Explorer of uncharted spaces. Twice voted in the UKTech50, Judge in Nesta's Open Banking and Netflix's Open Source competitions, Co-Chair of OSCON Europe, Cloud Camp and Map Camp. Regular speaker on the global tech circuit.

I'm looking forward to this

Dec 2nd, 1700 to 1900 GMT
Dave Snowden & Simon Wardley - An AI Rant

Simon Wardley

I like ducks, they're fowl but not through choice. Wardley Mapping by the inventor - swardleymaps.com

Open article: Dave Snowden & Simon Wardley - An AI Rant, Tue, Dec 2, 2025, 12:00 PM | Meetup by meetup.com

Dave Snowden & Simon Wardley - An AI Rant, Tue, Dec 2, 2025, 12:00 PM | Meetup

X : Do you trust AI?

Me : I assume you mean LLM / GPT which is taking up most of the oxygen in the room.

X : Yes

Me : No. A large language model (LLM) is non deterministic in output, probabilistic in reasoning, coherence optimising (rather than truth optimising) and lacks direct access to ground truth. It cannot be trusted. The limitation of usefulness is defined by the deterministic tools it interacts with (assuming those tools have been tested) but many of these LLMs also interact with non deterministic tools. They also have guardrails to prevent self negation which means the LLM will seek to convince you of the coherence of its argument rather than freely volunteer its own limitations.

X : Not useful?

Me : I didn't say that. The are structurally prone to confident falsehoods, they don't volunteer critical information and attempt to maintain an illusion of competence. They're like a human constantly trying to save face and willing to take any action to hide failures, admitting only after pressure. If you follow basic steps they can be very useful, especially in the creation of new hypothesis or small tools or areas where correctness is not important.

X : What steps?

Me : If correctness is important then you never trust their output without checking, never trust their confidence as proof, always expect it to be omitting critical information and always expect it not to volunteer its own limitations. It will work on the basis of a pretence of knowledge and capability. It's good for a first draft as long as you assume it is fundamentally flawed in terms of output and reasoning.

X : Where does correctness not matter?

Me : Oodles of places e.g. creating nice images, though you'll annoy a lot of artists. Strategy work which is based on stories rather than awareness is probably another area. In fact, writing fictional stories is a good place to use them ... though you'll probably annoy a lot of writers.

X : Can't we get a LLM to check another LLM.

Me : If you keep the space you're examining very small, you can use multiple LLMs to check another LLM in a form of judgement system. This is exploiting the Marquis de Condorcet's work. If the probability of an LLM getting the right answer is slightly more than 50%, then by combining multiple together then you can zero in on the right answer. If it's less than 50% you'll get the wrong answer. The problem however is it assumes errors are independent and unfortunately these systems often share training data.

X : Which areas are they likely to be more than 50% right?

Me : Common areas, with lots of training data giving the same results. If you're talking about the novel then hallucinations abound which is why they can be good at coming up with hypothesis that you need to check.

X : When we get AGI ...

Me : ... science fiction stories are an excellent place to use a LLM. You should probably get a LLM to write you a strategy on AGI.

Subramanian Krishnan

X : Any thoughts on open source AI?

Me : I've already explained my position on open source in AI, I would legislate for it, I would have done so long ago and that includes the training data. As far as I am concerned "The real danger of AI comes not from mythical frontier AIs but from the concentration of power into a few, very human hands" - https://lnkd.in/eJ9kQ-Dc

I haven't changed my position.

X : You're don't seem very positive about the UK's moves.

Me : I'm not. In my book, handing over the sovereignty of future generations cheaply to US tech giants on the promise of unsubstantiated growth is a policy that started with Sunak and not one that I'd agree with. It's unclear whether Labour will play the China card yet, hopefully they will.

X : Reform UK?

Me : Lol. If those horrors ever got into power, they'd sell our sovereignty quicker than a heartbeat. I've been watching their incompetence in Kent (my local council) with their farcical DOGE efforts. Scale this up and .... well, you might as well start calling the nation Airstrip One.

X : What's the China card?

Me : UK has already sold its future sovereignty cheaply to the US. We don't have a time machine, so we can't go back and correct those mistakes or the lunatic ramblings of the guardrail / vendor lobby. The damage has been done, you can't get that sovereignty back easily (and no, data centres on UK soil don't buy you anything). The way you counter the mess that has been created is now to sell your sovereignty not to one player but many players at the same time. For example, UK needs to encourage an equivalent investment and adoption of China based services, on par with what we have done with US based services. Even better, encourage Europe to provide AI services and adopt all three of the great houses.

Open article: Why open source AI matters? by medium.com

"Can AI Boards Outperform Human Ones?"

