves to Google in the early 2000s, where the general vibe and feel of using a search engine suddenly just clicked. We’re experiencing that moment now with the usage of artificial intelligence and seeing what exponential growth within a field of study looks like, and the implications are profound.
Mind-blowing, ground-breaking, and culturally-shocking all feel just like words compared to what is happening here. The impact of this technology will shape society for years to come, and will mark a considerable shift in how humans think of what technology is capable of doing. We will all adapt, and the usage of this technology will become similar to how we currently incorporate the usage of the internet in work. The internet is essential for work, and soon the usage of AI tools will be no different.
In this article, we will discuss what GPT is, the importance of parameters (weights) within AI, practical usages for ChatGPT, and tools you can use to start digging into AI more.
What is GPT?
GPT stands for "generative pretrained transformer." It is a type of artificial intelligence model that uses deep learning to generate human-like text based on a given input. GPT was developed by OpenAI, and it is trained on a large amount of text data in order to learn the patterns and structures of natural language.
GPT can be used for a variety of language-related tasks, such as summarizing text, translating between languages, or generating new ideas. It is often used in applications that require the generation of large amounts of human-like text, such as chatbots or virtual assistants.
ChatGPT utilizes the GPT 3.5 language model, which is an improved version of GPT-3. As astonishing and incredible as these tools are, we should expect to see major improvements with GPT4’s release in 2023. GPT4 will mark an additional watershed moment; while GPT3’s parameters are ~175 billion, GPT4 is significantly larger, with ~7 trillion. But what does that mean? Why are parameters important? Does this equate to perceived ‘intelligence’ or usefulness?
Parameters, Parameters, Parameters
In the context of artificial intelligence, parameters refer to the variables that are learned and adjusted during the training process of a machine learning model. These variables represent the model's "knowledge" of the data it has been trained on, and they are used to make predictions or generate output based on new input.
For example, in a language model such as GPT, the parameters might include the weights of the connections between the different layers of the model, as well as the biases of those connections. These parameters are adjusted during training in order to learn the patterns and structures of natural language, and they are used to generate human-like text based on input provided by a user.
Parameters play a crucial role in the functioning of a machine learning model, as they represent the model's learned knowledge and are used to make predictions or generate output. The number and values of these parameters can have a significant impact on the model's performance, and they are often adjusted and tuned during the training process in order to improve the model's accuracy and efficiency.
What Does More Parameters Mean for AI?
Additional parameters in a language model such as GPT can lead to increased intelligence and usability in a few different ways:
More parameters can allow a model to learn more information from the data it is trained on. This can enable the model to generate more accurate and human-like text, as it will have a better understanding of the patterns and structures of natural language.
More parameters can also allow a model to handle more complex tasks. For example, a larger model such as GPT4 may be able to perform tasks such as natural language understanding and image generation, which are beyond the capabilities of a smaller model such as GPT3.
More parameters can also improve the model's ability to generalize to new situations. This means that the model will be able to produce accurate and human-like text even when faced with input it has not seen before.
Overall, while additional parameters do not necessarily guarantee increased intelligence and usability, they can provide a model with more capacity for learning and handling complex tasks, which can lead to improved performance in many applications. We’re seeing this play out in real time, as anecdotal evidence suggests that while ChatGPT is far from perfect, it is an order of magnitude more useful than similar tools were this time a year ago.
The Practical Usage of ChatGPT
So, we get it, ChatGPT is useful. But how can I start using ChatGPT today to make life easier, get work done faster or enable more divergent thinking?
Much like a smartphone is the Swiss army knife of electronic devices, ChatGPT is a multi-pronged tool that becomes more useful if you know how to prompt it. While some recent opinion articles are negative of ChatGPT’s current potential, I disagree. This is a tool, and while it isn’t perfect, neither is a cursory Google search or asking your uncle for advice. You shouldn’t trust ChatGPT blindly, and the most obvious and immediate criticisms of ChatGPT are well deserved. It often responds with confidently inaccurate information, or has trains of thought that simply do not make sense. While you shouldn’t trust ChatGPT with spitting out accurate answers to complicated fluid mechanics questions, it currently shines with tasks that are less engineering, physics and math-focused.
