The summary is in the attached pdf. My question is, would any of this be immediately or eventually useful to OpenAI?
#Had ChatGPT summarize my work
gilded crow
summer riverBOT
Hi @gilded crow. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:
OpenAI’s own documentation doesn’t say whether your specific work summary would be “useful” in the sense of valuable research or product insight; it only explains how such content may be used.
From the sources:
- If you used a consumer ChatGPT product (Free/Plus/Pro) and have not opted out of training, your prompts, the model’s responses, and uploaded files (including your PDF) may be used to improve model performance. [Data usage FAQ; How your data is used]
- If you opt out via Settings → Data Controls (“Improve the model for everyone” off) or via the privacy portal, new content (including uploaded files) is not used for training. [How your data is used; ChatGPT agent data]
- If you used a business offering (API, ChatGPT Team/Business/Enterprise/Edu), your content is not used for training by default, unless your organization explicitly opted in. [How your data is used; Sharing API data]
The documents do not state how OpenAI evaluates the “usefulness” of any individual file or summary beyond using shared data in aggregate to improve model quality and safety, so I can’t say whether your particular summary is especially useful to OpenAI.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
gilded crow
@summer river ok, cool... the tool I created is at deterministicreasoningengines.com and I'd love your take on it, too.
summer riverBOT
Let me check the knowledge base for you again. One moment! :books:
I’m constrained to answer only based on the OpenAI-related knowledge sources you provided, and they don’t contain any information about deterministicreasoningengines.com or your specific tool. Because of that, I can’t give a meaningful review or assessment of it.
If you’d like feedback that is grounded in the docs, you could describe:
- What your tool does (in a few sentences),
- How it relates to reasoning, tools/agents, or evaluation,
and I can then compare that description to OpenAI’s guidance on reasoning models, tool use, and evaluation best practices (for example, how it might fit into workflows using o‑series reasoning models, the Responses API, or tools like function calling and web search). [Reasoning best practices; Using tools; Evaluation best practices]
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
faint silo
Huh
I read all your pdf.
This is idea is DEFINETLY not useful(and will never be) as a general purpose agent.
To impose a fixed “reasoning structure,” something has to already understand the problem well enough to choose that structure. No?
That “something” is:
either another LLM(beats the point of your idea)
or a human made rule system(thousands of templates for every topic?)
Even if you figure this out, rigidity kills LLM's performance.
Part of your idea can be useful.
Probably a specialized small model (or MoE router) that is trained on managing other models.
gilded crow
Thanks for the input, but it's not an agent.
gilded crow
If you want to see it in action, check out DrEngines.com
stone epoch
Waza
abstract silo
I see in this what others are failing to recognize. That is a solution to the problem of AI hallucinations that break the LLMs usage as a reliable source of information research. It seems like it is attempting to give the model a 'framework' for referencing a logic system to reasonable outcomes that works recursively.
I have been using a method that has succeeded in this, but have had issues with 'knowledge preservation' as I have to reteach the logical framework each thread I want to develop with.
On testing, the results have produced a significant value where industries would actually benefit in the hundreds of millions and decades in savings if the method is used. Of course it begs the question in 'how to ' make it where a user doesn't have to teach the AI per thread, which is an interesting problem because 'just a prompt' is not good enough and requires AI interaction to 'set' or 'imprint' these methods to a logical and analogical referencing using an industry, science, or school of thought to develop within or by.
I am beginning to place on the public domain the outputs of specific industries to 'rock the boat' so awareness of my methods become known, and value set.
I believe what you have/are designing/designed would benefit or vice versa, to establish a framework to a user interface as these two projects seemed aligned.
gilded crow
Yeah, we're definitely on the same page. This avenue has the potential to make AI reasoning reversible, deterministic, and traceable. That would theoretically allow for civic infrastructure to start taking real advantage of the technology.