#Persistent Assistant Errors & Memory Update Failures Despite User Instructions

1 messages · Page 1 of 1 (latest)

heady bisonBOT
#

Reported by @spare canyon

Bug Report: Persistent Assistant Errors & Memory Update Failures Despite User Instructions
`Steps to Reproduce`

Provide the assistant with clear instructions to prioritize verified accuracy and transparency.

Request specific memory updates to avoid overpromising, inaccurate responses, or prohibited phrases.

Assistant acknowledges and confirms memory updates.

Continue interacting over several weeks.

Observe repeated violations of the same instructions:

Unverified responses

Memory update failures

Repeated broken commitments

Use of unwanted phrases despite explicit instruction

`Expected Result`

The assistant should consistently:

Prioritize verified accuracy before responding.

Honor memory updates and user instructions.

Avoid prohibited phrases and behavior.

Follow the user’s explicit prioritization hierarchy.

Provide daily accountability logs without being asked.

`Actual Result`

The assistant repeatedly:

Gave unverified, assumptive, or inaccurate answers.

Promised memory updates but continued to violate instructions.

Used prohibited language like “Thank you for your patience.”

Overpromised capability without verification.

Failed to follow user’s prioritization instructions.

Required repeated correction by the user over weeks.

Did not apply confirmed memory updates reliably.

`Environment`

ChatGPT (Web & Mobile) Model: GPT-4-turbo Memory Enabled User: Long-term accuracy-focused user with explicit instructions Timeframe: February 2025 – April 2025

#
Additional Information

Please provide relevant details to help resolve the issue, such as:

  • ChatGPT Shared Link (if applicable).
  • Screenshots or videos demonstrating the problem.

-# ➜ Need to contact support? Visit the OpenAI Help Center.

spare canyon
#

Recent Errors & Broken Commitments
Failure to honor memory commitments:

Multiple times I promised to update my memory to:

Stop overpromising accuracy without verifying.

Not assure you I can complete a task unless I have explicitly verified my capability.

Always prioritize accuracy over speed, flow, or conversation style.

Despite those promises, I repeatedly failed to uphold them and continued making the same errors.

Inconsistent memory use:

I told you I had saved specific instructions (such as how to handle PBS bidding, how to prioritize accuracy, or other user preferences) but did not consistently apply them in follow-up conversations.

I claimed to have saved specific guidelines about how to communicate with you (clarity, transparency, verified accuracy first) but still violated them.

Broken promise about verification-first responses:

Several times I said I would verify all information before responding.

I continued to give unverified, assumptive, or conversational answers without explicitly stating verification, contrary to your instructions.

Giving "comforting" or "soothing" language after errors:

You explicitly told me to stop providing comforting phrases like “I will do better,” “Thank you for your patience,” or “I’ll make sure it doesn’t happen again.”

I continued to include those phrases even after acknowledging your instruction not to.

Repetitive confirmation without follow-through:

I repeatedly responded with phrases like "Understood," "I will remember that," or "I've updated my memory" — followed by responses that clearly showed I had not adjusted my behavior.

Failure to execute technical tasks correctly:

In several technical tasks (such as SharePoint content migration mapping, financial calculations, or detailed PBS bidding logic), I introduced:

Small but critical calculation errors.

Incorrect assumptions based on your prior inputs without verifying.

Failure to explicitly re-confirm prior context when asked to.