Quick caveat before diving in. This area of tech is moving insanely fast and while I’m aiming to stick to the foundations here, if you’re reading this post more than a month after it was published, check the details, things could be out of date.

So, first up. LLM stands for Large Language Model, and it’s the collective label for the chatbot tools like Chat GPT, Claude, Google Gemini, and others.

They’re being used for things from brainstorming support, to cheating on school homework, to coming up with computer code.

Here’s a quick rundown of the major players. I used another AI tool – Perplexity – to compile this, although I’ve put some of my own ‘translations’ in from some of the tech-speak I got back. (I’ll cover Perplexity in my next post.)

LLM	Chat GPT	Claude	Gemini	Mistral
Differentiator	The first one	The ethical one	The Google one	The French one
Key Features	– Versatile applications – Strong in content creation, coding, and task planning – Can do image and voice as well as text – High accuracy and coherence – Sort of connected to the web	– Excellent at following complex instructions – Strong analytical and technical performance – Good with large prompts and long conversations – Robust ethical safeguards – Not (currently) connected to the web	– Real-time conversational capabilities (as in voice input and output) – Excels in interactive applications – Strong contextual understanding – Seamlessly connected to the web	– Cost-effective solution (it uses smaller models which use less power) – Strong in consistent structured output – Not connected to the web
Primary Users	General users, developers, content creators, businesses	Academic researchers, technical professionals, ethically-conscious users	Businesses for customer support, interactive application developers	Companies seeking a balance between performance and cost
Company Reputation	OpenAI – Known for innovation, but has faced some controversies	Anthropic – Emphasises ethical AI development	Google – Established tech giant with a strong reputation	Mistral AI – Emerging player, gaining recognition for cost-effective solutions

Note: If you keep tripping over Microsoft Co-pilot in your daily computer rounds, it’s Chat GPT with a Microsoft skin over the top. I prefer going directly to Chat GPT as it’s more consistent in how it responds, but also because I hate it when massive companies try to force me into things.

And yes, they are all ethical (we can go down a shades of grey rabbit hole the size of Jupiter on this one), but Anthropic, the company behind Claude, was created with ethics as its cornerstone, so you’ll see it come up pretty constantly in commentary about it. I’ve also found it to be true in how it responds to some prompts.

Mistral is also maintaining its standards according to recent reports – and being in the EU means it’s subject to some pretty careful reviews and oversights that aims to ensure it walks its talk.

Using LLMs

The main way you’re going to work with an LLM is via prompts, mostly written at the moment, but spoken is getting easier and more popular.

The main tool we’ve been used to prompting up to now is Google, and they way we prompt that is single words and short phrases. LLMs work better with conversations. Some research done by the Google Gemini team found that:

(Based on the team’s research, the most successful prompts average around 21 words, yet people’s initial attempts are significantly shorter — usually fewer than nine words.) Plus, don’t forget you can ask follow-up questions. https://blog.google/products/workspace/google-gemini-workspace-ai-prompt-tips/

So that’s the first thing, go long! These tools work better with more information.

Here are some elements you might want to include in your prompt:

• Context – what situation has triggered this request?
• Intent – what do I want to do with this info?
• Specific info – are there details I either need included in the response, or want to supply to improve the response?
• Response format – how do I want the answer presented? (I tend not to worry about this one so much but it can be helpful)
• Persona – tell the tool it’s a certain type of expert, a language tutor, a marketing strategy expert, a developmental editor.

And some thoughts about using them responsibly:

• Be transparent in AI use – tell people when you’re using it
• Maintain human creativity and judgment – keep a human in the loop, making key decisions
• These things take a lot of power, so only use them when they can be useful
• Don’t use people’s names or copyrighted work in your prompts
• Prompt to combat stereotypes and biases
• Use smaller models when possible/appropriate – this is one of the things Mistral’s good at, but most of them have smaller versions
• Practice safe data – don’t put in Personally Identifying Information (PII), confidential, or sensitive data
• Verify the output – these things are built to be helpful, not accurate. They will lie with such joyful confidence you will want to believe them. ALWAYS CHECK YOUR FACTS!

I’m aiming to keep these pretty short but please let me know what else you’d like to know about these tools, and when and how to use them.

In the meantime, if you want to dig a little deeper, I recommend this recent post from Ethan Mollick on which AI model to use when.