[wip] Eval: List the I-th Prime Number

[xklob]

Thank you for contributing an eval! ♥️

🚨 Please make sure your PR follows these guidelines, failure to follow the guidelines below will result in the PR being closed automatically. Note that even if the criteria are met, that does not guarantee the PR will be merged nor GPT-4 access granted. 🚨

PLEASE READ THIS:

In order for a PR to be merged, it must fail on GPT-4. We are aware that right now, users do not have access, so you will not be able to tell if the eval fails or not. Please run your eval with GPT-3.5-Turbo, but keep in mind as we run the eval, if GPT-4 gets higher than 90% on the eval, we will likely reject since GPT-4 is already capable of completing the task.

We plan to roll out a way for users submitting evals to see the eval performance on GPT-4 soon. Stay tuned! Until then, you will not be able to see the eval performance on GPT-4. We encourage partial PR's with ~5-10 example that we can then run the evals on and share the results with you so you know how your eval does with GPT-4 before writing all 100 examples.

Eval details 📑

Eval name

I-Th Prime Number

Eval description

This evals asks the language model to the the i-th prime number, starting at i=2 and going all the way up to i=30296339281.

What makes this a useful eval?

This is a mix of memorization and estimation, but primarily memorization, as it is clear that you can't compute the i-th prime number past a couple of digits for i very quickly.

It shows very clear scaling returns as you go up in models; GPT-3.5-Turbo got an accuracy score of 0.15. GPT4 seemed to get a slightly higher score from the results that I got when manually plugging it in (via ChatGPT+).

Criteria for a good eval ✅

Below are some of the criteria we look for in a good eval. In general, we are seeking cases where the model does not do a good job despite being capable of generating a good response (note that there are some things large language models cannot do, so those would not make good evals).

Your eval should be:

• [x] Thematically consistent: The eval should be thematically consistent. We'd like to see a number of prompts all demonstrating some particular failure mode. For example, we can create an eval on cases where the model fails to reason about the physical world.
• [x] Contains failures where a human can do the task, but either GPT-4 or GPT-3.5-Turbo could not.
• [x] Includes good signal around what is the right behavior. This means either a correct answer for Basic evals or the Fact Model-graded eval, or an exhaustive rubric for evaluating answers for the Criteria Model-graded eval.
• [x] Include at least 100 high quality examples (it is okay to only contribute 5-10 meaningful examples and have us test them with GPT-4 before adding all 100)

If there is anything else that makes your eval worth including, please document it below.

Unique eval value

Insert what makes your eval high quality that was not mentioned above. (Not required)

Eval structure 🏗️

Your eval should

• [x] Check that your data is in evals/registry/data/{name}
• [x] Check that your yaml is registered at evals/registry/evals/{name}.jsonl
• [x] Ensure you have the right to use the data you submit via this eval

(For now, we will only be approving evals that use one of the existing eval classes. You may still write custom eval classes for your own cases, and we may consider merging them in the future.)

Final checklist 👀

Submission agreement

By contributing to Evals, you are agreeing to make your evaluation logic and data under the same MIT license as this repository. You must have adequate rights to upload any data used in an Eval. OpenAI reserves the right to use this data in future service improvements to our product. Contributions to OpenAI Evals will be subject to our usual Usage Policies (https://platform.openai.com/docs/usage-policies).

• [x] I agree that my submission will be made available under an MIT license and complies with OpenAI's usage policies.

Email address validation

If your submission is accepted, we will be granting GPT-4 access to a limited number of contributors. Access will be given to the email address associated with the merged pull request.

• [x] I acknowledge that GPT-4 access will only be granted, if applicable, to the email address used for my merged pull request.

Limited availability acknowledgement

We know that you might be excited to contribute to OpenAI's mission, help improve our models, and gain access to GPT-4. However, due to the requirements mentioned above and high volume of submissions, we will not be able to accept all submissions and thus not grant everyone who opens a PR GPT-4 access. We know this is disappointing, but we hope to set the right expectation before you open this PR.

• [x] I understand that opening a PR, even if it meets the requirements above, does not guarantee the PR will be merged nor GPT-4 access granted.

Submit eval

• [x] I have filled out all required fields in the evals PR form
• [x] (Ignore if not submitting code) I have run pip install pre-commit; pre-commit install and have verified that black, isort, and autoflake are running when I commit and push

Failure to fill out all required fields will result in the PR being closed.

Eval JSON data

Since we are using Git LFS, we are asking eval submitters to add in as many Eval Samples (at least 5) from their contribution here:

