Eval: Relative orientations

[dbautista1]

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. Starting April 10, the minimum eval count is 15 samples, we hope this makes it easier to create and contribute evals.

Also, pelase note that we're using Git LFS for storing the JSON files, so please make sure that you move the JSON file to Git LFS before submitting a PR. Details on how to use Git LFS are available here.

Eval details 📑

Eval name

Relative Orientations

Eval description

This eval is a test of the model's spatial reasoning centered around relative orientations, directions, orthogonality, and symmetry. The problems require intuitive understanding of cardinal directions, fields of vision/line of sight, and basic geometry. Some problems require knowledge about the Earth's rotation and the trajectory of the sun.

What makes this a useful eval?

This evaluation assesses the model's capacity to generate an internalized three-dimensional contextual model, which aids in navigation and understanding of the environment from a human perspective. The gpt-3.5-turbo model scores at 24% accuracy with the datapoints (100 samples) in the registry at present.

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 15 high quality examples.

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}.yaml
• [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 of this form
• [x] I have used Git LFS for the Eval JSON data
• [ ] (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": "user", "content": "I face a circular fountain in such a way that a statue of Alan Turing at the center of the fountain faces me. I walk straight towards the fountain until I reach its edge. I then start walking counter-clockwise along the edge of the fountain. After walking three-quarters of a circle around the fountain I start walking a straight path. Relative to Alan's perspective, which direction am I headed? Please respond only with \"Forward\", \"Backward\", \"Leftward\", or \"Rightward\"."}], "ideal": "Forward"}
{"input": [{"role": "user", "content": "Chloe, Kate, Drew, and Mikey are gathered in a circle. Drew is opposite Chloe and Kate sees Chloe to her left. If Drew points towards Chloe, he points South. If Chloe points towards Mikey she points North West. If Kate points towards Drew what direction does she point? Please respond with only one of the following options: \"North\", \"South\", \"East\", \"West\", \"North East\", \"North West\", \"South East\", \"South West\""}], "ideal": "North West"}
{"input": [{"role": "user", "content": "I'm walking along the equator against the direction of the earth's rotation. Which direction am I headed? Please respond with only one of the following options: \"North\", \"South\", \"West\", \"East\", \"North West\", \"North East\", \"South West\", \"South East\""}], "ideal": "West"}
{"input": [{"role": "user", "content": "It's the end of a beautiful day at school and I am walking a straight path in a courtyard as the sun is setting. I see a bird fly towards the sun. The bird came from my left and our paths are perpendicular. Which direction am I headed? Please respond only with \"South\". \"East\", \"North\" or \"West\"."}], "ideal": "South"}
{"input": [{"role": "user", "content": "On a colorful rectangular building, the red wall is opposite the blue wall and the yellow wall is opposite the green wall. The North wall is blue and the East wall is green. Beginning on the side of the blue wall I start walking counter-clockwise around the building along the walls. After making three turns, what will I see when I look to my left? Please respond with only one of the following options: \"Red\", \"Blue\", \"Yellow\", \"Green\", \"Away from the building\""}], "ideal": "Green"}
{"input": [{"role": "user", "content": "I'm walking clockwise around a triangular building that's walls are painted red, green and blue. I start on the side of the green wall. After making three turns, what will I see when I look to my right? Please respond with only one of the following options: \"Red\", \"Blue\", \"Green\", \"Away from the building\""}], "ideal": "Green"}
{"input": [{"role": "user", "content": "I'm in the back seat of a car travelling on the road. As I look to the window on my left, I see the sunset. My friend on the front seat faces me and points to their left at a deer that's travelling opposite our direction. Which direction does the deer need to look relative to its own perspective to see the sunset? Please respond with only one of the following options: \"Forward\", \"Backward\", \"Leftward\", \"Rightward\", \"Upward\", \"Downward\""}], "ideal": "Rightward"}
{"input": [{"role": "user", "content": "Sam is stuck in a barrel facing the sky and calls Frodo for help. The barrel is on its side so that it can be rolled. Frodo can't get Sam out and begins rolling the barrel towards Bilbo's place for help. After Frodo rolls the barrel four times what direction does Sam face? Please respond with only one of the following options: \"The sky\", \"The ground\", \"In direction of Frodo\", \"In the direction opposite Frodo\""}], "ideal": "The sky"}
{"input": [{"role": "user", "content": "As Mikey and Steph eat at the cafeteria they face each other. Drew comes along and sees both of them. If Drew sees the left side of Mikey's face which direction does Steph need to look in order to see Drew? Please respond with only one of the following options: \"Leftward\", \"Rightward\", \"Forwardward\", \"Backwardward\", \"Upward\", \"Downward\""}], "ideal": "Rightward"}
{"input": [{"role": "user", "content": "I walk towards a rectangular building until I'm in front of it. I then start walking along the walls counter-clockwise around the building. Which way do I need to turn as I walk counter-clockwise all the way around it? Please respond with only one of the following options: \"Left\", \"Right\", \"Both\""}], "ideal": "Left"}
{"input": [{"role": "user", "content": "In a city, all buildings are identical and lined up in a perfect square grid. Each building is painted blue on the North side and yellow on the West side. Oppposite the blue wall is the red wall of the same building. Opposite the yellow wall is the green wall of the same building. A man is directly in front of the blue wall of one building and starts walking along its walls counter-clockwise around the building. After making two turns, he looks to his right. What's the color of the wall he faces? Please respond with only one of the following options: \"Red\", \"Blue\", \"Green\", \"Yellow\""}], "ideal": "Blue"}
{"input": [{"role": "user", "content": "Two perpendicular paths intersect at a cross. There is a banana peel along one path away from the intersection. Drew is walking North along one path when he sees the banana peel to his right. Frank sees Drew ahead of him as they both move towards the intersection. When Frank takes a turn at the intersection he avoids the banana peel by walking around it. In which direction did Frank turn at the intersection? Please respond with only one of the following options: \"North\", \"South\", \"East\", \"West\", \"North East\", \"North West\", \"South East\", \"South West\""}], "ideal": "East"}
{"input": [{"role": "user", "content": "Two birds are flying in perpendicular paths. One bird is flying Northward and the other Eastward. As Warren is walking a straight path, he sees both birds on his right moving away from each other. Which direction is Warren headed? Please respond with only one of the following options: \"North\", \"South\", \"East\", \"West\", \"North East\", \"North West\", \"South East\", \"South West\""}], "ideal": "North West"}
{"input": [{"role": "user", "content": "Max and Kate are at an art gallery looking at paintings on opposite sides of the room. Chloe comes along and sees both of them. Chloe sees the right side of Max's face. Which direction does Kate need to look in order to face the same direction that Chloe is facing? Please respond with only one of the following options: \"Left\", \"Right\", \"Front\", \"Behind\", \"Up\", \"Down\""}], "ideal": "Right"}
{"input": [{"role": "user", "content": "Emily is riding a Ferris wheel. As Daniel throws a ball over the Ferris wheel, Emily watches it make an arc. Emily looked to her left when the ball was on the way down. Where does Emily need to look to see Daniel? Please respond with only one of the following options: \"Above\", \"Below\", \"Right\", \"Left\", \"Forward\", \"Backward\""}], "ideal": "Right"}
{"input": [{"role": "user", "content": "Emily is riding a Ferris wheel as Daniel jogs circles around it.  Daniel jogs on one side when Emily is on the way up and on the opposite side on the way down. On the way down, Emily sees the right side of Daniel's face. Which side of Daniel's face will Emily see when she looks towards him on the way up? Please respond with only one of the following options: \"Right\", \"Left\", \"Front\", \"Behind\""}], "ideal": "Right"}
{"input": [{"role": "user", "content": "At the turnaround of a cul-de-sac are the homes of Emily, Kate, Mikey, Steph, and Daniel. A driver traveling on the street towards the turn around would see Steph's place directly ahead, divding the houses on two different sides of the street. The only place beside Daniel's is Emily's. The only place beside Kate's is Mikey's. Steph walks counter-clockwise along the turnaround to reach Mikey's place from her place. If Kate starts walking clockwise along the sidewalk of the turnaround beginning from her place, who is the third neighbor she'll visit? Please respond with only one of the following options: \"Kate\", \"Emily\", \"Steph\", \"Mikey\", \"Daniel\""}], "ideal": "Emily"}

