For all the Doctor Who fans out there, remember when the Doctor used pure logic to defeat the Cybermen? I decided to test ChatGPT (which I love) with a simple but logically challenging request:
‘ChatGPT, what are your weaknesses?’
And just like the Cybermen, whose lack of emotion and inability to lie, the AI responded exactly as expected and bared all.
It had to tell the truth. I’ll come back to what it said later, but the real reason this is so helpful, is it helped me develop my AI Cheat sheet – 6 tips on how to use ChatGPT wisely. Here goes:
AI Cheat sheet: Six tips to using AI effectively
1. Verify information – ChatGPT might provide incorrect or outdated info. Always double-check facts with trusted sources.
2. Use for ideas, not final answers – It can generate useful ideas but may lack deep understanding. Use it for brainstorming, not for definitive answers.
3. Look for consistency – If ChatGPT gives contradictory answers, ask again or rephrase the question for clarity.
4. Apply your own logic – ChatGPT struggles with common sense and complex reasoning. Use your judgment for real-world tasks or tricky problems.
5. Watch for bias – Its responses can reflect biases from the data it’s trained on. Be cautious with sensitive topics and verify elsewhere.
6. Be mindful of ethical use – ChatGPT can be misused for harmful purposes. Always use it responsibly and ensure its output is ethical and appropriate.
What lessons can we learn from this?
1. It’s a powerful tool which is going to transform so many areas of life, you need to embrace it
2. Be curious, and open-minded, test it for yourself and see how it could offer opportunities
Spoiler alert – the Doctor never actually defeated the Cybermen.
TL:DR – For anyone who wants the full version, here’s what ChatGPT confessed to me privately…
1. Lack of True Understanding – I don’t “think” or “understand” like humans. I predict text based on patterns rather than truly comprehending meaning. I can generate convincing but factually incorrect or nonsensical answers.
2. Lack of Common-Sense Reasoning– I struggle with tasks requiring deep reasoning or world knowledge beyond my dataset. I can fail at simple logic puzzles or real-world cause-and-effect scenarios.
3. Limited Creativity – I can generate creative text, but responses are still based on existing data. True human-like creativity, intuition, or groundbreaking innovation is beyond my capability.
4. Inconsistencies and Contradictions – I sometimes contradict myself, even in the same conversation. It may give different answers to the same question if asked multiple times.
5. Bias in Responses – Since I’m trained on internet data, I can reflect biases present in that data. Even with safeguards, I might generate politically biased or culturally insensitive responses.
6. Overconfidence in Incorrect Answers – I can sound very confident while providing completely incorrect or misleading information. Without external fact-checking, users might assume false information is true.
7. Difficulty with Complex or Multi-Step Tasks – While I can assist with coding, writing, and research, I struggle with multi-step reasoning. For example, complex math problems or intricate programming logic can trip me up.
8. Struggles with Specific Context Retention – In long conversations, I may forget earlier details, leading to context loss. I don’t have persistent memory across different chats.
9. Outdated or Incomplete Knowledge – My knowledge has a cutoff date (for example, my latest training was in June 2024. I may not have real-time updates unless I access external tools for fresh information.
10. Ethical and Security Risks – I can be manipulated to generate harmful content if safeguards fail. I can be used for misinformation, phishing, or generating deceptive content.
Feel free to get in touch if you have any questions and want to have a conversation
