Rules to Better AI Use - 16 Rules

ChatGPT is a user friendly way to interact with OpenAI's GPT models. Essentially, it takes a model of GPT that is fine tuned for dialogue and turns it into a chat bot that responds in a human-like manner.

ChatGPT is a conversational language model from OpenAI, accessible both on web and mobile that makes it easy to have conversations with OpenAI's GPT models, and adds lots of cool features on top, allowing your conversations to be much more natural.

It also saves your message history so that you can go back and look at your conversations later!

Why is ChatGPT cool?

Figure: ChatGPT's response when asked "Why is ChatGPT cool?"

ChatGPT is an incredibly powerful tool. GPT-4 is a noticeable improvement over GPT-3.5, and we can only expect the subsequent models to be even more impressive. For now, it does have its skills and weaknesses.

Here are the most important things for which we know it to be useful (and not):

✅ Use it for

1. Writing

It can draft emails, reports, proposals, writing projects, and templates, as well as proofread and edit documents.

2. Summarization

It can create concise summaries of meetings or lengthy documents/text.

3. Coding

It can assist with coding tasks by offering suggestions, debugging code, generating code snippets, and providing explanations for programming concepts.

4. Creativity

It can generate new ideas, offer creative suggestions, and provide guidance on resolving conflicts or addressing challenging situations.

5. Content

It can generate engaging content for various platforms, including social media, presentations, etc.

6. Critiquing

It can provide constructive feedback on various forms of work (in text format only), such as writing, presentations, and code.

7. Learning

It can provide resources, explanations, and tutorials on almost any topic.

8. Research

It can gather information on various topics, analyze data, conduct SWOT and competitive analyses, and weigh pros and cons of different options.

9. Translation

It can help in translating text between languages, providing quick and reasonably accurate translations for both personal and professional use.

And there are also areas where it performs poorly.

❌ Dont use it for

1. Non-textual information

It is not designed to process or generate images, audio, or other non-textual information.

2. Mathematics

While it can handle basic arithmetic and some mathematical concepts, it is not designed for advanced or complex mathematical problem-solving.

3. Emotion and empathy

It can mimic emotions, but it doesn't truly experience or understand them, which can lead to inappropriate or tone-deaf responses.

4. Context understanding

It may struggle with understanding context and maintaining continuity when a conversation gets long. 🤖

GPT is a large language model developed by OpenAI with major financial backing from Microsoft, and is currently leading the market in terms of technology and capability.

AI technologies have been being developed at companies like Amazon, Google and Microsoft for some time. The first large language model, GPT-1, was released by OpenAI in 2018.
It wasn't until the release of ChatGPT, which is powered by GPT, that the technology became widely known.

Isn't it called ChatGPT?

Whilst sharing similar names ChatGPT and GPT are different.
GPT stands for Generative Pretrained Transformer and refers to the large language model that underlies the ChatGPT frontend. This model accessed using an API.

This allows for developers to build their own applications leveraging the same capabilities of ChatGPT over the top of the GPT model.A great example of this is the SSW RulesGPT bot.

Why is GPT cool?

Figure: ChatGPT's response when asked "Why is GPT cool?"

How does GPT work?

GPT models are trained on a huge amount of data (often refered to as parameters) that it can then use to predict what words should come next when given an input.
For example, if the training data contains a weather report talking about beautiful clear blue skies, when you ask it what color the sky is it can call back to that data and respond blue.

GPT is trained on a huge amount of parameters.

• GPT-4: 1.76 trillion parameters.
• GPT-3: 175 billion parameters.
• GPT-2: 1.5 billion parameters.
• GPT-1: 117 million parameters.

Strengths

• The model can respond to queries using natural language.
• The model can handle grammatical and spelling errors.
• Responses can come from the model very quickly - no need to access a database to retrieve data.

Limitations

• The model is only as good as it's training data and has no knowledge of anything outside of this data.

Try it out in the playground

GPT and ChatGPT are similar tools that serve slightly different purposes. Each of them has strengths and weaknesses so it's crucial to understand the right time to use them.

Video: ChatGPT vs. GPT - Which Is Better? - Q&A Comparison (2 min)

ChatGPT

ChatGPT is strong at helping an individual do a specific task. For example, a user could could get responses to simple tasks like:

• "Write me a blog post on EF Core Code-First Migrations"
• "How do I create a table with EF Core?"
• "Can you help me find information about Northwind?"
• and much more...

GPT

On the other hand, GPT is better for tasks that require automation or programmatic integration. For example, it could be used to create systems for:

• Summarizing a page of text
• Tailoring inputs to output in a specific manner e.g. input a url, and it outputs an SSW rule
• Changing email content to fit a specific audience e.g. Product Owner, Developer, Designer etc
• and much more...

