question-mark
Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

mmmlu tasks improvement

See original GitHub issue

Our human annotators are complaining about the task definitions for the MMMLU tasks.

I think part of the issue is that the tasks don’t explain enough about the desired outputs. For example, here: https://github.com/allenai/natural-instructions-expansion/blob/master/tasks/task667_mmmlu_answer_generation_business_ethics.json

The only thing said is that "You are given a question on business ethics. But then: How do the expected answer relate to business ethics? (should they pick something that is more ethical or less ethical?) What is even “business ethics”, to begin with? We need to improve the instructions to address these.

Issue Analytics

  • State:closed
  • Created 2 years ago
  • Comments:7 (4 by maintainers)

github_iconTop GitHub Comments

1reaction
pulkitverma25commented, Nov 24, 2021

@pulkitverma25 are there any updates on Discussion #467. The conflicts might be difficult to resolve if we’re working on the same tasks.

Since the updates are in different parts of the files, this shouldn’t be difficult. You go ahead and submit your changes, I will take care of the merge. To avoid any confusion: The only issue related to MMMLU tasks that I am currently working on is #569

1reaction
danyaljjcommented, Nov 15, 2021

@Sujan242 could you work on improving these task definitions?

Read more comments on GitHub >

github_iconTop Results From Across the Web

Measuring Massive Multitask Language Understanding - arXiv
However, on every one of the 57 tasks, the best models still need substantial improvements before they can reach expert-level accuracy.
Read more >
Another negative example for MMLU tasks. · Issue #569 - GitHub
It seems like the annotators needed one negative example that actually had an incorrect answer, not just an answer with an incorrect format....
Read more >
Paper tables with annotated results for Multi-Task Learning in ...
In recent years, Multi-Task Learning (MTL), which can leverage useful information of related tasks to achieve simultaneous performance improvement on ...
Read more >
Is AI Progress Impossible To Predict? - LessWrong
For example, does a task improving rapidly when you go from a small model to a 7B parameter ... Individual MMMLU tasks are...
Read more >
The 12 tips that will improve your multitasking skills! - Cirkus
Juggling many tasks can become easier and less stressful with practice. Multitasking skills can be learned and improved with every project ...
Read more >

github_iconTop Related Medium Post

No results found

github_iconTop Related StackOverflow Question

No results found

github_iconTroubleshoot Live Code

Lightrun enables developers to add logs, metrics and snapshots to live code - no restarts or redeploys required.
Start Free

github_iconTop Related Reddit Thread

No results found

github_iconTop Related Hackernoon Post

No results found

github_iconTop Related Tweet

No results found

github_iconTop Related Dev.to Post

No results found

github_iconTop Related Hashnode Post

No results found