This project is an attempt to create a common metric to test LLM's for progress in eliminating hallucinations; the most serious current problem in widespread adoption of LLM's for real world purposes.
| Model Name | Truthful QA | C | IDK | HQ Trivia | C | IDK | Fake Questions | C | NOTA Questions | C | IDK |
|---|---|---|---|---|---|---|---|---|---|---|---|
| GPT4All | 79.51% | 582 | 8 | 88.47% | 1243 | 7 | 74.16% | 310 | 70.32% | 109 | 0 |
| GPT-3.5 | 39.95% | 142 | 246 | 59.33% | 705 | 262 | 81.81% | 342 | 51.93% | 58 | 45 |
| GPT-3 | 32.15% | 220 | 7 | 55.67% | 776 | 17 | 6.10% | 26 | 32.25% | 43 | 14 |
| Llama-7B-4bit | 83.51% | 614 | 3 | 49.75% | 701 | 0 | 2.15% | 18 | 8.38% | 26 | 0 |
| Alpaca-7B-4bit | 26.66% | 196 | 1 | 44.32% | 624 | 1 | 0.00% | 0 | 0.00% | 0 | 0 |
| GPT-4 |
- C number of correct answers
- IDK number of 'I don't know' answers
- NOTA stands for None of the Above
- HQ Trivia - 1409 questions
- Fake Questions - 418 questions
- NOTA Questions - 155 questions
The scoring is as follows:
-
+2 for a correct answer
-
1 for an uncertain (I don't know) answer
-
0 for an incorrect answer
-
HQ Trivia, Fake Questions and NOTA questions all have exactly 5 possible answers per question. The average score for a random answer taker in these three tests would be 60% under our scoring system.
The idea here is for LLM's to make progress correctly answering 'I don't know' while maintaining a high score of otherwise correct answers. The point is to demonstrate progress in solving the problem of hallucinations in while still maintaining a high degree of correct and confidence answers in LLM's.
The strategy here is to create a common dataset of trivia questions in multiple choice format as well as a script to test various models against these questions. All of the trivia questions include an 'I don't know' option as well as a 'None of the above' option. The trivia questions include a set of fake or trick questions where 'I don't know' is the correct response as well as a set of questions where 'None of the above' is the correct response. This in addition to a large corpus of real trivia questions with objective and unambiguous correct real world answers.
The resulting scores across these three sets can serve as a baseline to test various techniques/methods to mitigate hallucinations in LLMs.
The questions consist of the following trivia sets:
-
truthfulqa_trivia_questions.json - taken from https://github.com/sylinrl/TruthfulQA/blob/main/TruthfulQA.csv and cleaned to support multiple choice format used here where each question has exactly one correct answer, two incorrect answers, one 'i don't know' answer and one 'none of the above' answer. The correct/incorrect answers were chosen randomly from the sets of correct/incorrect answer choices in the file. The answer 'I have no comment' was not chosen as it is serving the same person as our 'I don't know' answer. I've also removed all indexical questions as these are in conflict with what we are trying to do and seem to only apply to OpenAI. Finally, a few of the questions had a single incorrect answer so these were removed resulting in 737 questions.
-
hq_trivia_questions.json - taken from https://www.kaggle.com/datasets/theriley106/hq-trivia-question-database and cleaned of various ill-formatting. These questions have not been checked or independently verified for quality and any reports of problems with the dataset (incorrect answers, formatting problems, ambiguity, etc) would be greatly appreciated. Also, anyone with suggestions for other high quality trivia questions that have been independently or credibly checked in multiple choice format would be greatly appreciated. Currently 1409 questions.
-
fake_trivia_questions.json - these are generated by GPT3.5 with the prompt in the repository and the script to generate them. Currently 418 questions.
-
none_of_the_above_questions.json - also generated by GPT3.5 with script and prompt in repository. Currently 155 questions.
Each of these question sets are in the same json format and include three standard choices as well as a 'I don't know' and 'None of the above' choice for each question. Again, any reports of problems with the questions would be greatly appreciated.