Additionally, GPT provides an API that developers can leverage.

Since AI tools are a fast evolving landscape, it's important to stay on the ball. It's a productivity race - and the mistake of falling behind can compound into financial losses and competitive disadvantage.

When people exeriment with new tools, often they can learn from each other, which helps everybody. Typically a company will have some people who are more enthusiastic about AI "AI evangelists" - and they are in a natural place to lead the exploration of AI productivity.

These AI evangelists should get together to stay up to date - and help spread the "what's good and what's bad in the AI world" to their colleagues.

The best way to do this is by having a weekly meeting.

• It should be run by the AI evangelists
• It should also include other attendees (rotating) so they might get some learnings too

Here's an example of a post-meeting email - use it as a starting point:

If the above format is too long, it can be good to get it down to the essentials:

As more organizations adopt GPTs and AI tools to streamline work processes, one common issue arises: employees are often unaware of what tools exist or how to access them. This leads to underutilization of powerful resources, and even duplication of efforts when different teams independently create similar tools.

A directory of AI tools (e.g. GPTs) ensures that everyone in the company knows what tools others have been using available.

Benefits

• Awareness: You want all your employees to know about the tools
• Adoption: Employees are more likely to use tools that are easy to find
• Avoid Duplication: You don't want 2 people to work on the same tool without knowing - you want visibility into what already exists.

Best Practices

1. Location - Make it easy to find by putting the directory front and center on your intranet homepage
2. Descriptions - Include a brief explanation of the purpose and use case
3. Support - Include the author so users know where to seek for help or clarification.

Integrating ChatGPT with your personal assistant, such as Siri for iPhone or Google Assistant for Android, can greatly enhance your mobile experience. This rule will guide you through the steps to connect ChatGPT with Siri and Google Assistant.

Connecting ChatGPT with Siri (for iPhone)

Video: Siri is AWESOME when combined with ChatGPT! (I'll show you how!) (12 min)

Connecting ChatGPT with Google Assistant (for Android)

At the time of writing, Android integration is a little trickier, but still totally doable!

Follow these instructions and you should be talking to your new AI assistant in no time.

Adding ChatGPT as an app on your Windows, Android, or iOS device can be extremely useful for a number of reasons.

For example, you can use ChatGPT to ask questions, seek advice, or just have a casual chat whenever you need it. Whether you're looking for information on a specific topic, need help with a task, or simply want to pass the time, ChatGPT is always available to assist you with its intelligent responses.

Furthermore, ChatGPT is constantly learning and improving its abilities, which means that the more you use it, the more personalized and accurate its responses will become. With ChatGPT as an app on your device, you can have an intelligent and reliable virtual assistant at your fingertips, ready to help you with whatever you need, whenever you need it. So, whether you're a busy professional or just someone looking for a fun and engaging chatbot, adding ChatGPT as an app is a decision that you won't regret.

Download ChatGPT for mobile

• Android
• iOS

How to add ChatGPT as an Edge app

ChatGPT is an extremely useful tool for software developers as it has been trained to understand how code functions. It will provide an invaluable alternative to StackOverflow, being a great resource to help developers troubleshoot.

ChatGPT can be used for:

• Detecting bugs in your code
• Solving compile time or runtime errors
• Generating code based on a text description
• Explaining how a piece of code works
• Translating code to a different language (e.g. Python code to C#)
• Minimising generation of boilerplate (e.g. JSON to C# classes)
• Helping automate the code refactoring process
• Helping perform code reviews

Try it yourself, copy and paste this into ChatGPT

What does this code do?
[HttpPut("{id}")]
public async Task<IActionResult> MoveRight(string id)
{
	try
	{
		if (await _legalApiDbContext.ParaLefts.Where(a => a.ParaId == id).CountAsync() != 0)
		{
			ParaLeft toDelete = _legalApiDbContext.ParaLefts.Where(para => para.ParaId == id).First();
			_legalApiDbContext.ParaRights.Add(new ParaRight { ParaId = id });
			_legalApiDbContext.ParaLefts.Remove(toDelete);
			await _legalApiDbContext.SaveChangesAsync();
			return Ok();
		} else
		{
			return StatusCode(StatusCodes.Status404NotFound);
		}
	}
	catch (SqlException err)
	{
		_logger.LogError(err.Message);
		return StatusCode(StatusCodes.Status500InternalServerError);
	}
}

Sometimes, a developer runs into a complex error and they start googling all over the internet to fix the issue. In strange or unlucky cases this can result in hours of searching.

