- The Week in AI
- Posts
- Lawyer in trouble after ChatGPT makes up cases
Lawyer in trouble after ChatGPT makes up cases
Deep dive into ChatGPT & prompt engineering, Risk of extinction from AI, Japan's copyright vs AI ruling, ChatGPT's affect on learning, Big tech AI releases and more...
Featured Story:
A lawyer submitted a document to court where he used ChatGPT to come up with citations of previous decisions relevant to the case e.g, Varghese v. China Southern Airlines, Shaboon v. Egyptair, Petersen v. Iran Air.
The problem?
The cited cases were made up by ChatGPT and did not exist.
On being asked, ChatGPT (incorrectly) confirmed that cases were real and that they can even be found in reputable legal databases. [link]
Key Insight: ChatGPT is not an infallible source of truth. It is trained on tremendous amounts of data, which makes it give correct answers for many things. But other times, it can generate answers that sound right but are completely incorrect.
Other updates to keep you ahead of the curve:
Top AI scientists and leaders have issued this statement [link]“Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war”
Non-profit eating disorder helpline replaced 6 staff members and 200 volunteers with an AI chatbot. After it started giving harmful advice, it “fired” the AI too. [link]Insight: Every leader will need to embrace AI, but they must do so cautiously. They need to assess the risks and consider whether it’s better to augment human roles with AI rather than replacing them.
According to a survey by Pew Research Centre, only 14% US adults had used ChatGPT. Of those, 15% found it “extremely useful” and 20% “very useful.”Insight: Despite all advances, we’re very early in generative AI.
Japan has made a decision to exempt AI training data from copyright regulations [link]. Insight: Artists and writers have been calling for regulations to prevent AI companies to use their proprietary data. But this comes at a cost of slowing down AI advancement. Japan’s decision shows its clear prioritisation of AI.
According to a survey by Intelligent.com, a whopping 85% of high school and college students who used both ChatGPT and traditional tutoring sessions believe that ChatGPT is more effective for learning. [link]Insight: Learning will look increasingly different compared to before. Tutors can prepare for this by exploring ways to integrate AI technologies into their teaching methods to enhance engagement and effectiveness.
Meta will be bringing AI chatbots for messenger & WhatsApp and AI for photo editing in Instagram. [link] LinkedIn introduced an ad copy suggestion tool to enhance advertising campaigns using OpenAI’s model. [link]
AI Toolbox: ChatGPT & Prompt engineering (Part 1)
Most likely you would’ve heard about ChatGPT by now. It’s a chatbot that allows you to have natural, and engaging conversations. It has become really popular because of its intelligence and ability to answer questions that seemed impossible for chat models before it.
What can I use it for?
The most popular use cases for ChatGPT are -
General questions - recipes, travel itineraries, fitness workout planning etc.
Language tasks like generating creative writing, translating languages, generating content for blogs etc.,
Specific areas like programming tasks, helping in learning for students, and even mental health support.
Prompt engineering
ChatGPT will generally give you the answers that you’re looking for. But sometimes it just doesn’t work if you don’t ask your question the way “it likes”.The skill of phrasing the question well to get best answers is known as prompt engineering.
Here are some best ways you can ask questions to ChatGPT -
1. Include detailed explanation to tell ChatGPT what you want exactly
If you’re asking ChatGPT something but are not happy with the response, make the question detailed explaining exactly what you need. For e.g.,
Initial Prompt: Make a 4 day plan for a trip to Italy.Better Prompt with specific details: Make a 4 day plan for a trip to Italy. It should prioritise food, wine, experiencing the local culture and visiting spots with natural beauty. It must include a visit to Duomo in Milan.
2. Provide examples to teach ChatGPT what you want
If ChatGPT isn’t able to answer how you want it to, try providing it 2-3 examples that it can learn from and then ask your question.
Initial Prompt that gives a useless answer 😅: (In image below)

Better Prompt with examples that gives correct answer: (In image below)

3. For complex tasks, provide it smaller steps that can help it provide correct answers.
Consider two ways of asking the same task below -
Initial Prompt: Summarise these meeting notes in 2 paragraphs in SpanishBetter Prompt with smaller steps: You will be provided with some meeting notes - Summarise the meeting to keep only the important parts- Convert this summary from English to Spanish
When I tried the initial prompt, it incorrectly converted all meeting notes to Spanish without summarisation. But when I broke it down to two simple steps as shown, it gave exactly what I wanted. You can check the two conversations by clicking the links.