It’s time to start, or if you’ve been unimpressed with ChatGPT-3.5’s capabilities, it’s time to restart. Often, “forget your previous prompt” is used to steer conversations in a completely different direction; so “forget your previous experience” is the command I urge you to take before attempting another crack at using generative AI in your work.
With the release of GPT-4o (the “o” is for “omni”), OpenAI has democratized access to one of the most powerful generative AI models available. This leap forward offers professionals, including those in the fields of tax, accounting, and law, a chance to harness advanced AI capabilities that were previously out of reach. While many have dabbled with the free version of ChatGPT, the jump to GPT-4o promises a significantly enhanced experience, featuring more accurate responses, greater contextual understanding, and the ability to handle more complex queries and tasks.
Refresher on the Basics of Generative AI
What is Generative AI?
Generative AI refers to artificial intelligence systems designed to generate new content, such as text, images, and even music, based on the data they’ve been trained on. Unlike traditional AI models, which are often limited to recognizing patterns or making predictions in closed rules-based systems (think Chess and AlphaGo) , generative AI creates new possibilities for knowledge workers to augment their performance on any intelligence task, providing a powerful tool for innovation and creativity.
Why GPT-4o?
GPT-4o represents a significant step forward in the field of freely available generative AI; however, Google’s Gemini and Anthropic’s Claude are all very close together in their (paid) capabilities. Dabble with them all. You’ll find the free GPT-4o compared to the free GPT-3.5 sees some massive gains.
A quick summary of improvements compared to your prior experiences on the free version:
- Enhanced Capabilities: GPT-4o offers more sophisticated language processing abilities, making it better at understanding context and nuances in text.
- Greater Accuracy: With advanced algorithms and a larger training dataset, GPT-4o delivers more accurate and relevant responses.
- Complex Query Handling: The model is designed to manage more complex and multi-faceted queries, making it an ideal tool for knowledge workers dealing with a volume of detailed information.
- Custom GPTs: Packaging frequently used prompts into “custom GPTs” allows free users to build out more sophisticated and effective prompts and eliminates the “copy & paste” steps.
Getting Started
So you’ve been convinced but aren’t sure where to start. The transition from not using generative AI to making it a part of your daily operations might seem daunting, but it’s more accessible than you might think. Here are three simple guidelines to either get you off the ground or help you restart.
Guiding Principle 1 – Invite AI to the Table
1. Invite AI to the Table: Start by integrating AI into familiar tasks. This approach isn’t just about throwing technology at every problem but rather about understanding where it can enhance your existing processes. Begin with tasks you know well, where you can immediately see the benefits and learn from the interaction. This firsthand experience is critical. You’ll encounter successes that encourage further use, frustrations that teach limits, and gradual familiarity that breeds skill. This is how every great technology adoption begins—not with immediate perfection but through persistent experimentation.
2. Expand Gradually: Once you’re comfortable with AI in familiar contexts, begin to extend its use to areas that are less familiar. As you grow more accustomed to the capabilities and limitations of AI, you’ll gain the confidence to apply it in new, perhaps more complex scenarios. Push your boundaries. Try leveraging AI to enhance decision-making, streamline operations, and perhaps even redefine problems.
3. View AI as a Partner, Not a Replacement: While AI will change how you work, it’s not here to replace you—or anyone else. The skills, insights, and experiences you bring to your role are irreplaceable. AI is a tool, when used effectively, that can make you even more capable, more insightful, and more efficient. But just like any tool, its effectiveness largely depends on the hands that wield it.
Guiding Principle 2 – Be the Human in the Loop
AI currently works best with human help; being that helpful human will be reciprocated with better AI outputs. It’s critically important to learn to be the human in the loop.
1. Understanding AI’s Limitations: AI is a powerful tool, but it requires human oversight to function effectively. Large Language Models (LLMs) like GPT-4o don’t actually “know” anything—they generate text based on patterns in their training data. This means they can’t inherently discern truth from falsehood, making human intervention crucial. Don’t be easily convinced by AI-generated content without verifying its accuracy.
