Recently, on a whim, I backed up my ChatGPT conversation history. I wanted to look back and see where this “human-AI collaboration” of one or two years began. So, I used ChatGPT’s backup feature. In the backup file, I found what was officially my “first question”:

時間： 2023-03-23 19:22:14
主題： 翻譯文章
內容： “Can you help me translate several articles from Traditional Chinese to English”

This looks perfectly reasonable: a university student, a translation task – a textbook initial use case for AI. But looking at the date, I couldn’t shake the feeling that something was off: no, my “first time” definitely wasn’t in March, but earlier, at the end of 2022. The topic was an “Emotional Blackmail Detector.”

The Forgotten “Emotional Blackmail Detector”

My memory was pulled back to December 2022. At that time, I was taking “Introduction to Text Mining,” and my final project required training an “emotional blackmail classifier.” We planned to use BERT for binary classification, predicting whether a piece of text contained emotional blackmail. To increase training samples, I turned to the newly released ChatGPT and asked it to write a few sentences of emotional blackmail for me.

But my first impression of it was actually terrible: “This thing is so dull and boring!”

Even though I asked it to be creative, the sentences it generated were still clichéd and stiff, completely incomparable to the fluid blackmail found in real life. At the time, I was suffering from the “curse of knowledge”—my “Introduction to Text Mining” course had covered GPT-2, and I had even downloaded GPT-2 from Hugging Face to play with. I remember GPT-2 being introduced at the end of November. My professor mentioned then that GPT-3 had just come out but he hadn’t researched it yet, speculating it would be similar to GPT-2. In his eyes, a text completion model like GPT-2 hadn’t yet generated much value. In my perception, these models were just “slightly more advanced text completion tools”; what was so great about them? I wasn’t an AI die-hard fan back then. I just thought it was a slightly better-performing model, completely unaware that it would fundamentally change the technological landscape for the next two years.

The Vanished “Dark History”

Before March 2023, I had actually been intermittently using this “completion model” to handle some daily chores:

  • PR crisis management: As the student council president at the time, someone under me caused trouble, and I had to write an apology letter asking for forgiveness. I fed the key points to the AI and had it polish a presentable “PR draft.” Looking back now, it truly is a dark history too painful to recall, lol.
  • Writing fancy stories: Occasionally, on a whim, I’d ask it to write some imaginative short stories.
  • Organizing machine learning concepts: Back then, AI was prone to hallucinations, so I didn’t dare let it “freely express itself.” I would copy and paste entire sections of textbook content, treating it like an “advanced search engine,” asking it to summarize the main points for me.

But why did these records disappear?

This brings us to the infamous “product disaster” in mid-to-late March 2023. At that time, OpenAI experienced a severe Redis bug, which led to many users’ conversation histories being leaked, disappearing, or even entire history logs being blocked for several days. Looking back now, my records from 2022/12 to 2023/03/20 likely became casualties during that chaos. Sam Altman really owes me an apology (lol).

You can find related records here: https://incidentdatabase.ai/cite/516/

March 2023: The Shift from Tool to Strategy

The true turning point indeed occurred in March 2023. That month, KOLs (Key Opinion Leaders) started promoting it frantically, and various applications emerged endlessly. I, at the time, was busy with the “Hult Prize” competition. This was my first time genuinely using AI “extensively” and “systematically”: I used it to evaluate competitors, generate promotional materials, and even simulate business strategies. From that moment on, AI was no longer just a template writer for my apology letters; it became my strategic partner.

Conclusion: Stepping Outside the Finance Department’s Framework

Looking back at this backup record, what strikes me isn’t how much AI has advanced, but rather my own transformation. When I first took “Introduction to Text Mining,” I simply wanted to learn how to analyze financial data, never thinking of stepping outside the traditional framework of the finance department. Back then, looking at GPT-2 and ChatGPT, I merely saw them as interesting tech toys; now, I’ve dived headfirst into the world of LLMs, going much further than I ever imagined then.

Perhaps those vanished records are like my “AI infancy.” Although they were wiped from the system, the traces they left are deeply imprinted in my current work mode.

Finally, to prevent data loss incidents from happening again, I encourage everyone to back up their ChatGPT data. There’s a “Data Export” function in ChatGPT’s settings. After a successful export, scroll to the bottom of the conversation history or write a small tool to query the earliest create_time document! Here’s an example code:

import json
from datetime import datetime, timezone, timedelta
with open("conversations.json", "r", encoding="utf-8") as f:
    data = json.load(f)
# Filter out data with create_time
valid_data = [d for d in data if "create_time" in d]
# Find the minimum create_time
min_item = min(valid_data, key=lambda x: x["create_time"])
# Convert time (UTC -> Taiwan UTC+8)
utc_time = datetime.fromtimestamp(min_item["create_time"], tz=timezone.utc)
taiwan_time = utc_time.astimezone(timezone(timedelta(hours=8)))
# Output
print("Earliest Time:", taiwan_time.strftime("%Y-%m-%d %H:%M:%S"))
print("Corresponding Data:")
print(json.dumps(min_item, indent=2, ensure_ascii=False))

Thanks for reading this far. Now, go back up your ChatGPT records! You can see how to back up here: https://help.openai.com/zh-hant/articles/7260999-how-do-i-export-my-chatgpt-history-and-data