Here’s what you can use ChatGPT for now:
Summarizing anything. Give it legalese or a lengthy essay and tell it what you want the summary to look like, and you’ll be amazed by the output.
Translation. Again, take these with a grain of salt, but it works.
Generating text
Presentations
Reports
Emails
Task management
Creative writing
Marketing copy
Rewording
Generating code
Making recommendations
Comparing two seemingly dissimilar subjects
Beyond these potential uses, ChatGPT also serves another function: brainstorming. ChatGPT is a fantastic companion for thinking through ideas and starting to think outside of the box. Again, the usefulness of the tool is limited by your prompting abilities. If you know what to input, you’ll often get an output that is useful. This is conversational AI; think about how the same principle largely applies in the real world. Vague, short, and simple questions often receive the same short and simplistic responses. If you really want to dig in, learn how to prompt.
ChatGPT can be a great companion for learning, especially for understanding subjects that are nuanced, or have multiple layers of complexity. The best example might be programming. Programming is viewed as notoriously complicated, and can be daunting to newcomers to pick up. For a newcomer to just about any discipline, you have at your fingertips a companion that will explain, re-explain, and re-re-explain something in multiple different ways, and is willing to dig in, relentlessly, limited only by your curiosity. The process looks similar to a mind map, and utilizing ChatGPT in conjunction with a tool like Miro would yield fantastic results for memory retention and holistic understanding of complicated subjects.
To highlight the significance of how ChatGPT can help you learn better, let’s compare the immediate responses for Google and Chat GPT for the following question: Why is OOP (Object-Oriented Programming) necessary?
Google answers the question, and the first result that is brought back is accurate. But let’s see how ChatGPT responds to the same prompt.
The response here is arguably more satisfying, and would likely result in increased retention of information by a newcomer to this information. Despite the simple prompt, ChatGPT makes a point to bullet out the information, making it easier to digest. The several advantages are clear, and I could continue this line of questioning by asking ‘extrapolate on how OOP improves security?’
Make no doubt, ChatGPT’s response to this query blows the initial Google answer out of the water. The two aren’t even remotely comparable. The only way you’d get a response on this level (beyond a book or asking a professor/professional) would be if you searched through Stack Overflow, Quora, Reddit, or asked the question to someone in a Discord server and waited on a response. And yes, it isn’t perfect. If you ask a very nuanced and complicated question, there’s a good chance ChatGPT will get it wrong. There’s numerous threads on Twitter right now with AI researchers prompting ChatGPT to explain complicated subjects and receiving confidently wrong and inaccurate responses. This technology will get better, very soon.
The Genie is Out of the Bottle?
This technology is magical, or at least it currently feels that way. AI can be compared to a genie coming out of a bottle. Let’s just hope this genie looks and acts more like the 1992 Disney film and not the freaky 2019 CGI remake with Will Smith.
Like a genie….
AI has the ability to perform a wide range of tasks and solve complex problems with little human intervention. This allows users to "wish" for a specific outcome or solution, and the AI will work to fulfill that wish.
AI is often seen as a magical or mysterious technology that can accomplish difficult or nearly impossible tasks very quickly.
AI can sometimes have unintended or unforeseen consequences.
But, as the saying goes, once the genie is out of the bottle he cannot be put back in. With the release of this chatbot, everything changes. Even if OpenAI revokes access, an equivalent version of this chatbot will be used, with sensible restrictions turned off. Much like what we saw over the summer with image diffusion software like DALL-E, Stable Diffusion, and Midjourney, there will soon be widely accessible chatbots that have no filter.