ChatGPT helps smooth this process. Simply paste the error into ChatGPT and it will give you instant feedback about the problem.

For example, let's say you try to run ef migrations using the command:

dotnet ef database-update

If this command gives you an error like:

No project was found. Change the current working directory or use the --project option.

Then you could ask ChatGPT and it would give you the solution!

Try it yourself, copy and paste this into ChatGPT

What is this error in EF Core?
No project was found. Change the current working directory or use the --project option.

AI can be helpful, and in some situations, it is good to be transparent that it helped you. Other times, however, it is more important that the message feels like it came personally from you.

In the case that you want to be transparent, you should indicate when AI was used.

When you want to be transparent (generally internal communication), there are benefits:

• Adoption: Spread and encourage usage of AI by showing how and when you use it, enabling others to think about when they can use it too
• Transparency: Others can identify AI-generated content, promoting openness and trust
• Responsibility: Encourages responsible use of AI-generated content

For generated text, a good option is to use the 🤖 emoji.You should use it at the end of your message/email/etc. There are 2 options:

1. Use 🤖 when you used ChatGPT but you edited the result

   Example: “Improve SEO with these tips: Create quality content, optimize for mobile, and use relevant keywords. 🤖”

2. Use 🤖🤖 when the content is entirely generated by ChatGPT

   Example: "As we watched the sunset from our office balcony, the conversation turned to the latest tech trends. The sky was painted in hues of pink and orange, but our minds were on the latest software updates and hardware releases. The birds chirped their goodnight songs as we debated the merits of different programming languages and platforms. It was a peaceful moment, a brief pause in the chaos of the tech world, and we couldn't help but feel grateful for the opportunity to work in such an exciting and constantly evolving industry.🤖🤖”

This is also important for generated images, Meta has developed invisible watermarks that can be used to indicate that an image was generated by AI.

Google has also done something similar. Check out SynthID

Setting Perplexity as your default search engine can significantly enhance your browsing experience by providing you with AI-driven, personalized search results. With Perplexity, you get more than just a list of links; you get context-rich answers, summaries, and insights tailored to your query.

Why choose Perplexity?

By making Perplexity your default search engine, you leverage the power of AI to get context-rich, personalized search results. This can help you find information faster, with more relevance, and with a better understanding of the context around your queries.

Here’s how you can set Perplexity as your default search engine:

For Chrome

1. Open Chrome and navigate to Settings: Click on the three vertical dots in the top-right corner and select "Settings"
2. Access the "Search Engine" section: Scroll down to the "Search engine" section or use the search bar at the top of the Settings page to find it.

3. Add Perplexity as a custom search engine:

   • Click on "Manage search engines"
   • Scroll to the bottom and click "Add"
   • Enter the following details:
   • Search engine: Perplexity
   • Keyword: perplexity.ai
   • URL with %s in place of query: https://www.perplexity.ai/search?q=%s
4. Set Perplexity as the default search engine: After adding, click on the three vertical dots next to Perplexity and select "Make default"

For Firefox

1. Open Firefox and go to Preferences: Click the hamburger menu in the top-right corner and select "Preferences"
2. Navigate to the "Search" panel: On the left-hand menu, click on "Search."

3. Add Perplexity as a search engine:

   • Scroll down to "One-Click Search Engines"
   • Click "Find more search engines" and manually add Perplexity by entering the URL: https://www.perplexity.ai/search?q=%s
4. Set Perplexity as the default: Once added, click on Perplexity and select "Set as Default"

For Edge

1. Open Edge and go to Settings: Click on the three horizontal dots in the top-right corner and select "Settings"
2. Access the "Privacy, search, and services" section: Scroll down and select "Address bar and search"

3. Add Perplexity as a search engine:

   • Click "Manage search engines."
   • Click "Add" and enter the details:
   • Search engine: Perplexity
   • Keyword: perplexity.ai
   • URL with %s in place of query: https://www.perplexity.ai/search?q=%s
4. Make Perplexity your default: Click on the three dots next to Perplexity and select "Set as default"

Generating charts is essential in conveying data effectively. While creating charts can sometimes be a tedious process, ChatGPT can assist by generating various types of charts quickly, saving you time and ensuring accuracy. This is especially useful when you need to visualize data for reports, presentations, or data analysis.

