Install DeepSeek on your Mac or PC in 5 mins or less.

Explaining the hype with a hands on example

Well, well, well... it appears DeepSeek has rather cheekily wandered into the AI playground and started showing up the bigger kids! Let me paint you a picture of why this is causing such a delightful kerfuffle in the AI world.

There’s no substitute for experience and to that end, I’m going to show you how you can install and try DeepSeek on your own Mac or PC in just a few minutes, even if you struggle with getting the toaster working.

The secret sauce here is something called Chain of Thought (CoT) reasoning - imagine if, instead of just blurting out an answer like a contestant on a game show, the AI actually shows its work, rather like that one meticulous friend who insists on explaining exactly how they solved the Sunday crossword. This has been something of a crown jewel for OpenAI's premium models ($200/mo), particularly their o1 series.

But here's where it gets interesting (and why some executives might be choking on their morning coffee): DeepSeek R1 has managed to replicate this capability, and in some cases, they're doing it better.

What makes this particularly spicy:

1. The Performance Metrics DeepSeek R1 is beating o1 on some rather impressive benchmarks:

   • AIME (American Invitational Mathematics Examination)

   • MATH-500 (think workplace word problems on steroids)

   • SWE-bench Verified (programming tasks that would make a senior developer reach for the paracetamol)

2. The Cost Factor Here's the real kicker (and hence the panic in the markets) - they've done this with a fraction of the computing resources. It's like someone showing up to an F1 race in a souped-up family car and keeping pace with the pros.

3. The Open Source Angle They've released this under an MIT license, which in corporate terms is a bit like publishing your secret recipe book. Anyone can use it, modify it, or build upon it - commercially!

The Chain of Thought aspect is particularly fascinating. While o1 tends to be rather concise (think Sherlock Holmes delivering his conclusions), DeepSeek R1 is more like Dr. Watson - it walks you through every step of its reasoning process. Sometimes, this means it can be a bit verbose, but it's incredibly useful for:

• Complex mathematical problems

• Multi-step logical reasoning

• Programming tasks

• Step-by-step analysis

The worry for big AI companies isn't just that DeepSeek has matched their capability - it's that they've done it in a way that's:

1. More transparent

2. More cost-effective

3. More accessible

4. Potentially more scalable

So enough with the preamble, why don’t you try it for yourself? Remember - this is running locally on your PC, offline if you so choose.

What You'll Need

• A reasonably modern computer (if it can handle Chrome with multiple tabs, you're probably good)

• 16GB of RAM (8GB might work, but it'll be like running a marathon in wellies)

• About 50GB of free space

• 2-5 minutes of your time

The Two-Step Setup

Step 1: Install Ollama

1. Head to ollama.ai/download

2. Download and install:

   • Mac: Drag to Applications folder

   • Windows: Run the installer

Step 2: Install Chatbox

1. Visit Chatbox Releases

2. Download the latest version:

   • Mac: .dmg file

   • Windows: .exe installer

3. Install like any other app

Getting DeepSeek Running

1. Open Terminal (Mac) or PowerShell (Windows) and type the text below followed by return.

ollama run deepseek-r1:8b

1. Wait for the download (perfect tea-brewing time)

Note: The “DeepSeek-R1:8b” as the result of a rather clever bit of AI gastronomy - it's essentially a reduction sauce (Distillation) of the full DeepSeek model, but one that's managed to retain most of the flavor while being far more practical to serve up. Larger models require beefier hardware that’s beyond the scope of this quick example. So temper your expectations appropriately.

1. Open Chatbox

1. Click 'Settings' (⚙️)

2. Add Ollama:

   • Click 'Add New API Endpoint'

   • Select 'OLLAMA API'

   • API Host:

     http://localhost:11434

   • Save

Let's Test Drive Your New AI

Try these prompts to get a feel for what your new assistant can do:

For Writers and Creators

Can you help me outline a blog post about sustainable gardening?

Include 5 main points and some interesting statistics.

For Writers and Creators

I need to write a professional email declining a vendor's proposal

while keeping the door open for future collaboration. Can you help?

For Problem Solving

I'm planning a week of healthy meals for a family of four.

Can you create a shopping list and simple meal plan?

For Learning

Explain quantum computing like you're talking to someone who just wants

to understand why it's exciting, not the complex physics behind it.

What you’ll see is the ‘Chain of Thought” musings of the AI as it breaks down the problem.

Pro Tips for Happy AI-ing

• Start with clear, specific requests

• Use the chat history feature in Chatbox to keep track of good conversations

• If responses get weird, hit the "New Chat" button

• Save particularly useful conversations using Chatbox's export feature

When Things Go Wobbly

• If Chatbox can't connect: Make sure Ollama is running (check Terminal/PowerShell)

• If responses are slow: Close some memory-hungry apps

• If it crashes: Just restart both Ollama and Chatbox

• If answers get odd: Start a fresh chat

Why This Setup is Brilliant

• Works offline - perfect for planes, trains, and confidential automobiles

• No subscription fees - your wallet will thank you

• Keeps your data private - what happens on your computer stays on your computer

• Surprisingly capable - from creative writing to coding help

What's Next?

The DeepSeek story is rather fascinating when you step back and look at the whole picture. It's sparked what might be the tech industry's most expensive game of musical chairs, with NVIDIA shareholders briefly losing their seats (and $600B) at the mere suggestion that AI might not need quite so much silicon muscle after all.

But here's the thing - while DeepSeek has demonstrated that you can build a remarkably capable AI model for the cost of a luxury car rather than a small island, it hasn't quite triggered the AI apocalypse that some headlines might suggest. Instead, it's done something far more interesting: it's shown us a glimpse of AI's middle class future.

The market's initial panic about NVIDIA's future rather missed the point. Yes, DeepSeek managed to train their model for a relatively modest $5.6M in GPU rental costs, but they still needed those GPUs. What's changing isn't the need for specialized hardware - it's the democratization of access to AI capabilities. NVIDIA themselves pointed this out, noting that DeepSeek's success actually suggests a future where more organizations, not fewer, will be running their own AI models.

Meanwhile, the big commercial AI players like OpenAI aren't exactly quaking in their boots. Their $200 monthly subscription for o1 isn't just about raw computational power - it's about reliability, support, and an entire ecosystem of features. DeepSeek isn't so much threatening their business model as it is expanding the market, creating a new tier of AI accessibility for those who need powerful reasoning capabilities but can't justify enterprise-level spending.

This broader transformation hasn't gone unnoticed in Washington. The recent announcement of $500B in private-sector AI infrastructure investment through The Stargate Project, along with $140M for new AI research institutes, suggests that the U.S. sees this not as a threat but as an opportunity to accelerate AI development across the board.

What we're really watching isn't a zero-sum game between DeepSeek and established players, but rather the beginning of AI's maturation into a more diverse, accessible ecosystem. The future of AI isn't just about the penthouse suite of capabilities offered by OpenAI or the basement bargains of simple chatbots - it's about filling in all the floors in between. DeepSeek has simply shown us that the elevator can indeed stop at more levels than we thought.

And perhaps that's the real story here: not that DeepSeek is upending the AI industry, but that it's helping to fill out the missing middle - creating a future where powerful AI capabilities aren't just for those with enterprise budgets or deep technical expertise, but for anyone with a decent computer and a problem to solve. The market may have momentarily panicked, but it's starting to realize that this expansion of the AI ecosystem might just lift all boats, even if some of them need to adjust their sails.