While ChatGPT is incredibly clever, it's not intelligent in the way humans are; it won't have original thoughts and isn't a finisher. This leaves many of us wondering, what's the actual point?
Okay, it's a digital wordsmith that can write your emails, debate those awkward questions, and even (sort of) write poetry. But what is it, and how should we be using it? And what about its equally perplexing pals, the LLMs?
What Have LLMs Ever Done For Us?
Today's hyped-up AI revolution is essentially a rebranding of good old Machine Learning, or ML. As the name suggests, ML learns from massive datasets, finds patterns, and uses what it learns from one task to inform the next. LLMs are just a language-focused subset of ML.
Think of LLMs as those over-caffeinated librarians who've read everything. They've devoured mountains of text, from Shakespeare to Reddit threads, to understand the chaotic symphony of human language. They know everything, can quote anything, but how much do they actually understand? ChatGPT is just one of the more chatty librarians in this vast, digital library.
What LLMs have done brilliantly however, is revolutionise our ability to communicate with computers. Computers are miraculous, but most of us Normal People don't speak their language (code), so we've been unable to truly get them to work for us. LLMs change all that. Now, we can have normal language conversations with computers to tell them what we want. That's the revolution!
Transformers and Training
How do they do it? It's probably not important, but it's all about a clever gizmo (a "word-relationship learning engine") called a "transformer." This lets the model focus on the context of words. It allows the LLM to consider words within the context of the entire sentence it's building.
Then comes the "Training" part. Picture this: you feed a baby AI a zillion books and ask it to guess the next word in each sentence. It makes a lot of hilarious mistakes at first ("The cat sat on the... toaster?"), but over time, through trial and error (Machine Learning!), it gets the hang of it. This is how LLMs learn to predict, generate, and "understand" language, or at least predict the next word with remarkable accuracy.
A Menagerie of LLMs (Large Language Models)
LLMs have actually been around for a while, but since the public launch of ChatGPT in November 2022, all the major players have been scrambling to bring their versions online. Now we've got a whole zoo of LLMs, and they're definitely not created equal:
- GPTs (like ChatGPT)
The smooth talkers, great at generating coherent text. They're the life of the digital party. - Gemini and GPT-4o
The multi-talented ones. They speak, see images, hear sounds, and are quickly becoming all-in-one digital assistants. - LLaMA and Mistral
The open-source rebels. They're all about community and customisation, like a DIY punk band of AI. - Claude
Emphasis on safety and helpfulness, strong performance in enterprise applications, and very strong context windows. - DALL-E 3
Exceptional ability to interpret complex text prompts and produce high-quality, detailed, and creative images.
The differences? Think of it like cars. Some are fast (inference speed), some can carry a lot of luggage (context window), and some are better at off-roading (specialised tasks) or even image and video generation.
What Are LLMs Good At?
LLMs are popping up everywhere:
- Chatbots
They're revolutionising customer service, though they still make mistakes. The better LLM applications can work off secure datasets (RAGs), meaning they can be constrained to restricted information and don't risk releasing sensitive information. - Content Creation
They can write articles, scripts, and even marketing copy. Just remember, they're great at sounding smart, but not always being smart. Image generation is improving daily, and video generation is quickly catching up. - Code
They're helping programmers write code faster, which is great and efficient, but like everything else, they're replicating or merging previous code. They'll never generate anything genuinely original. - Translation
Breaking down language barriers, though they still get tripped up by idioms ("It's raining cats and dogs" becomes "Cats and dogs are falling from the sky"). But like everything else, they only need to learn the correct answer once!
What Are LLMs Not So Good At?
It's not all sunshine and rainbows. LLMs, like their creators, have their quirks:
- Bias
They can inherit the biases of their training data, leading to awkward (and harmful) outputs. - Hallucinations
They sometimes make stuff up, confidently presenting fiction as fact. Think of them as that friend with wild stories you're never sure are true. - Privacy
They can accidentally spill sensitive data, which is less "oops" and more "lawsuit." This is why the RAG (Retrieval Augmentation Generator) was developed, to enhance LLM responses by retrieving relevant information from secure knowledge sources.
In summary, the responses are brilliantly researched and delivered in seconds, but with so much information, they're liable to make mistakes, even with perfect prompts.
At Push, we teach the 10:20:70 rule: spend 10% of your time writing the perfect prompt, 20% checking and improving the answer, and save 70% of your time.
So, humans still have a role to play!
Google, Gemini, and the Future of Search
Google dominates search results, driving sales and opinions. You may have noticed a new box at the top of your Google results, "AI Overview." This is Gemini scanning websites and consolidating your answer on the results page (zero-click search). It's early days, but this could profoundly shift internet usage.
The Future: AI's Wild Ride
AI is evolving at warp speed, and businesses must adapt to stay ahead in the AI marketing revolution. They can understand and generate multiple forms of media, adapt to your style, and maybe even work in AI teams. Who knows, maybe one day they'll understand Monty Python, but that obviously won’t be an American LLM!
Large Language Models like ChatGPT are powerful tools that are changing how we interact with technology. While they're incredibly capable, they also have limitations—like biases and hallucinations—that require careful handling. As AI continues to advance, it's important to understand both the potential and the pitfalls of LLMs.
By embracing these tools with a critical eye, we can harness their power to enhance our productivity, creativity, and communication. Remember, the key to getting the most out of LLMs is to use them as collaborators, not replacements, for human insight and judgment. With this balanced approach, we can unlock the full potential of AI to improve our lives without losing what makes us uniquely human.
