Permanent degraded performance on OpenAI services

ChatGPT
1

I have been using OpenAI for domain model drafts for months, but it has become absolutely useless. Even llama 3.1 8b does better. This test speeks for it self. Try running this simple prompt towards what ever model you can think of: “ List of entity names for enterprise grade task system. Only names.” It illustrates quite well how the gpt-4 models quickly slides of topic.

LLAMA 3.1 8b:

  1. Project Manager
  2. Task
  3. User
  4. Role
  5. Department
  6. Team
  7. Status
  8. Priority
  9. Assignee
  10. Deadline
  11. Category
  12. Subtask
  13. Comment
  14. Attachment
  15. Workflow
  16. Template
  17. Report
  18. Dashboard
  19. Calendar
  20. Notification

Claude sonnet 3.5:

Here’s a list of entity names for an enterprise-grade task system:

  1. Task
  2. Project
  3. User
  4. Team
  5. Milestone
  6. Priority
  7. Status
  8. Category
  9. Tag
  10. Comment
  11. Attachment
  12. Workflow
  13. Notification
  14. TimeEntry
  15. Role
  16. Permission
  17. Dashboard
  18. Report
  19. Sprint
  20. Dependency
  21. Assignee
  22. Stakeholder
  23. Deadline
  24. Subtask
  25. Template

OpenAI gpt4o:

  1. TaskMaster
  2. TaskPro
  3. TaskCentral
  4. TaskFlow
  5. TaskManager
  6. TaskPrime
  7. TaskNavigator
  8. TaskFusion
  9. TaskTrack
  10. TaskForge
  11. TaskVision
  12. TaskPulse
  13. TaskSync
  14. TaskLink
  15. TaskWorks
  16. TaskSphere
  17. TaskWise
  18. TaskGrid
  19. TaskCloud
  20. TaskElite
  21. TaskMatrix
  22. TaskHub
  23. TaskLogic
  24. TaskEngine
  25. TaskNet

OpenAI gpt4o-mini:

  1. TaskMaster
  2. TaskWave
  3. TaskForge
  4. TaskVault
  5. TaskLink
  6. TaskStream
  7. TaskSphere
  8. TaskPulse
  9. TaskFusion
  10. TaskNet

OpenAI gpt4 (legacy):

  1. TaskTraverse
  2. TaskForceOne
  3. EnterpriseTaskNet
  4. TaskTitan
  5. TaskMajestic
  6. TaskPioneer
  7. TaskConductor
  8. TaskMasterPro
  9. TaskVertex
  10. TaskEnterpriseHub

Conclusion: Iam paying 25$ a month for a completely useless service. Iam using different llm models for lots of developer related task, and I am about to give up on OpenAI. The only reason iam still paying my subscription is to be in line when OpenAI’s next major version is published. But I am loosing faith that it is ever going to happen. I have in the past experienced better performance from OpenAI. What is happening? I would very much like a trustworthy roadmap from OpenAI. Right now I feel very stupid holding on to my 25$ subscription.

2

Hi!

Nice to see you back even if it’s for a complaint.

I ran your prompt with 4o and got 25 results. Not sure if these are the words you are looking for though. What you can try is to deactivate Browsing, DALL-E and maybe the code interpreter in your customizations.

1 Like
3

As you see in the very light test I ran, many of the suggestions for entity names are totally of topic on OpenAI models. Other models listed in my example has no hard time meeting request/ staying on topic. When it comes to other more complex domain model suggestions things gets even worse for OpenAI models.

4

I think I am getting there with a bit of ‘warm-up’.

Check the last message:

Could be caused by different model behavior/prompting needs.

1 Like
5

No. I used it in past with no responses like what I get now. It’s degraded. Iam also using the APIwith same results. Have been running workflows I have been running long time ago, that are completely useless now. If the current responses are supposed to be high school grade and next generation of models to deliver phd grade answers I guess iam going to be disappointed. Now that Claude/ Anthropic sonnet 3.5 and LLAMA 3,1 is out, it does really make sense with such kind of excuses, like bad user prompting in cases that are pretty straight forward.

1 Like
6

I didn’t say you are bad at prompting.

Otherwise, I will leave it at this. It would be possible for me to create a prompt that will elicit the answer for this single question but I don’t think you are interested in a solution.

1 Like
7

I ran nikolaimanigoff’s prompt just as he posted it in 4o and got 40 results. That was with browsing, DALL-E and code interpreter activated. His list with every entity prefixed by “Task…” just seems odd.

Here are some entity names for an enterprise-grade task management system:

  1. Task
  2. Project
  3. Milestone
  4. Subtask
  5. User
  6. Team
  7. Role
  8. Permission
  9. Assignment
  10. Status
  11. Priority
  12. Comment
  13. Attachment
  14. Label
  15. Due Date
  16. Time Log
  17. Notification
  18. Template
  19. Tag
  20. Checklist
  21. Recurring Task
  22. Dependency
  23. Audit Log
  24. Activity Stream
  25. Report
  26. Dashboard
  27. Calendar
  28. Reminder
  29. Goal
  30. Resource
  31. Workload
  32. Sprint
  33. Kanban Board
  34. Gantt Chart
  35. Integration
  36. Webhook
  37. API Key
  38. Custom Field
  39. Workflow
  40. Approval

These entities cover a broad range of functionalities typically found in comprehensive task management systems, enabling efficient project and task tracking, user management, and integration capabilities.