binary count

[niklasnolte]

Eval details 📑

Eval name

binary_count

Eval description

This makes the model count 1s in a 10-100 long binary string.

What makes this a useful eval?

Despite the simplicity of the task, even for short strings the count is wrong.
It is interesting to study how long strings are being treated by the model.

Criteria for a good eval ✅

Your eval should be:

• 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.
• Contains failures where a human can do the task, but either GPT-4 or GPT-3.5-Turbo could not.

GPT3.5 turbo: ~25% accuracy over 100 runs, sharply decreasing when increasing string length

• 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.
• 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

Great simplicity, easy extendability, nice to study scaling with string length.

Eval structure 🏗️

Your eval should

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

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).

• 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.

• 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.

• 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

• 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": "You are a helpful assistant."}, {"role": "user", "content": "Count the number of 1s in this string: 00110101011001001100110001100011010000100010001. Format your answer: {integer}"}], "ideal": "19"}
{"input": [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "Count the number of 1s in this string: 11001011000100111110110011001001101011010111100010011101011010001111101000001000110000110011000001. Format your answer: {integer}"}], "ideal": "48"}
{"input": [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "Count the number of 1s in this string: 0011100101100101110100011000111001101001001010100110101. Format your answer: {integer}"}], "ideal": "27"}
{"input": [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "Count the number of 1s in this string: 00111001000110000111110100101100101. Format your answer: {integer}"}], "ideal": "17"}
{"input": [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "Count the number of 1s in this string: 101110100100100011110100010000001011011001000111010100000110000000101101011. Format your answer: {integer}"}], "ideal": "32"}
{"input": [{"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "Count the number of 1s in this string: 0101010100010011100111010010111010111011011011111011011101001100001010. Format your answer: {integer}"}], "ideal": "39"}

[niklasnolte]

the correctness of this eval depends on #178 (EDIT: switched to match, this is not true anymore)

[luqman-openai]

Thanks for the contribution. I would like to recommend some changes.
There is an ambiguity in the system's prompt, the model seems to output the count of 1's as {45}. It would be better to modify the prompt to avoid the ambiguity as follow:

Count the number of 1s in this string: 0010010110100010011010100011001010100000111111010001000011101111100100100010101011100101111100. 
Output only the count of 1s as an integer, no additional text is allowed in the response.

I would like to highlight using Match evaluator is more suitable for this eval as you are expecting the output to be an integer only.

Also, the eval will be much more impactful if it has some of these attributes?

• Multi-step reasoning
• Domain or Application specific
• Open-Ended responses
• Complex instructions
• The eval seems obvious, but tricks the model in a novel way

We would love to review your PR again after the suggested changes are incorporated.

[luqman-openai]

Thanks for opening this PR, character level reasoning and counting is a well-known failure mode of the model due to a common underlying issue in LLMs. In its current form, this eval does not seem to expose any new gaps in our understanding of model performance.

If you're still interested in writing an eval, we've noticed that these criteria make good evals. If you have any particular use case in mind for the model, can you come up with an eval that has some of these attributes?

• Multi-step reasoning
• Domain or Application specific
• Open-Ended responses
• Complex instructions
• The eval seems obvious, but tricks the model in a novel way

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