Good morning. Welcome back to Scaling Greatness, a newsletter from Integreon focusing on amplifying business excellence and innovation.

Much ink has been spilled about the merits of ruthlessness in business. Audiences love a heel - the archetypal cold, calculating boss who sees fellow humans as disposable makes an excellent villain in almost any storyline. Think about Mr. Burns, Miranda Priestly and Logan Roy; and in real life, we devour stories about Steve Jobs’ on-the-spot firings and Jack Welch’s vitality curve.

Empathy, on the other hand, can be overlooked by business leaders – who tend to focus more on winning the deal and growing the bottom line rather than the finer points of human connection.

For most companies, however, empathy and openness are doing a massive amount of heavy lifting. While tales of table-pounding executives and apathetic firings make the news, much of the world economy actually runs on empathy. Developing a strong understanding of client needs, working closely with teammates and conducting efficient, friendly comms are the sorts of activities which keep companies moving and customers returning.

In fact, research has shown that empathy has a demonstrable impact on productivity, innovation, employee satisfaction and retention, which in turn can positively impact the company bottom line. A study published by Catalyst from Tara Van Bommel, PhD entitled “The Power of Empathy in Times of Crisis and Beyond” found that 76% of people with highly empathetic senior leaders reported being often or always engaged at work, compared to 32% with less empathetic leaders.

A successful firm may have a cold image – or an intimidating chief executive – but companies who want to build a culture that truly fosters growth and innovation incorporate empathy and understanding at their core. Many companies (including Integreon) offer empathy training to improve both employee and client outcomes.

Empathy is especially important in client work. Businesses and client-facing teams that want to grow long-lasting relationships will recognize the value of cultivating “trust records”, built on personal connection, genuine interest and mutual understanding. By focusing on practicing empathy, we can not only deepen our client relationships and boost employee satisfaction, but also create a path to steady growth and better outcomes for everyone.

⧉ A Moment of Greatness

Highlighting companies who highlight greatness.

Atlantic Coast Conference: Accomplish Greatness

Last year, the Atlantic Cost Conference (ACC) launched a creative campaign “Accomplish Greatness” to highlight their success in athletics, academics, community engagement and more. They continue to recognize greatness throughout their sports and academic programs.

Click here for video.

We love to find moments of greatness anywhere and everywhere. Let us know where you see greatness or how your company inspires great work.

⧉ 7 Ways to Address AI Hallucination

📝 by John Wei

Retrieval-augmented generation (RAG) is a process by which a large language model (LLM) references a knowledge base outside of its training data. Often based on real-time web searching, RAG can help models provide sources to support their responses. Some commentators have argued that RAG can solve the problem of LLM hallucination, while others have argued that the approach has flaws of its own.

But perhaps we shouldn’t want hallucination “solved” in the first place.

Hallucination and self-correction can represent a form of valuable intelligence; hallucination for one person can be creativity for another. It's worth noting that recognizing and fixing mistakes has long been a unique quality of human intelligence – until now, with the arrival of AI. For instance, when ChatGPT gives me an incorrect answer, I challenge it and explain my reasoning. ChatGPT corrects its responses and even apologizes for making mistakes.

AI models are designed based on the emulation of the human brain. As humans dream, LLMs may also dream or hallucinate.

In the legal domain, where the reliability of AI outputs is crucial, we must address LLM hallucinations while still harnessing their accurate outputs.

During our product development and LLM tuning cycle, we have applied various techniques to greatly enhance the accuracy of LLMs. Typically, LLMs used in the legal domain have an accuracy of around 50% to 60% out of the box. However, after tuning, we routinely see 90% or higher accuracy, with end-to-end process accuracy reaching higher levels on par with traditional automation processes.

What do some of these tuning techniques look like?

1. Leveraging existing legal quality control software and processes for validation
2. Soliciting detailed reasoning and references from LLM
3. LLM reviewing the output from another LLM
4. LLM prompted to brainstorm multiple inquiry approaches
5. LLM prompted to rephrase responses focusing on completeness, correctness, and concision – the “three Cs”
6. Teaching LLMs step-by-step using domain-specific legal knowledge
7. Deploying a "human-in-the-loop" approach

Hallucinations can be addressed through a thoughtful tuning process. Out-of-the-box LLMs, however, should be used with the utmost care before extensive tuning.