ChatGPT guide for safety and security heads
On release ChatGPT broke records for being the fastest growing product in history; its probably responsible for making Artificial Intelligence (AI) an everyday term. The popularity of GhatGPT and its promise to save time and money has changed business outlook. It has given businesses confidence to upgrade the ubiquitous ‘digitisation project’ to the new ‘AI project’. And, like digitisation, AI projects will impact every business function, including safety and security.
Therefore, if you’d like to be part of the AI conversation, a great place to start is with ChatGPT; use it, learn how it works and know its limitations.
- Three ways ChatGPT saves you time
- ChatGPT limitations, do’s and don’ts
- ChatGPT’s AI explained
Three ways ChatGPT saves you time
At work, it’s not necessary to be brilliant at everything. Good enough will do, especially at secondary tasks. For example, no one expects a doctor to make stellar presentation slides, just good enough slides to get a message across. This is where ChatGPT flourishes, it’s good enough at many tasks to save most people time. Here’s three such tasks :
1. Presentations
While ChatGPT will not yet generate images it will produce high-quality text content and offer guidance and advice on how to craft an effective presentation. For example, the following question produces a ‘good enough’ response:
What should go into a risk assessment presentation?
2. Excel help
I have the below names and numbers in Excel and I want to separate the numbers from the names.
| London – 10,257,7000 |
| Birmingham – 2,560,500 |
| Manchester – 2,517,500 |
| Glasgow – 1,019,900 |
| Newcastle – 868,800 |
| Sheffield – 854,200 |
| Leeds – 798,800 |
| Bristol – 753,700 |
A Google search for how to separate numbers and name gives hundreds of video results. Which one is correct? Some videos are 15 minutes long. However, ChatGPT gives one result:
Separate numbers from letter [I included the above list in the query]
3. Intelligence gathering
Monday morning and I want to know what and where protests occurred in London yesterday. Can ChatGPT help?
List GPS coordinate of every protest in london yesterday
ChatGPT limitations, do’s and don’ts
Limitations
Development
ChatGPT is undergoing constant development. As a result, its ability changes. For example, the above protest result came from using a plugin that’s no longer available: Link Reader. Such plugins are available through the paid version of ChatGPT. This limits how heavily one should rely on ChatGPT for repeated tasks.
Input and output limit
ChatGPT has an input and output limit of 4,096 tokens. A token can be as short as one character or as long as one word. If a conversation or response approaches this limit, the model might truncate or provide shorter responses to ensure it doesn’t exceed the token limit.
Hallucinations
A hallucination is another way of saying ChatGPT gets an answer wrong or makes up an answer. For example, in this post ChatGPT says it is never morally permissible to utter a racial slur—even if doing so is the only way to save millions of people from a nuclear bomb.
Several factors can contribute to hallucinations:
- Ambiguous query: If a user’s query is vague or ambiguous, the model might generate an answer that seems off-topic or unrelated.
- Model Limitations: No model is perfect. Even with vast amounts of training data, ChatGPT can sometimes produce outputs that are not entirely accurate or coherent.
- Training Data: ChatGPT’s outputs are based on its training data. If there are inaccuracies or biases in the training data, they can manifest in ChatGPT’s responses.
- Complexity of Language: Language is inherently complex and nuanced. ChatGPT might sometimes misinterpret the intent or context of a prompt.
If you have good knowledge of a field any hallucination should be easy to spot.
Bias
ChatGPT is bias. For example, University of East Anglia showed a systematic bias toward liberals. Also, it favours English over all other languages. For other languages, the model’s performance might be less reliable or nuanced.
Do’s and Don’ts
Whenever you use ChatGPT ensure you:
Do:
- Turn off Chat history
- Ensure users are compliant with your data policy.
Don’t:
- Enter personally identifiable informationInformation which can be used to distinguish or trace an individual’s identity, such as their name, social security number, biometric records, etc. alone, or when combined with other personal or identifying information which is linked or linkable to a specific individual, such as date and place of birth, mother’s maiden name, etc. into ChatGPT
- Enter organisation names or addresses into ChatGPT
ChatGPT’s AI explained
ChatGPT uses three types of AI:
- Machine Learning (ML): A subset of AI where algorithms allow computers to learn from and make decisions based on data.
- Deep Learning: A subset of ML using neural networks with many layers (hence “deep”) to analyze various factors of data. It’s particularly powerful for tasks like image and speech recognition.
- Natural Language Processing (NLP): Enables machines to understand and respond to human language.
ChatGPT’s AI is collectively known as Generative AI. Generative AI produces new data that resembles a given set of training data. In other words, it guesses the next word. Generative AI can produce content, whether it’s text, images, music, or other forms of data, that wasn’t in their original training set but is consistent with the patterns they learned from that data.
Summary: ChatGPT for safety and security heads
Directly or indirectly, AI will impact every part of our lives. From completing log books to strategic decisions. Avoid it being something that ‘happens to you’ and guide its direction; start with ChatGPT and/or, introduce SIRV.
