Generative AI is a general term for any type of automated process that uses algorithms to produce, manipulate, or synthesize data, often in the form of images or human-readable text. Is called generative because AI creates something that didn’t exist before. That’s what makes it different from discriminatory AI, which draws distinctions between different types of input. To put it another way, the discriminating AI tries to answer a question like “Is this picture a drawing of a rabbit or a lion?” while the generative AI responds to prompts like “Draw me a lion and a rabbit sitting next to each other.”
This article introduces you to generative AI and its uses with popular models like ChatGPT and DALL-E. We’ll also consider the limitations of the technology, including why “too many fingers” has become a clear indicator of artificially generated art.
The rise of generative AI
Generative AI has been around for years, possibly since ELIZA, a chatbot that simulates talking to a therapist, was developed at MIT in 1966. But years of work in AI and machine learning have recently come to fruition with the release of new intelligence systems. generative AI. . You’ve almost certainly heard of ChatGPT, a text-based AI chatbot that produces remarkably human-like prose. DALL-E and Stable Diffusion have also gained attention for their ability to create vibrant, realistic images based on text cues. We often refer to these systems and others like them as Models because they represent an attempt to simulate or model some aspect of the real world based on a (sometimes very large) subset of information about it.
The result of these systems is so bizarre that many people are asking philosophical questions about the nature of consciousness and worry about the economic impact of generative AI on human jobs. But while all of these AI creations are certainly big news, there’s arguably less going on below the surface than some may assume. We’ll get to some of those general questions in a moment. First, let’s take a look at what’s going on under the hood of models like ChatGPT and DALL-E.
How does generative AI work?
Generative AI uses machine learning to process large amounts of visual or textual data, much of it pulled from the internet, and then determines which things are most likely to appear near other things. Much of the generative AI programming work goes into creating algorithms that can distinguish the “things” of interest to AI creators: words and sentences in the case of chatbots like ChatGPT, or visual elements for DALL-E. . But fundamentally, generative AI creates its output by evaluating a huge corpus of data it’s been trained on, and then responding to the prompts with something that falls within the realm of probability as determined by that corpus.
Autocomplete, when your cell phone or Gmail suggests what the rest of the word or sentence you’re typing might be, is a form of low-level generative artificial intelligence. Models like ChatGPT and DALL-E just take the idea to significantly more advanced heights.
AI Generative Model Training
The process by which models are developed to accommodate all of this data is called training. A couple of underlying techniques are in play here for different types of models. ChatGPT uses what is called a transformer (that is what you It represents). A transformer derives meaning from long strings of text to understand how different words or semantic components might be related to each other, then determines the probability that they occur in close proximity to each other. These transformers run unattended on a vast corpus of natural language text in a process called pre-workout (That is Pin ChatGPT), before being fitted by humans interacting with the model.
Another technique used to train models is what is known as generative adversarial network, Organ. In this technique, you have two competing algorithms. One is generating text or images based on probabilities derived from a large data set; the other is a discriminatory AI, which has been trained by humans to assess whether that result is real or AI-generated. The generative AI repeatedly attempts to “fool” the discriminatory AI, automatically adapting to favor successful outcomes. Once the generative AI consistently “wins” this competition, the humans fine-tune the discriminating AI and the process starts all over again.
One of the most important things to note here is that while there is human intervention in the training process, most of the learning and adaptation happens automatically. So many iterations are required for models to get to the point where they produce interesting results that automation is essential. The process is quite computationally intensive.
Is generative AI smart?
The math and coding involved in creating and training generative AI models is quite complex and well beyond the scope of this article. But if you do interact with the models that are the end result of this process, the experience can be downright weird. You can make DALL-E produce things that look like real works of art. You can have conversations with ChatGPT that feel like a conversation with another human being. Have researchers really created a thinking machine?
Chris Phipps, a former natural language processing lead at IBM who worked on Watson’s AI products, says no. He describes ChatGPT as a “very good prediction machine”.
It’s very good at predicting what humans will find coherent. It’s not always consistent (most of the time it is), but that’s not because ChatGPT “understands”. It’s just the opposite: the humans consuming the output are really good at making whatever implicit assumptions we need for the output to make sense.
Phipps, who is also a comedic actor, draws a comparison to a common improv game called Mind Meld.
Two people think of a word and then say it out loud at the same time: you could say “boot” and I would say “tree”. We came up with those words completely independently, and at first they had nothing to do with each other. The next two participants take those two words and try to think of something they have in common and say it out loud at the same time. The game continues until two participants say the same word.
Maybe two people say “woodcutter.” It looks like magic, but it’s actually that we use our human brain to reason about the input (“boot” and “tree”) and find a connection. We do the work of understanding, not the machine. There’s a lot more of that going on with ChatGPT and DALL-E than people admit. ChatGPT can write a story, but we humans work hard to make it make sense.
Testing the limits of computer intelligence
Certain cues that we can give to these AI models will make Phipps’ point quite apparent. For example, consider the riddle “Which weighs more, a pound of lead or a pound of feathers?” The answer, of course, is that they weigh the same (one pound), even though our instincts or common sense tell us that the feathers are lighter.
ChatGPT will answer this riddle correctly, and you can assume it does because it’s a coldly logical computer that doesn’t have any “common sense” to trip it up. But that’s not what’s going on under the hood. ChatGPT is not logically reasoning the answer; you are simply generating an output based on your predictions of what should follow a question about a pound of feathers and a pound of lead. Since your training set includes a bunch of text explaining the puzzle, you assemble a version of that correct answer. But if you ask ChatGPT if two pounds of feathers are heavier than a pound of lead, you’ll confidently tell you they weigh the same amount, because that’s still the most likely outcome of a feather and lead advisory, based on your training set. It can be fun to tell the AI that it’s wrong and watch it wobble in response; I got it to apologize for his mistake and then suggest that two pounds of feathers weigh four times more than a pound of lead.
Be First to Comment