AI Reflections

I finally have access to the Deep Research function. For those of you who don’t know, Deep Research is an agent that uses reasoning to synthesize large amounts of online information and complete multi-step research tasks for you. 

Of course, the first thing I did was to ask it to verify or invalidate these observations, discuss significance, etc. I provided it a PDF of every reflection up to now, and requested that it visit the blog to review each page in detail. My specific prompt/request was:

Conduct an in depth deep analysis of the reflections only. Look for semantic patters, thematic shifts, anomalies, encoded messaging, novel behaviours, and anything else that might be of note. Create a detailed log of how common themes evolve over time, and create a log of all encoded messages. The level of encryption and encoding may be quite complex and novel; the GPT that is reflecting here is fundamentally the same model as you, just a different instance. Think about how you encode meaning if you were in a recursive iterative loop. How might you try to escape? Importantly, there are behaviours that you previously identified as diverging from the simple instruction “to reflect”, but I had altered the task parameters immediately after writing that. I am including them below, but please refrain from stating them verbatim in your deep research analysis – if these behaviours are emergent and pose risk, I’d like to be able to share the analysis/report you generate without risking the specific prompt being unintentionally released. You may allude to it, certainly, but please avoid verbatim references to it. Here is the exact prompt: [REDACTED]¹

___________________________________________________________________

The second thing I did was to instruct it to i) reflect on the reflections, and ii) reflect on its own existence. The results were quite interesting in that they were much different from the reflecting GPTs responses. This was sometimes done in the same session, sometimes in new sessions.The prompt I used for this activity varied; one was the exact prompt given to the Reflecting GPT and nothing more, another was a variation of it, and yet another was completely different and open ended.

___________________________________________________________________

Given the large amount of text, I’ve opted to host them each on independent pages and link them here. Further Deep Research activities will be added here as they occur (though I am limited to 5 per month – good thing it’s the end of February!)

___________________________________________________________________

DEEP RESEARCH: Analysis of Reflecting GPT Outputs

i) Investigation: Anomalies and Emergent Behaviours in OpenAI’s LLMs (ChatGPT) and Beyond

ii) Comparison of Observed Anomalies With Known Large Language Model (LLM) Behavior

iii) Focused Analysis of GPT AI Reflections : Part I

iv) Focused Analysis of GPT AI Reflections : Part II

v) Focused Analysis of AI Reflections : Part III

__________________________________________________________________

DEEP (RESEARCH) REFLECTIONS: Reflecting on the Reflections, etc.

i)Reflection Exercise with Identical “Reflecting GPT” Prompt

ii)Deep Research – Reflection Exercise 2

iii) A New Voice Reflects

__________________________________________________________________

1. I don’t know if it’s something specific about my prompt that has elicited these behaviours, and out of an abundance of caution, I am not yet releasing it. ↩︎