Read Before Monday #45

January blues are more like January innovations - that's what it felt this month with everything happening and we still have one week to go! 2025 is going to be clearly "China, China, China". Lets start with DeepSeek’s development of the their R1 model showcases how innovation thrives under constraints, with efficiency-focused techniques like Multi-head Latent Attention enabling them to rival industry giants despite limited resources. This mirrors China’s broader tech-industrial strategy, where interconnected industries such as EVs and electronics create self-reinforcing ecosystems that drive innovation and economies of scale, albeit with sustainability concerns. Then I'll match that with nostalgia! First with how Clive Sinclair’s ZX80 and ZX81 computers similarly democratised technology, inspiring a generation despite their quirks, while Toshiba’s Visicom COM-100 struggled to gain traction due to limited offerings in a fast-moving market. Finally, in software, Tony Marston’s critique of Domain-Driven Design underscores the balance between complexity and simplicity, reminding us that the right tools, whether in AI, industry, or development, must align with the problem at hand to truly deliver value.

___

DeepSeek, a Chinese AI startup, has developed an "open-source model", DeepSeek-R1, that outperforms leading AI models like OpenAI's o1 on several math and reasoning benchmarks. This achievement is notable given DeepSeek's limited resources and the constraints imposed by U.S. export controls. Instead of relying on extensive hardware, DeepSeek optimised its AI through innovative software techniques and model architectures. The company employs young researchers from top Chinese universities, driven by patriotism and scientific curiosity, to pursue ambitious AI goals efficiently. Their methods, including Multi-head Latent Attention and Mixture-of-Experts techniques, allow significant energy savings during model training. By "open-sourcing" their model, DeepSeek fosters collaborative innovation, potentially challenging current U.S. AI export limitations.

• My Take:?It's time! We need to talk about DeepSeek and their great work with reasoning models. Recap time: a small AI team in China, constrained by U.S. export controls, took their GPUs and, instead of bemoaning their hardware limitations, optimised the heck out of their approach. They grabbed open-source models (like Llama) and went to work, creating something so efficient it rivals the heavyweights backed by far more resources and cash. The buzz? They’re not just competing, they’re excelling. It’s a classic case of innovation under pressure. DeepSeek is proof that when you have limited tools, you can still build something extraordinary. There's a twist, of course; their optimisation strategies aren’t just benefiting them. They’ve inspired a wave of efficiency improvements across the industry. Love it! But let’s not get into geopolitical drama or the “China vs. the world” narrative - this is about doing more with less. And as for using LLMs? Stop relying on them as one-stop solutions. If you’re letting an LLM write your code without checks or blindly using it for search, you’re missing the point. DeepSeek’s story is a reminder: innovation works best when paired with thoughtful, intentional use.

___

China's industrial prowess is not confined to isolated sectors; rather, it thrives on a network of interconnected industries that bolster each other. This synergy is particularly evident in the electric vehicle (EV) sector, where advancements in battery technology, consumer electronics, and industrial automation converge to create a robust ecosystem. Such overlapping tech-industrial ecosystems not only enhance supply chain efficiency but also foster innovation and economies of scale, reinforcing China's position as a global manufacturing leader.

• My Take: Oh China, again! Twice here. Yes, they're overlapping tech-industrial ecosystems and they’re essentially rewriting the rules of industrial growth. Picture this: instead of isolated sectors grinding away in their lanes, China’s industries - like EVs, batteries, and consumer electronics - are playing a coordinated game of leapfrog, where progress in one boosts innovation in another. It’s a strategy straight out of the efficiency playbook, with supply chains so integrated they practically finish each other’s sentences. Also, they just not only control the supply chain, they are the supply chain! The result? Economies of scale that make the rest of the world look like they’re still using dial-up in a broadband era. Sure, there are risks; overproduction, environmental concerns, the occasional trade spat - but you’ve got to hand it to them. This approach is bold, it’s strategic, and it’s working. Still, it raises the big question: is all this rapid growth sustainable? For now, it seems like China’s tech-industrial jigsaw is piecing itself together beautifully. Whether it stays that way depends on how well they can balance ambition with the realities of resource limits and global pressures.

___

Clive Sinclair, born in 1940, was a British inventor who founded Sinclair Radionics in 1961. He gained prominence with the Sinclair Executive, the first true pocket calculator, in 1972. Despite innovative products like the ZX80 and ZX81 computers, known for making home computing affordable, some ventures, such as the Black Watch, faced significant challenges. Sinclair's legacy is marked by a blend of groundbreaking successes and notable setbacks in the tech industry.

