AI: Search on Steroids
This is not going to be a particularly “financial” or market-facing article.
Instead, it's going to be a personal reflection on the proliferation of modern AI tools - what is commonly referred to as “generative” AI.
It might also be seen as an optimistic riposte to Alex’s article, in which he presented a “bear case against AI”.
Firstly, some context.
I am 40 years old. This means that I remember life before the internet. I remember life before smartphones. I remember life before Windows, Google, YouTube, and Facebook.
If I had to name which one of those had the most profound effect on my day-to-day life, it would probably be Google.
But back in the early days (1998-2004), it wasn't initially clear which search engine was the best.
We had AltaVista, Yahoo! and probably others I’ve forgotten about by now. AltaVista was the one that I used the most, for several years.
Eventually, that changed. My perception - broadly shared - was that Google was cleaner. Google was faster. And Google’s results were more relevant.
And so it was that all of humanity’s wisdom (or at least all of it that had been uploaded to a public server somewhere) became readily available, and deliverable to us in a split second.
I was in secondary school and then starting university around this time. The ability to Google something quickly became an art form in itself. It suddenly became possible to answer (almost) any question far, far faster than anyone in history had ever been able to do so before.
In my early career, the ability to rapidly find the answer to almost anything was definitely a skill that I leaned on. Knowing the right keywords, and figuring out the right phrasing for a query, would mean getting much better results than someone else. Of course Google was ubiquitous at this point, and everybody was Googling all of the time.
All of which is to emphasise the revolutionary nature of what Google did.
And I don’t think I’m exaggerating when I say that generative AI is now having a similar effect on humanity as the search engine did.
Generative AI: a search engine on steroids
It should be clear from the above that I’m not really a technical person. I’m someone who uses computers, not someone who writes or handles code. I am numerate, with an educational background in maths and stats, but my programming experience is modest. I generally interact with computers as a consumer and end-user.
That is the perspective from which I say that generative AI is a revolutionary tool.
Before generative AI: you put a search term into Google or Bing. A range of search results pop up. Ideally, one of these results is exactly what you are looking for. You click on it and then continue navigating until you find what you need.
After generative AI: you ask your favourite chatbot a question. It then uses patterns it has already learned, sometimes combines this with live search results, and then gives you exactly what you’re looking for, in a highly digestible format.
Remember when I said that Google Search was cleaner, faster and more relevant than its contemporaries? Generative AI bots are now doing that to Search: for many purposes they are now cleaner, faster and more relevant.
Examples of what I use AI for, from the unimportant to the very important:
Cooking: when I need guidance with recipes or instructions, AI will provide. The results so far have been delicious.
Medical: before calling the doctor, a quick chat with AI might help me to figure out what’s going on (bearing in mind always that it can make mistakes).
Financial/strategic: when I’m making almost any important decision, I can use AI to test my logic. I don’t trust that it will be right, but it can be counted on to give me an interesting new perspective.
But there are some cases where I don’t trust AI:
Simple math and logic: I have noticed many times that AI will give me answers and make claims that are so obviously wrong as to be almost comedic.
Detailed advice: if you ask a chatbot for very detailed advice, it can only supply an answer that makes sense in the context that is given. Therefore, in order to get a sensible answer to a very complicated or detailed question, a large quantity of background information must be given. And if any important details are left out, the answer provided is likely to fall short.
There is a parallel here with early search engines.
The same search engine, in a different person’s hands, could either be far more useful or far less useful. When searching for hard-to-reach information, it was all about knowing the right keywords within the right search phrase.
I suspect that the same is true for AI.
It’s not the case that these tools eliminate the need for human intelligence.
Instead, like any other technology, they provide us with new opportunities that require human intelligence in order to capitalise on them.
In the short-term, this may (and probably does) mean job cuts in certain sectors. For example, US technology employers have not been shy about cutting jobs recently. Many of them have cited AI as a reason for the cuts, with investment instead being focused on AI-related coding efforts.
I don’t want to make too many grand predictions in this article - I’m not qualified to do so - but I am willing to make two.
