Why waiting for ChatGPT5 won’t save your business

I still remember handing ChatGPT4o to my daughter and watching her plan her 9th birthday party on the swing at the end of the garden chatting to her new best friend, Sky. Forget the Turing test this was the Taylor Swift themed, fancy dress, balloon fest test.

This age of ChatGPT is truly remarkable in its explosive progress. Many suggest that the shift from ChatGPT3 to 4 was like going from a preschooler to high-schooler, when performance is benchmarked some estimate a 2x Order of Magnitude jump. Those same analysts expect a much greater OOM leap by 2027. But there’s a problem - does the best Chat tool in human history help your business tangibly create the +5-10% revenue opportunity many are predicting or the margin improvements from efficiency promises made by GenAI cheerleaders?

?

Let’s break it down

Generative Pre-Trained Models apply Transformer models to vast amounts of training data, by consuming vast compute to create billions of parameters which in turn generate tens of thousands embeddings in an embedding space not like our three dimensional world but a >thousand dimension universe. Sprinkle in crowdsourced reinforcement learning through human feedback and we have an impressive large language model, which usefully answers our questions.?

I think it’s fascinating to watch the consumption pricing of later models where the disparity between improved performance and reduced cost signal to us the algorithmic progress in using compute more efficiently. We’re not seeing a 1:1 relationship in performance: compute.??Combining ever growing compute power with ever better algorithms creates big leaps. But these two great forces do not hold the key to Business results on their own, nor necessarily AGI.?

?

Tension in the Boardroom

In surveys of thousands of organisations we see the same GenAI use cases generally at pilot or in adoption at time of writing. Generating marketing content, summarizing meetings, or writing code. In essence these are extensions of what ChatGPT3 and similar models do. While these tasks are useful, they represent a narrow application of AI's potential. Investors and boards are increasingly demanding more substantial business applications of AI, yet the vast majority of pilots and scaled use cases remain stuck in these basic, albeit valuable, functions.?

Despite this modest initial return the only way is forward with one source claiming 80% of companies referenced GenAI in earnings calls in 2023 (GenAI not AI). This is previously unseen investor pressure eclipsing Cybersecurity, Cloud or Mobile scrambles of the past decade or so by investors. In fact many Private Equity firms now have a larger emphasis on GenAI than Cybersecurity in Due Diligence processes! What!??

?

Thinking beyond chat

Astounding advancements have been seen in GPT-4, Claude 3, and Gemini Ultra to name a few not least in their multi modality and ability to handle vast context.?

These models are still largely reactive—they generate responses based on what they’ve been trained on. They answer your questions as quickly as possible, unlike a human Scientific Researcher or Professional Knowledge Worker who would never give you the first answer that enters their head.??

To take the immense power of these Foundational Models we need to take more responsibility as businesses for applying them to real business problems. For most that means AI that can not just Generate but also Reason, Act and utilise the Tools of our business.?

?

Introducing Agentic AI (or for those who despise buzz words GenAI plus Agents)

This takes use from waiting for the next model to drop—to asking how to leverage AI as an active participant in our organization. This transition to agentic AI involves integrating three crucial components into our toolbox.?

?

1. Chains of Thought (CoT)

These enable AI to process tasks in a human-like, step-by-step manner, enhancing its ability to reason and make decisions.?

Simplistically, imagine ChatGPT muttering to itself for 5mims in the corner before answering. Or even imagine ChatGPT(a) is asked a question it then asks a question to ChatGPT(b). ChatGPT(a) takes the response to form a follow on question asked to ChatGPT(c) and so the chain goes on. By the end of this chain a more considered, filtered, refined response is formed.?

In fact recent developments by Devin (the leading software coding AI agent) in remediation of open source software bugs the Agentic tool spent an average of 10mins to resolve an issue with a maximum threshold of 40mins applied to it’s problem solving.?

?

2. Vector Stores

These are advanced databases that provide AI with rich, contextual data specific to your business, allowing it to make decisions informed by the nuances of your industry and operations. This can be as granular as vast CRM case histories, large ERP product catalogues or dynamic ITSM knowledge bases.?

Now, without re-training or fine tuning a LLM at huge expense we have simply augmented it with relatively small, even tiny amounts of real business data. Imagine hundreds of Terabytes of training data from OpenAI packaged as-a-service meeting just Kilobytes of your Master data and Gigabytes of your transactional data.?

?

3. Tools

To transition from generating responses to taking action, AI needs access to your enterprise tools. For instance, integrating AI with CRM, ERP, Analytics applications or indeed collaboration tools like Slack, Teams or indeed Email. This now enables it to perform tasks such as managing customer relationships, optimizing resources, or generating insights autonomously.

In fact current benchmarking undertaken by GAIA (General AI Assistant) benchmark is seeing a 3x performance uplift when GPT-4 is given a combination of COT, Context and Tools. The benchmark uses test questions which humans perform with 92% success but Generative Pre Trained models tend to achieve <5% success. These tests are in nature multi-modal, complex and require additional context. The seemingly most useful tool in these scenarios is simply giving an Agent access to a web browser.?

?

You’re probably not Microsoft

Tech giants like Microsoft, Google, and OpenAI provide Generative Pre Trained models—backed by vast compute power, cutting-edge algorithms, and extensive pre-training data. But while these models are powerful, they are still general-purpose. The responsibility of transforming these models into valuable assets that can truly impact your business lies with you.

Yet, how many of us have resident experts in this frontier technology sitting around waiting to be asked? These same people also happen to be increasingly expensive resources. Why not, then partner with a services organisation with these expertise to accelerate your innovation - don’t reinvent the wheel or start from a blank page.?

Creating space for controlled experimentation in the spirit of co-innovation might just be the single most important decision your business makes in 2024. Not because the AI revolution has??arrived but because it is continuous and accelerating.?

?

Start now, but don’t expect to finish

Not many of us can claim to have attained the value promised by Business AI but those who have started the journey in 2024 stand a far greater chance of success in 2025, 26, 27. And who knows by then AI success may be an existential consideration.?

Many will remember the early advantages of the winners of Cloud - those who went first. But many will also remember the relative speed at which other caught up once SaaS became the standard. This is not the same. AI is far more personalised in nature than infrastructure.?

Business AI is always going to require ‘last mile’ efforts from your business to make it land with meaning. That last mile gets easier and cheaper the more your business exercises the art of experimentation, iterative learning and adoption.?

So don’t wait for ChatGPT(n) to catch up with your industry, your business, your problems. Start your journey today whilst the burning platform beneath your feet is only lightly smoldering and ask yourself: what small Agentic Proof of Concept could I show my boss by Christmas??

After all, even if I’m wrong about 2027 you might just find you’ve given your boss, CEO and Chairman the best Christmas present they could have hoped for.?

?

Looks like you might need a new suit for that January earnings call!?