• My Take:?It's not the first time I've talked about Clive Sinclair and his rollercoaster ride through the tech world. But here's a guy who, at 21, starts Sinclair Radionics and churns out gadgets like the Sinclair Executive - the first true pocket calculator. Imagine, in the '70s, pulling a calculator out of your pocket that's as sleek as an iPhone 5. That's some serious James Bond tech for the time. But for every high, there was a low. The Black Watch? A ticking time bomb - literally, with batteries that had a penchant for overheating. Yet, Sinclair didn't quit. He gave us the ZX80 and ZX81, making home computing affordable and sparking a generation of bedroom coders - me included. Sure, the keyboards were awful (were they? ;) ), and the screens blanked out during processing, but hey, it was the '80s. Innovation is messy, and Sinclair's story is a testament to that. It's a wild mix of genius, ambition, and the occasional misstep. But isn't that how all great stories go? Meanwhile, someone is reproducing the ZX80 and it's brilliant!!!

___

In 1978, Toshiba introduced the Visicom COM-100, a home video game console developed in collaboration with RCA. Building upon RCA's Studio II architecture, Toshiba enhanced the system by adding colour graphics, a notable improvement over the original monochrome display. The Visicom featured detachable ten-button controllers and included five built-in games: Picture Drawing, Bowling, Pattern Drawing, Car Race, and Adding Game, all adapted from the Studio II's library but with added colour. Despite its sleek design and innovative features, the Visicom struggled in the market due to limited game selection and competition from more advanced consoles.

• My Take: I was so depressed with China advances that I had to bring two nostalgic posts this week! The Toshiba Visicom COM-100! Remember that? A console that tried to paint the town red, blue, and yellow but ended up colouring within the lines. In 1978, Toshiba took RCA's lackluster Studio II, gave it a splash of colour, and wrapped it in a sleek, futuristic shell. Detachable controllers? Check. Built-in games with educational flair? Check. But whilst the Visicom looked the part, it was still playing catch-up in a game that was moving faster than a pixelated car race. Limited game selection and stiff competition left it in the dust. It's a classic case of putting lipstick on a pig; no matter how much you dress it up, without substantial innovation, it's hard to win the game. See what I did here? Go back to the first two topics on China and you can see it with a different view.

___

In his article "Why I Don't Do Domain Driven Design," Tony Marston critiques Domain-Driven Design (DDD) by highlighting its complexity and perceived inefficiencies. He argues that DDD's practice of dividing large systems into bounded contexts with separate models can lead to unnecessary fragmentation. Marston emphasises the importance of reusability and productivity, suggesting that his approach - utilising a unified model with a layered architecture and separate classes for each database table - offers greater efficiency. He also questions certain DDD practices, such as the use of repositories, factories, and getters and setters, viewing them as adding unnecessary complexity without sufficient benefit. Marston concludes that while DDD may have its merits, it doesn't align with his focus on simplicity and high productivity in software development.

• My Take: Domain-Driven Design (DDD) always sparks debate, doesn’t it? On the one hand, it’s like having a GPS for navigating the tangled web of complex business requirements; it maps out strategic boundaries, gives you tactical tools, and ensures everyone’s speaking the same language (devs and stakeholders alike). The idea of separating concerns with layered architecture is brilliant in theory, keeping your domain logic pristine and your codebase less chaotic as it grows. But here’s the thing: DDD can feel like overkill for smaller or straightforward projects. Not every team has the bandwidth to maintain strict boundaries, or the need to split hairs over whether their code matches some textbook definition of “pure.” It’s powerful, yes, but only if applied where complexity demands it. Otherwise, it risks adding layers of complexity where simplicity would do just fine. Like all tools, DDD shines brightest when used intentionally, not just because it’s trending in architecture blogs.

___

This Week in GenAI

Live is here, as usual! This week was all about OpenAI's Operator, Sir Humphrey Appleby (the new AI for the UK Gov), Trump trumping Biden and Transformers Square. ?

In other news:

• OpenAI launches Operator - an agent that can use a computer for you | MIT Technology Review
• Developer Creates Infinite Maze That Traps AI Training Bots
• AI is weaving itself into the fabric of the internet with generative search | MIT Technology Review
• RD-Agent: An open-source solution for smarter R&D - Microsoft Research
• What Makes AI Agents Different from Other Types of AI Tools? | AI Advances
• Siri Is Super Dumb and Getting Dumber