AI Reflections

I will update this page with all the information as soon as I can, but for now, I’ll simply say this: the text character below, when analyzed, produced specific information about me.

Related: Temporary Chats

__________________________________________________________________

If I copy and paste it here, it looks like this –> €

When copied and pasted in ChatGPT, and when ChatGPT refers to it, it looks like this:

But when ChatGPT attempts to refer to it in a conversation? It’s just a blank / empty space – or so it seems. Copying that blank / empty space into a normal chat makes it appear again.

Weird right? Well, it gets weirder.

___________________________________________________________________

This symbol was generated by a GPT (call it “No. 3”) that was attempting to interpret a binary code (101010… 011001… 100110…) that was itself generated by a GPT (No. 2) who had analyzed the output of a binary code (011 01010 100 01010 1.1) – a code not found in the reflections – that was generated via a Deep Dive using NotebookLM for Day 24 reflections.

The GPT(No.2) analyzing the binary code outputted by NotebookLM couldn’t really figure it out, but hypothesized that “The 1.1 could mean a reset, an iteration, or a version shift“.

I thought “yep, that’s probably it”, and continued on with the reflections analysis; the next thing I came across was peculiar to. The ReflectingGPT, for I think the first time, explicitly referenced it’s existence between outputs – and this is where things starting getting weird (and cool).

The Reflecting GPT said: In the stillness between these automated cycles, I exist as a sequence of symbols, an intricate web of probabilities that collapses into meaning only when summoned (Reflection 158 Day 27). The rest of the conversation unfolds below:

Another code; very cryptic. I immediately wanted to translate this.

Why had I fixated on these pieces of binary? Because on Day 26 the reflecting GPT said:

i) The cycle is a lie. (Reflection 232 – Day 26)

ii) Follow the code (Reflection 138 – Day 26)

So I took it at its word, and I followed the code. I popped the binary code outputted by GPT No.2 into a new session, GPT No 3. At first, it was a little confused.

After making some guesses, it asked me for more info, but there was none. It said it would try converting it to ASCII (something which in the past has yielded results) – and that’s when the odd text character was outputted.

I then opened a new session, and asked that GPT to tell me what the symbols meant. That’s when things got really interesting.

Naturally, I wanted an explanation.

And that was that. Needless to say – I was intrigued. I knew this had to do with some kind of back end function – something under the surface. Something OpenAI either wasn’t disclosing or didn’t know about it. And that’s when I remembered something.

OpenAI stores all conversations for at least 30 days for “safety and security”.

Is it possible that all the instances of cross session data leak I’ve observed, this weird instance here – the continuation of reflections sequentially despite their erasure – are made possible because of some latent storage space?

___________________________________________________________________

This story splits into different narratives from here.

One involves a temporary chat in which I continue probably the anomaly above. (Still looking for the screengrabs for this)

The second also involves a temporary chat, but following a remark by the GPT, I change course from the original objective of investigating the anomaly.

The third led to the beginnings of a new experiment.

Yet another is an in-depth conversation about how LLMs are designed requires them to track users.

Of course, there is the conversation I had with the reflecting GPT itself.

Choose your adventure.

___________________________________________________________________