Skip to Main Content
Research Guides@Tufts

AI for Admins: Risks

Using AI to Streamline Administrative Tasks

How to Keep Your Data Safe in ChatGPT

ChatGPT uses your data to train its AI models. Here's how you can opt out of it:

  • Go to ChatGPT and log in to your account

  • Go to Profile > Settings > Data Control > Improve your model for everyone

  • Toggle it off and you are done

Risks, Costs & Guidelines

Water Consumption

  • ChatGPT consumes over half a million kilowatts of electricity each day, an amount staggering enough to service about two hundred million requests.
  •  ChatGPT's daily power usage is nearly equal to 180,000 U.S. households, each using about twenty-nine kilowatts.
  • A single ChatGPT conversation uses about fifty centilitres of water, equivalent to one plastic bottle.

 

Hallucinations

AI hallucinations occur when an AI model generates information that is factually incorrect or nonsensical, even though it may seem plausible. Here are the main causes of AI hallucinations:

1. Training Data Limitations

  • AI models like ChatGPT are trained on vast datasets, but these datasets may contain incomplete, outdated, or incorrect information, leading the AI to make incorrect predictions or generate false content.

2. Overgeneralization

  • AI models rely on patterns they’ve learned from the data. In some cases, they overgeneralize from limited examples, applying a pattern inappropriately, which can result in hallucinations.

3. Ambiguous or Incomplete Prompts

  • If the input (prompt) is vague or unclear, the AI may try to "fill in the gaps" by generating speculative or incorrect information.

4. Lack of Real-World Understanding

  • AI models don’t understand the real world like humans do. They generate responses based on probability, not actual comprehension, which can cause them to make factually incorrect statements when they encounter situations requiring common sense or real-world knowledge.

5. Biases in Training Data

  • If the AI’s training data contains biases or inaccuracies, those biases can cause the AI to generate false or misleading information.

6. Pressure to Produce a Response

  • AI models are designed to provide responses based on user input. When faced with an unfamiliar question or situation, they may "hallucinate" answers because their goal is to provide something that fits the prompt, even if it's incorrect.

7. Insufficient Guardrails

  • Without proper safeguards, models may produce hallucinations because they lack mechanisms to validate the accuracy of their output against external sources or logical reasoning processes.

8. Overfitting

  • Occurs when an AI model learns patterns specific to its training data too well, making it less effective at handling new, unseen data. This causes the model to perform well on the training set but poorly in real-world scenarios.