Large Language Models (LLMs) have now trained on such large data sets that they've run out of data. One of the reasons OpenAI, Claude and other vendors now create synthetic data. By definition this creates a dumbed-down version of the data analysed- it is an average. That is why recruiters are experiencing a flood of CVs, even cover letters, that look the same. 

Add another factor- the high costs of training LLMs for generic application. In The Economist of 20th Sep 2024, it states that "In a recent blog Ethan Mollick of the Wharton School at the University of Pennsylvania grouped state-of-the-art LLMs into four loose “generations”, each requiring ten times more computing power and data than the last. He calculated that in 2022, when ChatGPT was released, models typically cost $10m or less to train. Some cutting-edge models developed since then may have cost $100m or more. Those coming soon could cost $1bn. He thinks training costs will eventually exceed $10bn. As pundits quibble over how predictable these scaling laws really are, the cost of training continues to rise (see chart).

That is why there is increased attention on training small language models (SLMs) on proprietary data together with a great deal of human oversight, and domain expertise applying reinforcement learning. 

Forces Horizon, founded and managed by former members of the Armed Services, has proved the efficacy of leveraging proprietary databases and SLMs to solve a problem experienced by veterans in the UK and USA. Used to speaking and writing in highly specialised language to fellow service people who understand the terms and three-letter acronyms they face a barrier with potential employers and recruitment agencies in ‘civvy-street’. 

How do you match the skills of a sniper to the roles available in London Markets, technology companies, or banks?  Hint- one of the most effective project managers at Lloyds is a former British Army sniper. It took a very long time for his current employer to realise that over a protracted hiring process in which the applicant and employer sought a meeting of minds and understanding. Luckily. like many enterprises in the UK, Lloyds signed the ‘Covenant of Duty’ with the Ministry of Defence (MOD) This is a moral obligation owed to all who serve or have served in the UK's Armed Forces. Lloyds spent the time to find out the potential offered by this sniper but other enterprises may have given up earlier in the process.

Forces Horizon makes the whole process faster, simpler,and effective using various data points like rank, role, qualifications, service history, and experience to suggest optimal roles in civilian life and prepare a personalised CV (resume) that paints a clear picture for potential employers and recruiters.

The Forces Horizon platform was built to map professional language, skills, and experience from Defence and Security to civilian equivalents. This is achieved by leveraging LLMs, a glossary of over 100,000 terms, and a proprietary database of roles and ranks in the military branches and security forces. As you can imagine there is a great deal of human oversight involved as well as training LLMs on this proprietary database.

Another key focus has been to experiment and optimise user interfaces (UIs) to bridge the gap between military and civilian audiences and avoid the problem highlighted in the Financial Times article. This ensures the FH platform avoids generic GenAI tones of voice in both the generated outputs and editing functions. 

Translating experience is the most challenging element of the military-civilian professional transition. By empowering veterans to successfully achieve this in resumes, professional profiles, and their own minds, Forces Horizon is leveraging the power of LLMs to serve this community whilst avoiding the well-known dangers of inaccuracy, misinformation, hallucinations, and security breaches.

Forces Horizon is partnering with enterprises like Honeywell,  commited to hiring veterans, and recruitment organisations like Veteran Hiring Solutions. With around 250,000 vets leaving the US and UK armed forces each year it is delivering a solution to a compelling problem. In the future it will extend this to other national militaries plus emergency services.

A proven use case for AI, LLMs, human oversight, and domain experience.

Companies like Global Logic apply the same strategy to serve markets such as insurance and financial services. Its insurance leader Chris Gill, himself an ex-Royal Marine, said: -

 “Our focus is on creating models that understand the unique contexts of various sectors, such as insurance and finances, allowing us to provide bespoke solutions for complex challenges. By collaborating closely with industry experts and integrating proprietary data, we aim to deliver AI-agnostic solutions that address the use case at scale, going beyond generic results and truly serve the needs of our clients."

Answering a question I put to him about harnessing the data from Building Management Systems (BMS) as part of an insurer's risk prevention services he replied:

“….harnessing data from Building Management Systems (BMS), such as those provided by Hitachi (Stewart Reeder Michael Elliott), can indeed unlock significant opportunities in risk prevention and claims resolution. By integrating GenAI and SLMs with BMS data, we can enable insurers to monitor real-time operational metrics and environmental conditions of properties. This can facilitate predictive maintenance, identifying potential issues before they result in damage or loss, which not only mitigates risk but also reduces maintenance costs.”

AI company Aiimi has announced its partnership with RiverStone International – a leading global acquirer and reinsurer of legacy and discontinued insurance business – to manage and unlock valuable insights across the firm’s vast data estate.

This new partnership is part of an initial three-year agreement, whereby Aiimi was selected as RiverStone International’s data and AI partner following a competitive bidding process.

Aiimi use AI, including GenAI, LLMs and SLMs/proprietary databases  to help organisations find, make sense of, and retain control of their data. Their technology and services are used by the likes of FTSE100 companies, the UK government, and the FCA. As part of the new partnership, the Aiimi Insight Engine will be working to discover and manage RiverStone International’s immense stores of information and legacy data; unlocking value from previously inaccessible insights, and mitigating security risks associated with sensitive data at the same time.

Aiimi achieves that without the need to integrate all the data in warehouses, lakes etc, so insurers and enterprises can leave the data where it currently is but still access, label, and make available for corporate insights and the various core platforms, applications,  and overall ecosystem many are deploying. Aiimi connects the data dots at the ‘data intersection’  between core systems of record, policy admin systems, claims, underwriting, counter-fraud, and supply management systems- to name a few. 

Authorised users will gain instant access to the information they need to handle the challenges associated with complex legacy insurance liabilities, ensure compliance in the ever-changing regulatory market, and steer insightful business decisions that drive forward RiverStone International’s rapid global expansion. This partnership will build upon Aiimi’s wealth of experience within the insurance industry, which spans over 10 years and includes work with the likes of Liberty Mutual and AEGIS.

These three examples show how pioneers like Forces Horizon, GlobalLogic and Aiimi are innovating fast with cost-effective solutions to levarage the benefits if Generative AI, small and large language models, and also anticipate the well known downsides of LLMs and GenAI. 

The Economist astutely observes that: -

"If the trend towards smaller and more specialised models continues, then the ai universe could contain a constellation of models, instead of just a few superstars.

This means that investors are in for a rocky ride. Their bets on today’s leaders look less certain. Nvidia could lose ground to other chipmakers; OpenAI could be supplanted. The big tech firms are hoovering up talent, and many of them make the devices through which, they hope, consumers will reach their ai assistants. But competition among them is fierce. Few firms yet have a strategy for turning a profit from generative ai. Even if the industry does end up belonging to one winner, it is not clear who that will be."

Forces Horizon in its specialised field, and both Aiimi and  GlobalLogic with their insurance and financial markets expertise aim to be winners.