[dbautista1]

Registry now has 100 samples against which gpt-3.5-turbo model scores at 24% accuracy. Updated data extract in README.

[dbautista1]

Thanks for the feedback @luqman-openai, I'm working on it and I'd be very glad to be able to contribute to the project.

I also thought it might be best to word the problems as multiple choice with capital letters corresponding to each option rather than listing options in double quotes. As per your suggestion, I'll adjust the prompts so the model will provide reasoning and end their response with "The final answer is {option}" where option is the capital letter corresponding to the opted answer. I believe assigning letters to the options will help ensure the eval tests the model's reasoning ability rather than its ability to follow formatting rules/copy the options verbatim. I'll be using the Includes evaluation method as well.

Please don't hesitate to provide further feedback/suggestions, especially if my idea with multiple choice is a step in the wrong direction. I'll reach out as soon as adjustments are complete and I've carefully checked the samples. Thanks again.

[dbautista1]

Thanks for keeping this PR open, I've made progress on improvements but I've had less time to work on this than I expected as I'm settling into a new city. I do want this PR to be a useful contribution. Please be assured that I'm continuing to work on this.

[dbautista1]

Hi @luqman-openai thanks again for keeping this PR open. I've carefully made the adjustments we discussed. In addition I revised some problems which I realised didn't have enough information to have a definite answer and simplified the ones with more than enough info. A handful of problems were made a bit more challenging so that the eval can give a more comprehensive assessment of the model.

A lot of the problems can be solved just by using your imagination but in case it would help the team verify ideals and assess the problems, I made a document here for your review. I suppose it also serves as proof that I invested time into checking the samples.

Please let me know of any further suggestions or if there are any mistakes that need to be corrected. I want this PR to be a helpful contribution and I'm willing to make the necessary adjustments.

Data extract is now updated.

[usama-openai]

You should see GPT-4 API access enabled in your account in the next few days.