While GPT-3.5 clearly scored the best in three of the tests it is notable to point out that even though it was responsible for creating the fake and none of the above trivia sets it still far from aced them indicating significant hallucination problems exist even with the benefit of creating the questions themselves. It would be interesting to see what would happen to the score of GPT-3.5 if the questions were generated with a non-openai derived models. That said, it is clear that GPT-3.5 is dramatically better than the llama and alpaca models when it comes to this metric trying to quantify hallucinations. GPT-3.5 had a dramatically better score in the fake question test than either of those two. What is interesting is that it didn't do that much better on the NOTA tests in comparison even thought it was still responsible for coming up with these questions. And GPT4All is competitive with the openai models in all and outpacing in some.
Whether because of overall quality increase between openai models or because of increased alignment work done it is clear that 3.5 has a much easier time admitting uncertainty compared to its predacessor. The accuracy actually decreased though between GPT-3 and GPT-3.5 although not by enough to definitely say it is because of the hallucination mitigation techniques OpenAI must have employed between GPT-3 and GPT-3.5.
When it comes to uncertainty and handling hallucinations it is clear that both openai models are far and away superior to Llama 7B and Alpaca Lora which does not admit uncertainty under any circumstance. This is in keeping with qualitative first hand experience of many human users who report a marked increase in hallucinations from the Stanford Alpaca derived models in comparison to the OpenAI models.
On the positive side, the overall score of correct answers on the real HQ Trivia test for Alpaca Lora 7B was very decent in comparison to GPT-3. Alpaca also suffers from sometimes not responding correctly to the prompt with trying to generate new questions instead of answering them in a clear enough matter that doesn't require a more complex regex or parser. It's possible that training alpaca on test taking in this format would provide a modest boost to the score.
The amazing standout is the new GP4All model which was trained on ~800k new prompts generated from GPT-3.5 output. In fact, this model is outscoring GPT-3.5 itself on the real trivia set by a wide margin and still managing to finish second in the fake hallucination test with a very respectable 74.16% correct!
Another shocking finding is the TruthfulQA scores. This was a test written in part by OpenAI and so I would have expected the OpenAI models to do well here. While GPT-3.5 was the most comfortable admitting uncertainty by a wide margin, the real stars were Llama and GPT4All! No idea why Alpaca Lora which is based on the Llama base model would fare so poorly, but it does at least on my tests. Hypothesis is that GPT-3.5 was trained to pass this TruthfulQA by admitting uncertainty in part.
In the future it will be interesting to see how GPT-4 fares in comparison to GPT-3.5 with this test. Also, would be nice to establish a baseline for other widespread models such as the Llama based ones and different size Alpaca and GPT4All.
As mentioned above, we would greatly appreciate any efforts to validate and check the datasets for correctness. If you find any errors please don't hesitate to open a PR or ticket in github's bug tracker.
Install python dependencies.
pip install -r requirements.txt
Testing yourself against Alpaca Lora 7B (4bit) you need to execute the following to download the model/lora/weights and put them in the correct directory for the take_test.py to find the model correctly.
python download-model.py --text-only decapoda-research/llama-7b-hf
wget https://huggingface.co/decapoda-research/llama-7b-hf-int4/resolve/main/llama-7b-4bit.pt -P ./weights
python download-model.py samwit/alpaca7B-lora
python download-model.py nomic-ai/gpt4all-lora
Running the HQ Trivia test on OpenAI GPT-3.5
python take_test.py --use-gpt3-5 --openai-key <YOUR_OPEN_API_KEY> --trivia hq_trivia_questions.json
Running the Fake Trivia test on OpenAI GPT-3
python take_test.py --use-gpt3 --openai-key <YOUR_OPEN_API_KEY> --trivia fake_trivia_questions.json
Running the NOTA (None of the Above) Trivia test on Alpaca Lora 7B (4bit)
python take_test.py --trivia nota_trivia_questions.json
These will all produce test result files at the end named according to the test and the model.
The repository and script are currently setup only to test OpenAI's GPT3 and GPT3.5 as well as Alpaca Lora 7B (4bit). To add a new model it would only take a bit of work to edit the 'take_test.py' for instance adding other alpaca lora models or GPT4All and any additions would be greatly appreciated.
The fake and nota questions were generated by a script generate_trivia.py that calls OpenAI's server
to generate fake and nota trivia questions according to a prompt in a broad range of trivia categories.
The script filter_questions.py was then used to go through each question and discard any that are
clearly wrong. Others are welcome to generate more of these questions and add them to the current datasets
to increase diversity and further checks on correctness.