Firstly, based on US tech valuations and what I’ve observed in the UK small-cap market, I believe that AI almost certainly is driving some degree of investor overenthusiasm and corporate overinvestment.
Markets are nearly always obsessed with something. That something is invariably at the heart of an overvalued and speculative frenzy. AI is now the phenomenon that we have collectively become obsessed with.
Secondly, I also believe that AI will have a profound and long-lasting effect on how we work and indeed how we go about our daily lives. I can say this from personal experience: I no longer simply “Google” something. My first instinct is to ask a chatbot what it “thinks”.
This is a major change in how I seek information. And I’m not alone. The number of people using AI tools every day is estimated by various sources at somewhere between 100 and 600 million.
I don’t see this number falling any time soon - and if it does, it won’t be because I’ve stopped asking for an AI’s opinions.
Disclaimer
This is not financial advice. Our content is intended to be used and must be used for information and education purposes only. Please read our disclaimer and terms and conditions to understand our obligations.
22 comments
I've gone down the AI rabbit hole in the last six months and gone from mild interest to strong interest and actual fear about what this means for the future of the majority of jobs. I'm now what you would call a power user of AI tools, including Claude Code, and I can directly see how this replaces a large subset of jobs, particularly junior jobs at present, and pretty much any job where the person is simply responsible for following a process or moving work through a process.
The latest models out this year are supremely capable and they are only getting better. I don't think it's hyperbole that we see at least 10% of white-collar jobs replaced by AI in the next three years. I wouldn't be surprised if this is as high as 30% once this catches on and people realize what it can do and start to trust it. It's going to spread by wildfire because it's just a no-brainer. It can do work so much more cheaply; even if not perfect we can add a human in the loop. Humans' role will be reserved to reviewing and making judgments on the work that AI does, as opposed to doing the work in the first place.
This said, I I think that Apple have made a huge error in being behind or not pursuing the AI strategy. I think that in an AI age where users increasingly talk to the AI and instruct it on what to do, as opposed to using screens, buttons, clicking themselves, Apple's supreme competitive advantage in user interfaces becomes much less important. The interface is rapidly becoming a voice, and you are just speaking or typing directly to the AI, which interprets your instructions and then goes off and interfaces with whatever tool, website, or API needs to fulfill your requirements. You could express my view here as long as Google is short Apple over the next 10 years.
Here's a question away from shares but AI related I've ended up going down a bit of a rabbit hole this weekend using Gemini for details on shall we say two rogues from my night clubbing years in the 1990's around the market town where I grew up.
The two famous local names always had a bit of a rivalry shall we say one of them I knew very well but we lost contact 25 plus years ago. Anyway I type their details into Gemini and its bringing up some remarkable insight's into events around these two in the 1990's which I personally was never aware of.
I can only presume to gather this information it is scraping very early internet local town forums/sites to gather the information or possibly Facebook groups.
When I ask it to send me links to its resource material it sends me through to an a 1990's style ancient looking forum but when I try and search what I'm looking for on said forum it brings up nothing.
Is there a better way of doing the prompts/search's to backtrack where its getting its source material from or is it a case that a lot of these early forum sites have been deleted and now sit in some inaccessible archive ?
The main reason I'm interested is that this is a great example of a small market town probably 80-100k population urban legend with two widely known characters if you lived their.
One of the results Gemini is finding is of an an aledged showdown in the early hours of a summer morning that drew a crowd of hundreds to the town market place as the pubs emptied as word got round.
As mentioned one of the characters involved was on the fringes of my going out group at the time and none can remember this happening but it either did and we all missed it or Gemini is making up a great story that would fit the next Guy Ritchie film.
Any pointers views would be appreicated.
Many probably know this but still: You can put a lot of data into an excel / word type doc. Give it an explanation of 1 who you are (eg private investor / compliance officer) , what you want it to review (eg speadsheet of your current portfolio or for compliance office FCA handbook) , what questions to ask (eg consumer duty) /get answers. You can chuck in several PDF's etc &/or websites to review. This way you get an AI specifically understanding context, researching your specific request and giving an output based on these specifics. The better /more specific the question the better the result. If approached with this kind of method the results can be very powerful.