2. Addressing Hallucinations: One significant issue with AI is hallucination, where the AI generates plausible but incorrect information. While newer frontier models are better at minimizing these errors, they can still fabricate convincing but false citations and facts. For example, there have been instances of fabricated case references in legal settings, highlighting the need for vigilance. Apply your critical thinking and verify the information provided by AI, ensuring it aligns with the appropriate context.
3. Providing Oversight: To effectively be the human in the loop, you must actively check the AI’s outputs for accuracy and relevance. Your role involves providing oversight, offering your perspective, and injecting critical thinking skills. This collaboration enhances results and keeps you engaged in the AI process, preventing over-reliance and complacency. Aim for a balance where you are 30% doer and 70% reviewer—never 0% reviewer. Your oversight is essential to ensuring the AI’s contributions are valuable and accurate.
Guiding Principle 3 – Anthropomorphize AI (but be careful)
This is a squishy guideline because AI is not a person, has no emotions, no empathy and no consciousness. There is no “what it’s like to be an AI” experience, it is NOT a person; however, its ability to mimic humanity through language can be leveraged in surprisingly powerful ways.
1. Leveraging Human-Like Interactions: Attributing human-like qualities can profoundly influence how we interact with these chatbots. LLMs are trained on our languages, a fundamental way we communicate with each other. This training results in AI mimicking aspects of human experience and communication and sharing similarities with human strengths and weaknesses; anthropomorphizing bizarrely can result in stronger outputs. (Tell it to take a deep breath before responding)
Similarities with Humans:
- Creativity: Just as humans can generate innovative ideas and solutions, LLMs can produce creative and sometimes even original content.
- Errors and Misconceptions: Like humans, LLMs can “misremember” facts or generate false information confidently, akin to human errors or misconceptions.
- Empathy: LLMs can mimic empathy by tailoring responses to fit the emotional tone of a conversation, similar to how humans adjust their communication based on social cues.
However, treating LLMs as traditional software can lead to misconceptions about their capabilities. Unlike conventional software, which follows strict, predictable logic, LLMs can adapt and respond in varied ways, making them seem more human-like. But this also means they can make unpredictable errors, so it’s crucial to maintain a balanced view of their capabilities and limitations.
2. Assigning Roles and Personas: To effectively anthropomorphize AI, consider assigning it a specific role or persona. In our department, a few of the various personas we use include a Comprehensive Tax Reviewer, an international tax lawyer and CPA called Intaxa, and an Executive Coach. Each persona helps tailor the AI’s responses and functionalities to suit specific needs. For instance, the Excel Maestro is not just a tool; it’s a persona specialized in Excel-based tasks, making the interaction feel more intuitive and responsive.
3. Enhancing Interaction and Output Quality: By giving AI clearer guidance, you narrow its focus from a broad, general-purpose tool to something closer to a specialist you might consult. This can simplify interactions and it improves the quality of the output by directing the AI’s attention and capabilities toward a defined area of expertise. To further enhance interaction and output quality consider a mixture of:
- Assign a Persona: Define a specific role for the AI, such as a tax reviewer or an executive coach.
- Set Clear Goals: Provide the AI with a clear objective for each task. For instance, “Generate a detailed summary of tax analysis report for our CFO.”
- Provide Context: You know the facts, the AI doesn’t. Help it out.
- Provide Examples: Providing general instructions that apply to all examples is generally more efficient than demonstrating all permutations of a task by example. However, in some cases, providing examples may be easier. For instance, if you intend for the model to copy a particular style of responding to user queries, which is difficult to describe explicitly, this is known as “few-shot” prompting.
- Outline Interaction Flow: Instruct the AI on how the interaction should proceed. For example, ask the AI to: “begin by asking two clarifying questions and then, based on my responses, pose a third clarifying question to refine your understanding further”.
Embrace these practices, and you may find that AI can not only mimic empathy and creativity but also deliver enhanced value in a way that feels more natural and engaging. Remember, while AI can take on any persona—from a venture capitalist to a newspaper editor—it remains a tool, reflecting the instructions, personalities and even weaknesses we imbue it with.