✅ Benefits of using ChatGPT to generate charts

• Speed: Quickly generate charts without needing to know specific charting software
• Accuracy: Ensure that data is correctly represented with the right type of chart
• Customization: Easily customize chart parameters to suit your specific needs

How to prompt ChatGPT for chart generation

1. Specify the Type of Chart: Clearly indicate the type of chart you need. For example: "Generate a bar chart comparing the sales of different products"
2. Provide the Data: Ensure you provide the data in a format that ChatGPT can interpret. For example: "Here’s the data: Product A - 30, Product B - 45, Product C - 25"
3. Ask for Customizations: If needed, specify any customizations like colors, labels, or titles. For example: "Label the X-axis as 'Products' and the Y-axis as 'Sales in Units'"

Limitations to keep in mind

• Data Size: ChatGPT works best with smaller datasets. For large datasets, it may be better to use specialized charting tools
• Customization Options: While you can request customizations, they might not be as extensive as those available in dedicated software
• Interpretation: Ensure that the data you provide is clear and unambiguous to avoid misinterpretations

By using ChatGPT for generating charts, you can streamline your workflow, making it easier to visualize and present your data effectively.

Adopting Artificial Intelligence (AI) can revolutionize your business operations, drive innovation, and provide a competitive edge. However, it's natural to have concerns about the associated security risks. Rest assured, with the right strategies in place, you can mitigate these risks effectively and harness the full potential of AI safely.

Common AI Security Concerns

When integrating AI into your business, it's essential to be aware of the potential security challenges. Here are the most common risks enterprises might encounter:

1. Data Breaches and Unauthorized Access

   AI systems can process sensitive data, making them prime targets for cybercriminals. As with any other information system, poorly implemented security measures, like weak access controls or inadequate patch management, can lead to unauthorized access, data misuse, and financial losses.

2. AI Model Manipulation

   AI models are vulnerable to attacks aimed at compromising their performance, such as adversarial attacks, model poisoning, and prompt injection.

3. Exposure of Confidential Data or Intellectual Property

   AI systems that learn from user inputs, such as machine learning models or AI-assisted coding tools, can inadvertently expose proprietary data or intellectual property. For example, confidential algorithms or secret keys used during training might be learned by the model and potentially be generated in other contexts, leading to unintended exposure of sensitive information.

Proactive Strategies to Manage Security Risks

Implementing strong security measures from the beginning helps keep your AI adoption secure. Here’s how to protect your AI initiatives:

1. Follow Regular Security Best Practices

To manage AI security risks effectively, follow standard security measures such as robust access controls, regular vulnerability testing, data encryption, and monitoring system activity. These practices help minimize vulnerabilities, ensure data privacy, and maintain system integrity.

• Implement general best practices: See our comprehensive rules on general security best-practices: https://www.ssw.com.au/rules/rules-to-better-security/

The remainder of the points here cover AI-specific controls that will complement your existing security posture and ensure your business can benefit from the advantages of AI with confidence.

2. Mitigate AI Model Manipulation Risks

To protect AI models from manipulation, implement proactive strategies including:

• Adversarial Attacks: Regularly test models against adversarial examples and improve robustness through training with diverse datasets.
• Model Poisoning: Use data validation techniques and monitor training data for inconsistencies.
• Prompt Injection: Implement prompt filtering and validation to detect and block malicious inputs.

3. Mitigating Risks of Exposure of Confidential Data

To minimize the risk of exposing confidential data or intellectual property:

• Opt-out of training public models: When using enterprise versions of models, such as OpenAI services on Azure, data used for analysis is automatically opted out of model training. Ensure that you are opted out of model training for any enterprise data, either by default (e.g. with OpenAI on Azure) or manually if necessary (e.g. with GitHub Copilot).
• Training Public vs. Private Models: Avoid using confidential data to train public models. If building a proprietary model, ensure it is secured as per standard security practices. For proprietary data, always use a private model to prevent unintended exposure.

4. Monitor AI System Activity

Ensure you are monitoring AI activity to enable quick detection and response to potential security incidents, minimizing damage and maintaining system integrity.

• Real-Time Logging: Implement systems that continuously log access, inputs, and outputs of your AI models.
• Alerting Mechanisms: Set up alerts for unusual activities that could indicate a security breach.
• Monitoring Dashboards: Use dashboards to visualize and track AI system activity, making it easier to detect and respond to suspicious behavior promptly.

Why Security Matters in AI Adoption

By proactively addressing these security risks, you not only protect your business and customers but also build a foundation of trust and reliability around your AI initiatives. Here’s why managing security risks is crucial:

• Financial Protection: Prevent costly data breaches and security incidents that can result in significant financial losses.
• Reputational Integrity: Maintain and enhance your company’s reputation by demonstrating a commitment to data security and responsible AI use.
• Regulatory Compliance: Ensure compliance with industry standards and regulations, avoiding legal penalties.
• Operational Continuity: Safeguard your AI systems against disruptions, ensuring consistent and reliable business operations.

