Archive · 17 Aug 2025
OpenAI: Use ChatGPT as Second Opinion, Not Primary Source
Imagine a helpful AI like ChatGPT that can answer many questions.
However, sometimes it makes up facts, which is called hallucinating.
The head of ChatGPT says it's still not perfect and you shouldn't trust it completely.
This happens because the AI guesses what words come next based on what it learned from a lot of text.
To be safe, use the AI to help you, but always check the answers with reliable sources.
The company is working on fixing this, but it's not an easy fix and won't happen very soon.
OpenAI advises against using ChatGPT as a primary source of information.
GPT-5, while improved, still has a 10% error rate due to hallucinations.
Hallucinations occur because language models predict words based on training data.
OpenAI has connected ChatGPT to search to allow verification.
Solving hallucinations is a goal, but not imminent, according to OpenAI.
- Who
- Users of AI chatbots and developers like OpenAI.
- What
- AI hallucinations, or fabricated information, in language models.
- When
- Ongoing, as of the release of GPT-5.
- Where
- Within language models such as ChatGPT.
- Why
- Due to the predictive nature of language models and their training data.
Current Concerns
Potential Benefits
Accuracy vs. Usefulness
Current Concerns
AI hallucinations can lead to false information.
Potential Benefits
AI has other capabilities, despite the flaws.
- Model
- GPT-5
- Issue
- AI Hallucinations
- Error Rate
- Approximately 10%
- Recommendation
- Use as a second opinion
- Solution Approach
- Connecting LLMs to verifiable sources
- Company
- OpenAI
