The best laid plans …

Now, whist I would obviously promote using AI to support you in business and life generally, I still prefer to write my posts and newsletters unaided, so that you know these are my thoughts and not just information simply scrapped from the internet and thrown together without consideration.

This leads me to my admission that I got distracted by various things over the last 3 weeks, including taking some time off to celebrate my 60^th birthday, and as a result I failed miserably to post even once a week and I’ve now got a lot of catching up to do.

In that sense this reflects the usual use of the phrase “the best laid plans …” which indicates things have gone wrong. However, it’s also appropriate as for some time I’ve been planning on doing more about sharing my enthusiasm and knowledge around artificial intelligence. I’m therefore very pleased to announce that along with my brother Vince, our business has been accredited by the Chartered Institute of IT to deliver their AI training courses. As a result, I’ve been completely wrapped up preparing the courses and working towards the delivery of the first training in the coming months.

Meanwhile, the World of AI has gotten even crazier, as the pace of developmental change increases at an astonishing rate. In my view the key emerging issues are Model Code Protocols (MCPs) and the first real steps towards truly independent agents with the release of Manus … more on that later.

So, let’s get started.

Big Sharks vs. small fish

The big thing for me is the continued jostling for position as the market has adjusted to the arrival of DeepSeek AI, and the subsequent disruption to the status quo. If anything, things have become even more complicated as other new players from China appear on the scene. One of the key things that DeepSeek did was to completely upend the pricing models, and the view that success in AI reasoning power required massive investment (as in billions of dollars).

Whilst DeepSeek and other new models were not cheap (millions is still millions), their approach to how things worked was significantly different, even if they did copy or use parts of existing systems. The net result is that this has forced the other big players to change their approach, releasing new models earlier, and making more existing models free to access.

DeepSeek has even faced new competition from other Chinese companies, with Baidu’s ERNIE X1 model being 50% cheaper again, whilst Alibaba’s QWQ-32B is said to score significantly better tests results (even if it doesn’t trip off of the tongue as a name!).

The other big players have also not stood by idle. Anthropic has launched Claude 3.7 Sonnet and Claude Code, and we’ve seen similar developments in models with Microsoft Co-pilot and Google’s Gemini.

For me however, the next big leap was the arrival of Manus. This is probably the closest we’ve got so far to true Agentic Agents, which function with little or no human input. I finally managed to get access to the beta version (there’s a waiting list) last week. What I’ve seen has impressed me, although the context window significantly limits what you can do in terms of scale. Obviously, this will change, but using a fairly simple prompt it went off and undertook a complex data-scrapping task, including writing its own code and using its own internal browser. It decided what it needed to do to complete the task and what tools were required, and changed its approach when it encountered problems – watching this happening on my laptop in real time was mind-blowing, and gave a glimpse at what is going to be possible very soon.

AI infrastructure continues to also make the news, with the UAE announcing a $1.4 Trillion investment into the USA over the next decade. Intel is also making investments in order to counter Nvidia’s dominant position, and Meta also announced plans to produce their own chips.

Legal issues and real-World applications

The debates around AI legislation continues, with particular focus on the issue of copyrighted materials being used for training and the issue of privacy. Even within the EU we saw the issue highlighted with two contrasting events. The first was an edition of one Italian paper being created entirely by AI (albeit with human oversight), whereas Spain introduced a new law requiring strict labelling of AI content, with hefty penalties for infringement. Amazon also released their latest version of Alexa, but caused controversy by removing the privacy opt out, meaning your new device will capture everything that you say or do and share it with Amazon, raising obvious privacy concerns.

As for the impact of AI, there have been a number of events, which illustrate both the positive gains that can be made with AI, but also the potentially negative impact on both employment but also in the field of warfare.  

Starting with the “negative”, the Chinese smartphone company Xiaomi announced the opening of a fully automated factory that can produce one new phone every second, which is managed entirely by AI. The Ukrainian military also released data showing that AI enabled FPV (First Person View) attack drones achieve a success rate of 70% - 80%, compared to 20% - 30% for human-only operated drones.

In contrast however, there have been some excellent positive examples of AI helping to advance healthcare, particularly in relation to cancer treatment. One group of universities and research institutes had managed to use AI to assist with cancer screening tests, achieving from 97%+ to 99.26% accuracy. More impressive however was an AI tool that can be added to an iPhone with a magnifying lens connected to the camera, that can tell with 99.9% accuracy whether a melanoma is cancerous or not. Within a short period of time, this has already helped to identify 13,000 cases of skin cancer at a very early stage.

Regards

Tom Carter