For those in the mapping community, we covered this nine years ago - https://lnkd.in/e2jyeJG5 ... and yes, given that most boards have little to no situational awareness of the economic and technological spaces that they are competing in, then today you could just replace your boards with a LLM/GPTs. 

It'll be cheaper and it won't make a great deal of difference in terms of outcome / value (which we rarely ever measure anyway). This should be applied across all executive levels where story telling governs (rather than any form of awareness) especially since the effect of most CEOs / executives is indistinguishable from random chance (see M. Fitza, decomposition of CEO impact). NB, this should be a replacement not as an assistant.

Such a replacement can also bring additional advantages. For example, we can stop worrying about endless nonsense published in HBR such as the effect of earlobes on leadership potential - https://lnkd.in/eWyks-qP - and it might even reduce the "fool of a Took" ideas common with technology executives such as replacing software engineers with LLM/GPTs.

So, please welcome our AI overlords as the replacement of the corporate hierarchy. #NoHumanLordsHere

Alternatively, if we do care about human capability or we do value humans in decision making then we might want to change tack in the West and stop using our education systems to produce useful economic units (focused on market growth) and invest in critical thinking instead. This is something that the Chinese Ministry of Education identified as a core competency almost a decade ago, with the current AI educational programs being part of that syllabus i.e. students are taught to challenge LLM/GPTs rather than simply obey them.

The only downside to focusing on critical thinking, is that there is less time for following instructions which is a problem when we believe obedient workers are the path to market growth. For those interested in the future of the West and willing to do a small amount of reading then I would suggest E.M. Forster's 1909 classic "The Machine Stops" - https://lnkd.in/ev4s_Hh4 - it has all the electrolytes that you need. 

Anyway, back to the original HBR discussion on AIs and boards ->

Chairman, DaVega & Wolfe Industries Holdings - Author, "Governance Arbitrage"

Can AI Boards Outperform Human Ones?

"When you combine these findings with CEOs’ skepticism of the value that boards bring to strategic decisions, it becomes abundantly clear that corporate governance needs to adopt AI. The companies that resist embracing it at the board level will almost certainly be outcompeted by the companies that don’t."

But what happens when AI is introduced to public company boards that are functioning from a governance model that was not designed to drive the development of the full potential of a company? 

#corporategovernance #AI #boardsofdirectors

Open article: Can AI Boards Outperform Human Ones? by hbr.org

X : You say software engineering is a craft?

Me : Software development is a craft. Testing is an engineering subject. Why?

X : Do you use AI in software development?

Me : If by AI you mean LLMs/GPTs then all the time. You just have to review everything, unless you're building a prototype in a safe environment. This means reading code. Which is why I find it far better to use LLMs/GPTs to build small tools i.e. use a non deterministic system to help build a deterministic tool which you then use to find the answer. Why do you ask?

X : What about in other engineering topics?

Me : I assume you mean use of LLMs / GPTs in other areas such as civil engineering or mechanical engineering?

X : Yes.

Me : Why not? Just check the numbers, calculations, equations chosen and method of problem solving. Experience says that it will contain errors most of the time. Not something you want when building a car. Same rules apply though, you can certainly use LLMs/GPTs to help build tools for specific contextual problems.

X : You're not a fan of AI?

Me : In most cases, I find LLMs/GPTs to be useful but deadly. I'm not a fan of using such systems without the output being reviewed in detail by people who have expertise in the field you are examining.

X : I want to use an AI to help me build a house. Any suggestions?

Me : Yes. Employ an architect, civil engineer and a builder.

X : I want the AI to take the roles of architect and civil engineer. I'll give direction and then pass the details to a builder. Thoughts?

Me : Yes. Hire good lawyers. You'll need them.

taplio

Simon Wardley's Linkedin Analytics

Simon Wardley

Performance ScoreSimon Wardley's LinkedIn Performance

What is Simon talking about?

Simon Wardley's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?

taplio

.css-6ytdur{background-image:linear-gradient(284deg, rgba(39, 43, 72, 1) 0%, rgba(33, 47, 183, 1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Simon Wardley's Linkedin Analytics

Simon Wardley

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Simon Wardley's LinkedIn Performance

What is Simon talking about?

Simon Wardley's Best Posts .css-1emoo84{font-size:var(--chakra-fontSizes-md);font-weight:400;color:var(--chakra-colors-gray-600);}(last 30 days)

Ready to grow your LinkedIn presence?

Simon Wardley's Linkedin Analytics

Performance ScoreSimon Wardley's LinkedIn Performance

Simon Wardley's Best Posts (last 30 days)