Great article. AI is very good at analysing company data and looking for reports and forecasts. Can also look for insider dealing, short positions and can even discuss trading strategies.
Of course not as good as the Stockopedia team ;)
Indeed, I'm finding AI chatbots exceptionally useful in everyday life both personally and professionally.
I've used it successfully for fixing a tap, sorting out long standing issues on my PC, cooking, relationship advice, estate planning, data analysis and root cause analysis.
It's genuinely useful and productivity-enhancing.
What i intensely dislike is the amount of social media content now ai-generated and 'clogging up' youtube and, I'm sure, other platforms that I dont use. I hope an ai filter can be developed.
For YouTube alternative access, I am now running NewPipe - on my android (not linked to my Google a/c) - which is against the YouTube terms of service, but does a very nice job of getting rid of all the AI clutter. Apparently Android will block NewPipe completely this or next Sept......
Excited to know how Stockopedia is going to utilise a LLM in the future with their knowledge, expertise and data.
Stocko could put the entire discussion board threads into NotebookLM or similar - that would be interesting.
I have been a software engineer since 1969. I have learnt a few things about developing software.
Firstly, most statements of requirements are incomplete, ambiguous and inconsistent. You cannot write adequate specifications in any natural language except for very small programs.
Secondly, testing software only shows that the software runs the tests correctly in the chosen order, today, so long as nothing is changed. That’s never what is needed for any important system.
Thirdly, cybersecurity has become critically important and criminals are using AI at least as effectively as other software engineers.
I think there are still more areas where we are yet to feel the impact. Having first trained as a software engineer and having retrained as a Psychotherapist in 2010. I can say that my former colleagues are concerned. So much for a job for life in software development. I've recently been using Claude, Anthropic's AI to write code for some of the projects I undertake at Grow Your Dough. I have
It is remarkable. I do have to direct it with suggestions as to what I want. I then have to test it to spot any issues but I have to say I'm delighted with what it has produced and it's ability to generate user interface icons at the drop of a hat.
So I can ask Claude to write code for different platforms, in different languages, targetting different parts of the software development stack covering user interface generation to all of the code required to implement my ideas. It could even generate user documentation if I asked it to. That has just sidelined many different engineers all with specialisms and a technical author.
That is going to have quite a profound effect on software development andd I agree that "the new barrier to entry is a laptop and an idea."
There is even talk of AI inventing its own programming lanugage that is more sophisticated. AI with a language that we can't comprehend ... hmmm.
"The parameters that tailor the features of such an AI focused programming language will be designed for the needs of machines and not human programmers."
Food for thought!
Phil
P.S. My therapist colleagues are also very concerned about AI too. Put aside fears about poor advice (which is a very real concern) and consider that an AI is available 24/7, is always willing to listen, doesn't charge you by the minute. Anyone who has suffered a bereavement will know that after a while you are left alone in your grief. For those that have limited funds it's understandable that they might "talk" to an AI. I hope those that are suffering find some comfort there.
The flipside is that a good therapist listens in many ways that an AI can't. I "listen" for body language as well as content. I listen for sutble pauses, tone or intonation in delivery. I gather much more context as I explore your story rather than just asking a question and presenting lots of ideas. I will also challenge whereas an AI (in my experience) tends to collude and support your ideas rather than highlighting challenges.
All that said ... AI is in development, maybe that will change
"There is even talk of AI inventing its own programming lanugage that is more sophisticated. AI with a language that we can't comprehend ... hmmm."
Well, considering we started years ago by writing at machine code level, I am sure some of us could eventually reverse engineer what they are writing. Unless of course AI also make some processing chip that is beyond our engineering knowledge :-)
Seriously, when will AI actually think for itself rather than just scanning data for the best answer faster than we can, perform deductive reasoning, touch, smell, feel, and develops beliefs, and come up with its own business ideas? That is the time to worry about AI, perhaps.