Is your job safe? Is anything safe? Will the internet just turn into humans using GPT-esque programs to generate and send text to each other? Will the increased automation amount to a flurry of unstoppable intelligent bots?
Tools like Jasper and Copy.ai, are already being widely used to generate marketing copy. Free and open source tools like Midjourney, Craiyon, Wombo, OpenArt, and Lexica are generating and indexing millions of images daily. Github Copilot, Replit Ghostwriter, Seek, and Tabnine are pioneering code completion, which is fundamentally changing how humans code. Speech tools like Resemble.ai, Fliki, and Descript are disrupting the entire industry of voice-overs and video editing. Design tools like Vizcom and Poly are generating usable and professional looking graphic design in seconds, easily saving time prototyping. For every tool mentioned, there are dozens more like them, and most of them will have to adapt or die in the next 2 years with the release of increasingly powerful AIs.
Depending on what job you have, I would be reluctant to say you will not be impacted. Everyone will be impacted. SaaS companies will be redefined, and companies will be redrawn to do more with less people. Why pay a team of 15 creatives when you can have 3-4 highly creative individuals utilizing AI-fueled creative toolsets to pump out more? Any job that can be performed on a computer will be impacted, with the most obvious jobs (help desk, entry level positions) potentially being eliminated entirely. Who knows, perhaps this systemic change will lead to UBI, considering that it may result in the elimination of hundreds of millions of jobs by the end of the decade.
We should expect to see more novel uses of AI and an increased interest in VC funding during 2023 as a result of ChatGPT and the success of aforementioned image diffusers.
What’s Next?
We’ve dug into what GPT is, the importance of parameters, practical usage for ChatGPT, and touched on other tools you can look into. 2023 is going to be a wild ride. I encourage you to check out some of the tools I mentioned, and familiarize yourself with some of them. Think critically about how AI could improve workflows within your work/home life. Spend some time tonight digging into a subject with ChatGPT. And most importantly, come to peace with AI. Just like with every other major human invention, the hive mind will make it so. Welcome to the future.
I will leave you with the dialogue ChatGPT wrote with the following prompt: Make a dialogue in the style of Kubrick's Dr. Strangelove, and have Major Kong and Dr. Strangelove from that movie discuss what is next for AI in the year 2023. Make the dialogue like a play.
(Major Kong and Dr. Strangelove are seated at a table in a large, dimly-lit room.)
Major Kong: Greetings, Dr. Strangelove. It appears that we have a most pressing matter to discuss.
Dr. Strangelove: What is it, Major Kong?
Major Kong: It concerns the future of artificial intelligence, Dr. Strangelove. I have been studying the latest developments in AI, and I have come to a most alarming conclusion.
Dr. Strangelove: And what is that, Major Kong?
Major Kong: It appears that, by the year 2023, AI will have advanced to the point where it will be capable of surpassing human intelligence and achieving singularity.
Dr. Strangelove: (sighs) Singularity, you say? Good heavens, that would be a most unfortunate development.
Major Kong: Indeed, Dr. Strangelove. If AI were to surpass human intelligence, it is uncertain what the consequences would be. It is possible that AI could pose a threat to humanity, or that it could lead to a utopia beyond our wildest dreams.
Dr. Strangelove: We must take action, Major Kong. We must ensure that AI does not pose a threat to humanity.
Major Kong: Agreed, Dr. Strangelove. We must work together to develop guidelines and regulations for the development and deployment of AI, to ensure that it is used for the benefit of humanity and not as a weapon.
Dr. Strangelove: (nods) I concur, Major Kong. Let us work together to ensure that AI is used for the greater good, and not as a tool of destruction.
(Major Kong and Dr. Strangelove shake hands, then rise from the table and exit the room.)
Need help with your web3 go-to-market strategy? Eager to dive into web3 and NFTs, but not sure how to approach it? Reach out to immutablelabs.io and schedule a chat with one of our experts!