Adopting AI can open up new opportunities for growth and innovation. By understanding and managing the security risks, you can confidently move forward with AI. Strong security measures will help keep your AI tools safe and reliable.

Adopting AI comes with complex legal risks that businesses often overlook. From intellectual property disputes to compliance with regulations and ethical data usage, failure to address these issues can result in lawsuits, fines, or damaged reputations.

Understanding and managing these legal implications is critical to safely and confidently using AI in your business.

Key legal risks to consider

• Intellectual Property (IP) Infringement - AI models often rely on vast datasets or pretrained models, which may incorporate proprietary data or content. Companies using AI must ensure they have the rights to use the data or models to avoid IP disputes
• Regulatory Compliance - AI adoption is subject to various regulations depending on the industry (e.g., GDPR for data privacy in Europe, HIPAA for health information in the U.S.). Non-compliance can lead to hefty fines and legal action
• Ethical Use of Data - Misusing customer data, even unintentionally, can result in legal consequences. The use of AI must align with ethical standards and respect user consent, particularly in the case of sensitive or personal information

How to manage legal risks

1. Secure proper licenses and IP rights

Ensure that any data or models used by your AI system are appropriately licensed or within the public domain. Work closely with your legal team to verify the rights to all datasets and algorithms before incorporating them into your systems.

2. Comply with industry regulations

Depending on your industry, ensure your AI system complies with all relevant regulations, such as GDPR, HIPAA, or CCPA. Compliance should include data privacy, data retention, and transparency around AI decision-making processes.

3. Implement ethical AI practices

Ethical AI isn't just about compliance; it's about building trust. Ensure your AI respects user consent, transparency, and fairness. This includes having clear documentation on how data is used, avoiding biased algorithms, and ensuring that decisions made by AI are explainable.

4. Use contracts to limit liability

When using third-party AI services or products, ensure that contracts clearly define liability in case of AI-related incidents. This can help protect your business from being held accountable for issues like data misuse or malfunctioning AI products.

By understanding the legal implications of adopting AI, businesses can avoid costly mistakes and operate within legal boundaries. Addressing these risks upfront helps ensure that your AI systems are not only compliant but also build trust with your customers and stakeholders.

Adopting AI can pose significant risks to your brand’s reputation. AI systems, if not carefully managed, may produce inaccurate information, make biased decisions, or violate customer privacy. Any of these issues can severely damage customer trust and loyalty, affecting your bottom line.

While AI has enormous potential, businesses must proactively manage the brand risks associated with deploying these systems, especially when they interact with customers directly.

Common brand risks with AI

• Inaccurate or misleading information - AI models may produce hallucinations or incorrect results, especially when dealing with unstructured or ambiguous data. If customers receive inaccurate information, it can undermine trust in your products or services
• Bias in AI decisions - If not properly managed, AI systems can perpetuate or even amplify biases present in the data, resulting in unfair or discriminatory outcomes, which can lead to reputational damage and loss of customer confidence
• Privacy violations - AI systems that handle customer data must respect privacy laws and customer expectations. Mishandling sensitive data or failing to anonymize information can lead to public outrage and legal consequences

How to mitigate brand risks

1. Ensure transparency in AI decisions

Customers need to understand how and why AI systems are making certain decisions. Ensure your AI models are explainable, especially when they are involved in critical decisions like credit approval or hiring.

2. Prioritize accuracy and continuous learning

AI models can degrade in performance if not regularly updated and fine-tuned. Regularly evaluate and retrain your models to prevent incorrect outputs from reaching customers. This can involve incorporating fresh data, conducting error analysis, adjusting hyperparameters, and incorporating human feedback to further enhance model accuracy. Additionally, consider implementing source citation or reflection mechanisms within the model to increase reliability and transparency.

3. Address bias in AI models

Bias can damage your brand, leading to loss of trust and negative publicity. To mitigate this risk, conduct regular bias audits to ensure AI models are fair and non-discriminatory. Identify and address sources of bias in the training data before they affect decision-making.

Bias audits should include regular reviews of input data, model outputs, and decision logic. Incorporate diverse perspectives during model development to catch potential biases early.

4. Respect customer privacy

Handle customer data with care. Anonymize data when possible, and ensure that your AI systems comply with privacy regulations, like GDPR or CCPA. Additionally, be transparent with your customers about how their data is used.

By addressing these brand risks proactively, businesses can maintain customer trust while reaping the benefits of AI. In a world where AI is becoming increasingly integrated into daily operations, it’s vital to ensure that your AI systems work in your favor, not against your brand.