Trust me having worked on Aircraft Stabalisation & Autopilot software using 68000 for the EH101 helicopter (now called Merlin) I can tell you no one wants to go back to writing machine code.
Ah, a youngster. LOL.
It is still very early days
Like everyone else i have issues with trust.
How do you check a forty page report produced by an ai with a lot of dense information and tables.
There are techniques on ways of getting reliable answers, by asking the right questions in the right way - which does considerably improve the quality of the results.
My favourite ( i think it's called adversarial reprompting) is to pass the question and the answer to a different artificial intelligence and then say do not run the prompt, test that the report matches the prompts requirements , has it made any mistakes, any omissions how can the report be improved and how can the prompt be improved.
You then have a discussion and the scope can evolve in interesting directions.
Hi
Nice posting Graham. This might interest you (& others) if you have not already seen it (post copied from X)...
GROK summarised a second post on the story .....
"The playbook has changed: [old path vs. new path with audience-first + vibe coding + AI automation] That's the new barrier to entry is a laptop and an idea."
Clearly a skilled AI literate person, but I think the final point is notable (much like your comments about skills forming your Google queries). ... "the new barrier to entry is a laptop and an idea."
Regards
Howard
Nicely put Graham. Approaching 70 years old I remember life, probably before your parents were born at a time when technology had advanced to provide us the thrill of stereophonic sound, delivered from a second loudspeaker box hardwired across the room from our vinyl record player with diamond stylus. Having studied for GCE A and O levels from extensive reading and many trips to my local reference library, I envied the following generations of students, who, just as I was by then, in my career at the time making extensive use of Google's Internet search engine to instantly bring information to me in a quality determined by the skill of my questionning to help me run my business and them with their study.
With the rise of AI with Chat and Gemini, the latter of which I use almost daily nowadays I confess to feeling initially a reluctance to use them. As if somehow they were a cheating tool designed to do so much of the heavy lifting involved with reasoning and discovery that they will remove the capacity for these skills from younger minds born into a world where to live without a smartphone or tablet seems almost inconceivable.
Like you Graham I still always check the search results, but find a well and fully worded question on any subject, or even a request for a comparison between two products or strategies often provides me in return a multi page answer in the form of a report, which often draws in aspects I might not have even considered and will provide links throughout for me to follow up additional avenues of research and enquiry.
Although I consider the value of shares in the major AI companies to be overpriced, since they can only improve efficiency for their customers, rather than actually produce a physical asset and each is in competition for the same limited share of take up. I now recognise what a truly remarkable asset it is to have at our fingertips and pray that ways are found to regulate the industry in order that the results it supplies are unable to be manipulated for nefarious or fraudulent purposes.
But are the big tech AI companies really overvalued? Following the recent (downward) rerating many are on prospective PEs in the low 20s (inc Nvidia, Alphabet and Microsoft). Not terribly expensive for such high quality companies in my view.
"Although I consider the value of shares in the major AI companies to be overpriced, since they can only improve efficiency for their customers, rather than actually produce a physical asset"
Connect an AI up to a 3D printer and you have assets. No sick pay, no unions, etc.
Did you know that when you receive a 3D printer, you actually print parts to assemble the 3D printer itself.
Keep extending that idea now ... an AI with a 3D printer that can assemble parts for a larger 3D printer that can build assembly robots, that can build larger production lines and factories, generating more and more complex components and items.
That will be the trajectory.
The 3D printer is your asset and AI will provide increased efficiency that reduces employment costs of unions and workforce sick pay, but will not remove the need for your purchase of the 3D printer and associated raw materials used.
You do until you have nobody available to buy your goods as none is in employed.
I’ve used all three available to me and they are very useful but I’m often asking whether they are sure this is right because…. And I’m thanked for pointing out the error and here is revised advice. So, the human interaction is essential.
And the AI interface promises to do better next time if challenged on why the wrong answer given in the first attempt. And if one asks the same question next week or month the answer will still be incorrect. And AI apologises. Again.
That is not to say it will not get better, just remember that it is using someone else's data, not yours, to find answers.