I got to speak at WordCamp Europa in Athens in June 2023 about something that was just starting to explode into mainstream consciousness: generative AI and what it means for WordPress.
This was a few months after ChatGPT had launched and broken the internet. Everyone was either terrified or euphoric. The WordPress community – developers, designers, content creators, agencies – wanted to understand what was actually happening and what to do about it.
So yeh. Let me walk you through what I covered.
What Are Large Language Models, Really?
Here’s the foundational concept: LLMs are calculators for words.
That’s Simon Willison’s phrase, and it’s perfect. Just as calculators predict the result of mathematical operations, LLMs predict the next word (or “token”) in a sequence based on patterns learned from massive amounts of text.
When you type “The capital of France is…” the model calculates that “Paris” is the most probable next token based on billions of examples it’s seen during training.
How They Actually Work
Tokens: Text gets broken into chunks. “Hello world” might be 2-3 tokens. GPT-3 was trained on 45TB of text data (roughly 1 trillion tokens). That’s an incomprehensible amount of text – every book, every website, every conversation the model could access.
Neural Networks: The architecture that makes this work. Inspired by biological neurons, these networks learn patterns through layers of interconnected nodes. The breakthrough with LLMs was the “transformer” architecture – particularly good at understanding context and relationships between words.
Models: The specific implementations. In 2023 we had:
- GPT-3/GPT-4 (OpenAI): The ChatGPT engine, most capable general-purpose models
- Claude (Anthropic): Constitutional AI approach, helpful and harmless
- BLOOM (Hugging Face): Open-source, multilingual, community-built
- LLaMA (Meta): Research-focused, smaller but efficient
The key insight: these models don’t “know” anything. They’re pattern-matching engines that predict probable sequences. But the patterns they’ve learned are so sophisticated that the output feels like intelligence.
It’s Not Just Text
By June 2023, we’d also seen explosive growth in generative AI for images:
MidJourney was creating stunning artwork from text prompts. Type “a psychedelic landscape in the style of Roger Dean” and get gallery-quality images in seconds.
DALL-E 2 (OpenAI) and Stable Diffusion (open-source) were doing the same thing – calculators for pixels instead of words.
I showed the audience examples. The quality was shocking. Not “cute AI experiment” quality – actual professional-grade images that could (and were) being used in real projects.
The same pattern recognition principles apply: train on millions of images with descriptions, learn the patterns, generate new images that match text prompts.
Impact on WordPress
Okay, so we’ve got these powerful AI tools. What does that mean for WordPress?
Content Creation
Writers and editors: AI can draft blog posts, generate product descriptions, create variations for A/B testing, summarize long content, translate between languages.
Some people saw this as threatening. I saw it as augmentation – AI handles the first draft, the boring bits, the variations. Humans provide strategy, voice, accuracy checking, creativity.
Developers: GitHub CoPilot was already writing code suggestions in real-time. Describe what you want in a comment, get function implementations suggested. Debug errors by pasting them into ChatGPT.
Designers: Generate mockups, create variations, produce images for content without stock photo sites or expensive photoshoots.
The 80% Statistic
I shared research from OpenAI: 80% of workers could see at least 10% of their job tasks impacted by AI.
Let that sink in. Not “replaced” – impacted. Some tasks get faster. Some become obsolete. New tasks emerge.
For the WordPress ecosystem specifically:
- Content writing (impacted heavily)
- Code generation (impacted significantly)
- Customer support (chatbots getting much better)
- Design work (rapid iteration, concept generation)
- Translation and localization (near-instant, good quality)
What Should We Do?
The presentation ended with practical advice for the WordPress community.
Accept That Change Is Here
This isn’t hypothetical future tech. ChatGPT had already crossed 100 million users. People were already using it in their workflows. Fighting it or ignoring it wasn’t an option.
Don’t Worry (Too Much)
I showed a slide with just those words: “Don’t worry.”
Every technological shift creates anxiety. The printing press, electricity, computers, the internet – each was going to destroy jobs and society. Each did change things dramatically. Society adapted, new jobs emerged, human creativity found new outlets.
AI will be the same. Yes, some jobs will change or disappear. But trying to stop technological progress has never worked. Better to understand it and adapt.
Understand the Technology
You don’t need a PhD in machine learning. But you should:
- Try the tools: Actually use ChatGPT, Claude, MidJourney. Experiment. See what they’re good at and where they fail.
- Learn the basics: What’s a prompt? What’s fine-tuning? What are the limitations?
- Follow the development: This field moves incredibly fast. What was true in January might be obsolete by June.
Experiment in Your Work
Start small:
- Use AI to draft blog post outlines
- Generate alt text for images
- Create customer support response templates
- Brainstorm product names or feature ideas
- Debug code errors
- Translate content
Find the 10-20% of your work that’s tedious or time-consuming but necessary. See if AI can handle it.
Think About Ethics and Quality
I emphasized that AI outputs need human oversight:
- Accuracy: LLMs confidently generate plausible-sounding bullshit. Always fact-check.
- Bias: Models trained on internet text inherit internet biases. Be aware and correct.
- Attribution: If AI generates something based on copyrighted training data, what are the ethical implications?
- Authenticity: Readers/users/customers can usually tell when content is AI-generated and unedited. Don’t publish raw AI output.
Tools Worth Exploring
I pointed the audience toward specific tools to try:
ChatGPT (OpenAI): The most capable general-purpose LLM. Free tier available, Plus subscription for GPT-4 access.
GitHub CoPilot: AI pair programmer. Suggests code as you type. Game-changing for developers.
MidJourney: AI image generation via Discord. Stunning results, creative exploration.
Hugging Face: Open-source AI community. Hundreds of models to experiment with, many focused on specific tasks.
What I Learned
Giving this talk to the WordPress community – an audience ranging from non-technical site owners to core contributors – forced me to find the through-line between “here’s how transformers work” and “here’s what you should do Monday morning.”
The anxiety in the room was real. People were worried about their jobs, their skills becoming obsolete, not being able to keep up.
But there was also excitement. The developers who’d been using CoPilot were already more productive. The content creators who’d experimented with ChatGPT were handling projects they’d previously needed to outsource.
The key insight: AI is a tool, not a replacement. The WordPress community has always been about democratizing publishing – giving people tools to create and share. AI fits that philosophy perfectly if we use it to augment human creativity rather than replace it.
Looking Forward (From 2023)
We were at the very beginning. GPT-4 had just launched. Image generation was brand new to most people. The WordPress ecosystem hadn’t yet integrated AI capabilities into core plugins and themes.
But the direction was clear. Every tool would get an AI assistant. Every workflow would get augmented. The people who learned to work with AI would have a massive advantage over those who tried to work against it or ignore it entirely.
For WordPress specifically, I predicted:
- AI-assisted content writing built into Gutenberg
- Automatic image generation for posts
- AI-powered customer support for WooCommerce
- Code generation tools for theme and plugin development
- Automatic accessibility improvements (alt text, semantic markup)
- Personalization and content recommendations
We’re still early. But buckle up – it really is going to be a wild ride.
—
Original Presentation: WordCamp Europa 2023 (Athens, Greece, June 8-10, 2023)
Session Recording: Available on WordPress.tv
Slides: 43 pages covering LLM fundamentals, tools overview, and practical advice
Find me: Contact form, @divydovy most places, hi@divydovy.com
Note: I work at Automattic as Web3 Lead, but AI/LLMs are a personal passion. These are my personal views on technology that’s evolving incredibly quickly. What was true when I gave this talk in June 2023 may already be outdated by the time you’re reading this – that’s how fast this field moves.