New eval for math contests (AMC 10/12)

[emptycrown]

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

Math Contests

Eval description

The eval tests models on AMC 10/12 problems from the last 10 years. The AMCs (American Math Contests) are popular national math contests with multiple-choice answers. Questions that involve figures or images are not included in the samples. Math formulas in both the questions and choices are written out in LaTeX.

What makes this a useful eval?

These math problems test advanced reasoning capabilities. It is not easy for even a human to solve many of these problems, especially those later in the test. For each exam, there are 2 tests per year, each with 25 questions, making this a clean and structured dataset to run evals on. Furthermore, the accuracy of the response can be easily and quantifiably measured since each answer is one of 5 choices: A, B, C, D, E. Finally, note that gpt-3.5-turbo is remarkably bad at solving even the easier problems. It tends to generate a lot of text to explain its reasoning, but it's not logically sound and usually gets the wrong answer.

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

Contest math is a fairly unique test of logical thinking and chain-of-thought, with multiple steps. Math was mentioned as one of the areas in which evals were sought.

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
• [ ] (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":"Your job to solve math problems from a national math contest. Each problem is multiple choice with five possible answers: A, B, C, D, E."},{"role":"user","content":"PROBLEM: 'A choir director must select a group of singers from among his '$6$' tenors and '$8$' basses. The only requirements are that the difference between the number of tenors and basses must be a multiple of '$4$', and the group must have at least one singer. Let '$N$' be the number of different groups that could be selected. What is the remainder when '$N$' is divided by '$100$'?\\n'\n=======\nCHOICES:\n[A]: 47\n[B]: 48\n[C]: 83\n[D]: 95\n[E]: 96\n"}],"ideal":"D"}
{"input":[{"role":"system","content":"Your job to solve math problems from a national math contest. Each problem is multiple choice with five possible answers: A, B, C, D, E."},{"role":"user","content":"PROBLEM: 'A laser is placed at the point '$(3,5)$'. The laser beam travels in a straight line. Larry wants the beam to hit and bounce off the '$y$'-axis, then hit and bounce off the '$x$'-axis, then hit the point '$(7,5)$'. What is the total distance the beam will travel along this path?\\n'\n=======\nCHOICES:\n[A]: 2\\sqrt{10}\n[B]: 5\\sqrt2\n[C]: 10\\sqrt2\n[D]: 15\\sqrt2\n[E]: 10\\sqrt5\n"}],"ideal":"C"}
{"input":[{"role":"system","content":"Your job to solve math problems from a national math contest. Each problem is multiple choice with five possible answers: A, B, C, D, E."},{"role":"user","content":"PROBLEM: 'Semicircle '$\\Gamma$' has diameter '$\\overline{AB}$' of length '$14$'. Circle '$\\Omega$' lies tangent to '$\\overline{AB}$' at a point '$P$' and intersects '$\\Gamma$' at points '$Q$' and '$R$'. If '$QR=3\\sqrt3$' and '$\\angle QPR=60^\\circ$', then the area of '$\\triangle PQR$' equals '$\\tfrac{a\\sqrt{b}}{c}$', where '$a$' and '$c$' are relatively prime positive integers, and '$b$' is a positive integer not divisible by the square of any prime. What is '$a+b+c$'?\\n'\n=======\nCHOICES:\n[A]: 110\n[B]: 114\n[C]: 118\n[D]: 122\n[E]: 126\n"}],"ideal":"D"}
{"input":[{"role":"system","content":"Your job to solve math problems from a national math contest. Each problem is multiple choice with five possible answers: A, B, C, D, E."},{"role":"user","content":"PROBLEM: 'Let '$d(n)$' denote the number of positive integers that divide '$n$', including '$1$' and '$n$'. For example, '$d(1)=1,d(2)=2,$' and '$d(12)=6$'. (This function is known as the divisor function.) Let'\\[f(n)=\\frac{d(n)}{\\sqrt [3]n}.\\]'There is a unique positive integer '$N$' such that '$f(N)>f(n)$' for all positive integers '$n\\ne N$'. What is the sum of the digits of '$N?$'\\n'\n=======\nCHOICES:\n[A]: 5\n[B]: 6\n[C]: 7\n[D]: 8\n[E]: 9\n"}],"ideal":"E"}

[Ein-Tim]

Please add some examples of evals to the respective field in the PR description.

[emptycrown]

Please add some examples of evals to the respective field in the PR description.

Just edited the original post!

[usama-openai]

Closing the PR due to inactivity; please feel free to reopen if you get a chance to address comments.