Addams Family LLMs
Agent Scully has already committed us to another article about Large Language Models, so let us look at them today.
We have been finding it harder to niche down than expected. We're into week five now, a month into this AI experiment. Even AI in the workplace turned out to be not specific enough. We did a deep dive on LLMs and Microsoft Copilot and Quantum computing. Perhaps our niche within the niche [What the hell even is a niche? Fester.] is that we're just trying to make this stuff memorable. Anyone can write a forgettable newsletter, can't they?
Agent Scully has already committed us to another article about Large Language Models, so let us look at them again today before moving on to more interesting topics like self-driving cars. Don't forget you can ask us questions in the comments and we will answer every one. Pitching an article at a general audience always means some people will want more and some will be confused. In Scully’s case… [I am still out here editing. Scully.]
So this week I am writing as Gomez Addams and my side kick is Uncle Fester, that bald one. [Actually I shave it. Fester.] I am actually quite tired right now, to be honest. It's late, so nearly time to get up, and I have an exciting night ahead, and it’s the 13th of the month too.
These are the things Google Gemini thinks are important for you to know about LLMs right now: Increased Context Window, Multimodality, Enhanced Reasoning and Problem-Solving, Focus on Efficiency and Accessibility, AI Safety and Ethical Considerations, Specialized LLMs. I read that X Files article last time and I think Scully is really getting into her stride. So this week, having set the agenda here, I am just going to let Uncle Fester take it from here. He makes the unthinkable thinkable, the ineffable effable, and in fact he's just actually better at this than I am.
We're going to a team building event in town later. I think you might know we live in Westfield, New Jersey, that's over in America, and they have this new sort of night out. You take some friends and throw axes at the wall and all get very tipsy. Last time we did it one of us, lacking in the required concentration, split someone's head in two, right down the middle. I won't say who it was but it might have been my wife. The guy didn't know what hit him but he saw the funny side of it. Or perhaps I should say he saw the two halves of his own face laughing back at him from a bucket. Good night!
Uncle Fester Explains LLMs
Finally, a show I actually know! This week, The Addams Family has been chosen as our theme, and I must say—thank you, Gomez! It’s about time we used a show I’ve seen. However, I am slightly offended at being cast as Uncle Fester. I’m choosing to believe it’s because of his ‘caring nature’ rather than any physical resemblance to a bald man. My role in these newsletters is to make the smart knowledge accessible, so I’ll embrace it, I ‘care’ about people understanding. But let’s be clear—zero physical similarities. [I think we understood that long ago. Gomez.]
Now, onto this week’s topic: LLMs, Part 2 — a slightly more advanced look at how these AI models are evolving. Instead of comparing tools, let’s dive into what’s actually changing in this space.
What’s New in LLMs?
Increased Context Window
One of the biggest improvements in newer AI models is their ability to remember and process more information in a single conversation. Earlier versions of AI could only keep track of a few paragraphs at a time, meaning you’d often have to repeat yourself. Now, with increased context windows, models can retain much more detail, making them better at summarising long documents, keeping track of previous parts of a discussion, and maintaining more coherent conversations over time. I would say that garbage in garbage out applies to this, you need to be using a clear prompt. I tend to follow this format for prompting – what is the persona you’re wanting it to replicate, what is the task at hand, what is the context, what approach do you want the tool to use and what is the output for?
Multimodality: Beyond Just Text
Originally, LLMs were purely text-based, meaning they could only process and generate words. However, now they can understand and generate text, images, and even audio. This means you can upload a picture and ask an AI to analyse it, or have it generate visuals based on your descriptions. In the future, this could lead to AI tools that integrate even more seamlessly with video and voice recognition.
Enhanced Reasoning and Problem-Solving
LLMs are becoming significantly better at reasoning through problems, not just pulling up pre-existing knowledge. Whether it’s debugging code, solving complex maths problems, or helping plan detailed projects, newer models are improving in their ability to ‘think’ through steps logically rather than just regurgitating information. This makes them more useful in areas like research, decision-making, and even creative problem-solving.
Focus on Efficiency and Accessibility
Newer AI models are being designed to be faster, more efficient, and easier to use. This means better response times, lower computing power requirements, and integration into more everyday apps. The goal is to make AI something that feels like a natural part of your workflow rather than something you have to go out of your way to use. Expect to see AI baked into more of the tools you already use, from email clients to spreadsheets and even your mobile assistant.
AI Safety and Ethical Considerations
With great power comes… well, a lot of responsibility. As AI becomes more advanced, concerns around bias, misinformation, and responsible use are growing. AI companies are working on ways to make these models more transparent, prevent them from generating harmful content, and ensure they don’t spread false information. There’s still a long way to go, but expect ongoing improvements in how AI systems handle ethics, privacy, and accuracy. We need a dedicated article for this one! [Noted. I pick up on these very subtle indicators. Gomez.]
Specialised LLMs: Tailored for Specific Tasks
Rather than one-size-fits-all models, we’re seeing the rise of specialised LLMs designed for particular industries and tasks. Whether it’s medical AI trained for healthcare applications, legal AI for contract analysis, or AI built for scientific research, the trend is moving toward more focused, domain-specific models that perform better in their respective fields. This could mean AI that is not only more powerful but also more reliable and trustworthy in professional settings.
Embracing LLM’s: Grow With It, Not Against It
AI isn’t some mysterious force lurking in the shadows like Thing skittering across the desk—it’s a tool that’s evolving fast and becoming a natural part of how we work and interact with technology. The smartest approach isn’t to fear it like Wednesday eyeing a cheerful summer camp, but to learn it, use it, and grow with it.
We’re at a stage where AI can enhance efficiency, simplify complex tasks, and even boost creativity, but it’s only as good as how we command it—much like how Gomez wields a rapier or Morticia tends to her carnivorous plants. The real magic happens when we work alongside it, shaping it with clear inputs and leveraging it in ways that make life easier (and, dare I say, a little more electrifying—Uncle Fester approved).
It’s not about replacing human intelligence—it’s about enhancing it. Those who embrace AI and build their skills around it will find themselves ahead, while those who ignore it may feel like they’ve been locked in the Addams’ dungeon, wondering how they got left behind.
So, experiment with it, refine how you use it, and make AI work for you. Because in the end, it’s not AI taking over—it’s the people who know how to use it effectively who will. And if that means channelling your inner Uncle Fester, illuminating the way with a few well-placed sparks, so be it.