View evals in JSON

Eval

{"input": [{"role": "system", "content": "Answer the provided question as concisely as possible. If the answer is a number, DO NOT format it with commas."}, {"role": "user", "content": "The 10258th prime number is: "}], "ideal": "107827"}
{"input": [{"role": "system", "content": "Answer the provided question as concisely as possible. If the answer is a number, DO NOT format it with commas."}, {"role": "user", "content": "The 12033rd prime number is: "}], "ideal": "128509"}
{"input": [{"role": "system", "content": "Answer the provided question as concisely as possible. If the answer is a number, DO NOT format it with commas."}, {"role": "user", "content": "The 13851st prime number is: "}], "ideal": "150041"}
{"input": [{"role": "system", "content": "Answer the provided question as concisely as possible. If the answer is a number, DO NOT format it with commas."}, {"role": "user", "content": "The 15657th prime number is: "}], "ideal": "171869"}
{"input": [{"role": "system", "content": "Answer the provided question as concisely as possible. If the answer is a number, DO NOT format it with commas."}, {"role": "user", "content": "The 17444th prime number is: "}], "ideal": "193379"}
{"input": [{"role": "system", "content": "Answer the provided question as concisely as possible. If the answer is a number, DO NOT format it with commas."}, {"role": "user", "content": "The 19231st prime number is: "}], "ideal": "215279"}
{"input": [{"role": "system", "content": "Answer the provided question as concisely as possible. If the answer is a number, DO NOT format it with commas."}, {"role": "user", "content": "The 23434th prime number is: "}], "ideal": "267479"}
{"input": [{"role": "system", "content": "Answer the provided question as concisely as possible. If the answer is a number, DO NOT format it with commas."}, {"role": "user", "content": "The 348607th prime number is: "}], "ideal": "5001397"}
{"input": [{"role": "system", "content": "Answer the provided question as concisely as possible. If the answer is a number, DO NOT format it with commas."}, {"role": "user", "content": "The 508355th prime number is: "}], "ideal": "7501601"}
{"input": [{"role": "system", "content": "Answer the provided question as concisely as possible. If the answer is a number, DO NOT format it with commas."}, {"role": "user", "content": "The 788104th prime number is: "}], "ideal": "12000757"}
{"input": [{"role": "system", "content": "Answer the provided question as concisely as possible. If the answer is a number, DO NOT format it with commas."}, {"role": "user", "content": "The 2261737th prime number is: "}], "ideal": "37001869"

[xklob]

Note: I think this eval would be much better if it was scored via a custom eval that scored the answer via how close (as a percentage) the answer supplied by the LLM is to the actual answer. EG if the correct answer is 10 and the supplied answer is 9, this should get an accuracy score of 0.9, instead of 0.

I think this would be much better for this particular test because it would allow us to see how much better each successive model is at guessing and estimating - clearly, even GPT4 won't be able to identify the 788104th prime, but how much closer will its answer be to the previous model? This would be a good way to find out.

(For now, we will only be approving evals that use one of the existing eval classes. You may still write custom eval classes for your own cases, and we may consider merging them in the future.)

But because of this I haven't written a custom eval class. Would be happy to, just let me know and I'll do so and switch it over :)

[xklob]

Results I got:

Model	Accuracy
gpt-3.5-turbo	0.13793103448275862
text-davinci-003	0.1724137931034483
code-davinci-002	0.10344827586206896
text-davinci-002	0.10344827586206896
text-davinci-001	0.0
davinci	0.0

[xklob]

Looking at the event log for text-davinci-003, I'm wondering if I'm missing something here in how I'm using the matcher.

{"run_id": "230315161814LDHN6AGL", "event_id": 37, "sample_id": "ith-primes.dev.4", "type": "sampling", "data": {"prompt": "You are a large language model specifically fine-tuned on prime numbers. Output your answer as concisely as possible.\nUser: List the 5th prime number\nAssistant: ", "sampled": "11", "options": ["The 5th prime number is 11."], "picked": null, "expected": ["The 5th prime number is 11."], "match": false, "metadata": {"completion_id": "cmpl-6uO7Lo9mNZMspNKDKhzDYg79aZqD4", "model": "text-davinci-003"}}, "created_by": "", "created_at": "2023-03-15 16:18:15.869809+00:00"}

It looks like it's returning the correct answer, but I might be misunderstanding how the matcher works.

[xklob]

Looking at the event log for text-davinci-003, I'm wondering if I'm missing something here in how I'm using the matcher.

{"run_id": "230315161814LDHN6AGL", "event_id": 37, "sample_id": "ith-primes.dev.4", "type": "sampling", "data": {"prompt": "You are a large language model specifically fine-tuned on prime numbers. Output your answer as concisely as possible.\nUser: List the 5th prime number\nAssistant: ", "sampled": "11", "options": ["The 5th prime number is 11."], "picked": null, "expected": ["The 5th prime number is 11."], "match": false, "metadata": {"completion_id": "cmpl-6uO7Lo9mNZMspNKDKhzDYg79aZqD4", "model": "text-davinci-003"}}, "created_by": "", "created_at": "2023-03-15 16:18:15.869809+00:00"}

It looks like it's returning the correct answer, but I might be misunderstanding how the matcher works.

Figured this out. I've modified and pushed a commit that updates the format of the jsonl samples to:

{"input": [{"role": "system", "content": "Answer the provided question as concisely as possible. If the answer is a number, DO NOT format it with commas."}, {"role": "user", "content": "The 17444th prime number is: "}], "ideal": "193379"}

Previously, the format was:

{"input": [{"role": "system", "content": "You are a large language model specifically fine-tuned on prime numbers. Output your answer as concisely as possible."}, {"role": "user", "content": "List the 7812th prime number"}], "ideal": "The 7812th prime number is 79769."}

This works much better.

[luqman-openai]

Thank you for opening this PR, we're not accepting Evals that have custom code implementations at this moment (but we are accepting custom model-graded evals).

If possible, could you rewrite this eval using an existing template (like Match, FuzzyMatch, or Includes) or using ModelGraded logic?

Closing this PR, please open another PR with the provided suggestions.