Spicerack's Founder, Dave, talks to MD Justin about AI in practice - what it means for our clients and what the wider opportunities are as AI technologies improve.
The AI projects we're working on with our clients are critical to their ongoing success. Could you provide a general overview of their needs and how we typically respond to them?
Our clients' needs typically fall into the following categories:
- Great UX and UI design for complex tasks, such as querying rich data sets, configuring workflows and creating content
- Data cleaning, taxonomy and manipulation
- Automation
- Product development, which includes exploring new products and features that leverage emerging technologies or mitigate their associated risks
We help our clients develop their AI strategies, especially in product development. By identifying potential use cases, and working through agile R&D cycles, we evaluate the value of AI integrations in their product suites, enabling them to monetise solutions faster.
What's the potential of AI for businesses like our clients over the next couple of years?
Initially, we’ll see more task automation of less complex workflows, resulting in greater efficiency and cost savings.
There will also be better integration of AI components into customer and internal user journeys, reducing task friction and improving knowledge retrieval and transfer.
We'll see remote agents working alongside their human counterparts, as these models are augmented with memory, more reinforcement learning from human feedback (RHLF) and better data. These could be researchers, customer sales representatives, triage nurses, code reviewers or security experts, to name a few. This will be a game changer for businesses, freeing people up to deliver better products and services through creative collaboration with their AI counterparts.
And what is the current pace of the development of AI?
AI and gen-AI models have gone from being able to complete tasks at a pre-school level in 2019, to a sixth-form level (through GPT4) since 2023. The latest Claude model consistently scores 60% on the Graduate-Level Google-Proof Q&A (GPQA), currently the most challenging postgraduate benchmark. In comparison, a PhD subject matter expert scores around 80%, while most graduate students struggle to achieve more than a few percent.
Listening to experts like Sam Altman and Eric Schmidt, we should expect another leap forward as GPT5 and equivalent models come online over the next 12 to 24 months. These models will be able to complete 5-hour tasks, as opposed to the current 5-minute range. This will take us from simple summarisation-based workloads t-based workloads. That's a game changer, especially when combined with increased context length, real memory, and access to other tools like servers to run code on, and other software to carry out tasks.
To make all this happen, there’s a big race to buy up power contracts and the latest GPU chips to build out ever-more-powerful compute cluster data centres to train and run the models. Nvidia stock is up a staggering 22,080% over the past decade, and huge investments are being made by the AI firms themselves to attract the best brains and push innovation.
And what kind of pace are we seeing with our clients in research and education, for example?
The businesses we work with have been aware of AI capabilities for some time. Most are using some form of AI tooling in existing workflows, whether starting to use CoPilot in their daily tasks, or integrating AI tooling into their platforms and services at a deeper level.
The hesitation to fully adopt generative AI largely centres upon precision/hallucination, privacy and IP concerns, and to some extent knowing when to go all in. Is it better to wait for the next generation of models, or to try and capitalise on early adopter advantage?
I’d say that our clients are already seeing the benefit of an early adoption pace, in providing a competitive edge, and in some instances establishing market presence before their competitors. The pace of AI in business is something we can talk about with confidence. So what about the pace for humanity?
Well, that shifted the gears a bit, mate. For a start, I’d say the democratisation of powerful algorithms, and the tools to gain maximum value from them, puts power in the hands of the many. These tools often enable us to solve problems as individuals, or small collections of individuals, rather than being the preserve of big corporations.
Developing faster and cheaper medicines, as well as other therapeutic and assistive technologies, should help improve the quality of life of many more people and reduce the burden on our healthcare systems.
And global warming...
Well yes, various AI technologies have the potential to impact positively on global warming. There’s real potential for better energy efficiency, for example through efficient battery storage, greater accuracy in weather modelling, and innovative weather intervention measures such as Marine Cloud Brightening.
Do you think AI will ever be able to mirror human traits and human genius? It clearly has the potential to simulate both, but will it ever fully replicate the complexity and depth of human cognition?
If we achieve AGI (artificial general intelligence), then I believe it will become almost indistinguishable in most areas and surpass us intellectually in many ways.
Although isn’t the way genius has long been perceived just a myth? Perhaps a better question to be asking is whether these algorithms are capable of true original thought and creativity?
