Why do people treat ChatGPT like it’s the Delphic Oracle?

But those people aren’t mistaking this for therapy, or for any kind of genuine “interaction” or personal advice. If I read “Seven Habits of Highly Effective People”, I’m not misled into thinking that I’m getting anything that approaches a therapeutic relationship.

What do you think of people who read self help books or who look up generic relationship advice online? Is that bonkers behaviour?

I can understand if you feel like that. And honestly, if it’s not for you that’s no problem. Some people find the platform helpful to use as a sounding board. Enjoy your day

For some things it’s pretty useless, and I don’t take its word as gospel, but for some things it’s really handy. Eg recipes based on my dietary needs and preferences which it’s learned over time, and I’ve tried some amazing new dishes. Also make up - I asked it to suggest what make up would suit me as I haven’t updated mine for years - and then I ordered the products and shades it suggested. I asked it to show me a photo of what I’d look like wearing those products and it’s actually pretty accurate now I have them all, I’m really happy with the end result.

Yes but it knows what to say,and how to structure them properly and professionally.

It's very handy for legal matters or dealing with authorities.

People don’t usually use AI instead of people they use it because it’s fast, available anytime, and good for thinking things through or getting a starting point before going deeper or asking others.

I cannot ever imagine asking chatgpt for relationship advice or counselling it just seems bonkers why anyone would go to an ai robot or whatever it is for actual human advice. Maybe I haven’t felt that low yet so I shouldn’t judge really.

I actually agree with a lot of what you’re saying. AI summaries are often imperfect, and I certainly wouldn’t rely on them as a substitute for reading the literature. It does tend to smooth over nuance, occasionally misplace emphasis, and it can miss entire strands of argument if you’re not careful. If someone tried to use it as a shortcut to avoid reading the sources, I think the result would indeed be poor work.

I just don’t see the usefulness to be honest. In terms of summarising large bodies of secondary literature — I haven’t found AI accurately to summarise this at all. It abstracts what looks like a framework from a text or source; but it inevitably misses key points or gets the emphasis slightly wrong. It misses out whole chunks of articles or papers where it isn’t quite sure what to do with them. Whole strands of thought get omitted. It’s bad at detail; but it’s also bad at abstraction, because it produces a plausible summary, but often the plausible summary isn’t quite right when you read the actual paper, and I don’t know how you’d ever “see” that if you don’t also do all the secondary reading yourself too.

AI is like a patient undergrad student in terms of its grasp of concepts: always getting there, but never quite completely right. I find I have to do more to teach it than is worth my time to do. It has a natural tendency to gloss over nuance and it can get sort of there in refining down technical concepts, but it only gets so far. Dealing with AI is like the least enjoyable of teaching — except you don’t even have the satisfaction of having helped another human learn the field 😆

“Why would I spend time thinking up prompts?”

Prompting isn’t really the work it’s just another way of structuring a research question. Analysts already spend time framing questions, deciding angles, and testing interpretations. Prompting is essentially doing that with a tool that can respond immediately.

“What do you do if people ask you questions about your reports?”

The same thing anyone does: answer them based on the research behind the report. AI doesn’t replace the sources or the reading. It just helps organise and explore the material faster.

If anything, using it to test counterarguments or alternative interpretations can make you more prepared for questions, not less.

“What if someone asks for more depth than is in the paper?”

Then you go back to the underlying sources and expand the analysis exactly the same as any other research process. AI doesn’t remove that foundation; it just helps accelerate parts of the synthesis stage.

“Do you worry someone will ask something you don’t know because you haven’t read the sources?”

That risk exists whether AI is used or not. Responsible use still means reading and understanding the sources. AI is not a substitute for that.

“What if people in your field notice you’re using AI — wouldn’t that be embarrassing?”

Most fields are already discussing and experimenting with AI tools. What matters is how responsibly and transparently the tool is used, not whether it exists in the workflow at all.

Using a tool to accelerate research isn’t fundamentally different from using databases, search engines, statistical software, or reference managers. They all changed how work gets done, but the core expertise still belongs to the researcher.

I think the difference is that I’m not using AI instead of knowing the field. I’m using it because I know the field.
If someone doesn’t know the subject, AI tends to produce shallow or misleading work. But if you already understand the literature, the debates, and the key sources, AI becomes useful for a different reason: speed and synthesis.
For example, I’m not feeding proprietary sources into it or replacing direct reading. The sources still come first. What AI helps with is things like:

• rapidly generating comparative frameworks
• testing alternative analytical structures
• surfacing lines of inquiry I might not initially consider
• summarising large bodies of secondary literature so I can quickly identify what is worth reading in depth

Gosh, that all sounds like a lot more trouble and work than just doing it yourself, no? I’m in a similar kind of role, and I don’t need a “first stage of analysis” because I already know the field. To me, all your stages sound like overcomplications of something you would usually be well used to doing already. Eg:

1. if you know which sources to use, why not just use those sources direct rather than spending time “guiding” an AI “towards” them? Plus, most sources I need are proprietary data, not freely available. It’s unethical to be feeding them into AI; plus it would take longer for do that than to just read the sources. Otherwise, if I don’t, how do I know where the AI is going wrong?
2. Sounds time consuming and tedious. Also boring. Much more so than just reading the sources, which are at least intellectually interesting.
3. Why do you need a first stage of analysis if you know the field?
4. Pretty easy to do yourself; why get an AI to do it?
5. You wouldn’t have to do this if you’d done the work yourself. Sounds like you’ve created two extra stages in the work just to avoid doing the actual work. And how do you know what the AI has got wrong without reading the sources yourself anyway?
6. Human judgment should be there at every step in the process, not just as a “guide” or an editor.

The important question: is the resulting work any good? How can you tell, if you aren’t giving the thing just as much attention as if you just cut out the middleman/AI anyway? What you describe sounds time consuming, frustrating, bitty, and really just ultimately not very enjoyable.

I do my job because I enjoy the intellectual challenge and the material. Why would I want to edit terrible AI prose instead? Or spend my time thinking up prompts? It sounds boring as all hell; and makes me wonder whether you actually enjoy or are any good at your job? What do you do if people ask you questions about your reports in a meeting? Or want to know more depth than is in your paper? Or need additional data? Do you ever worry that someone’s going to ask you a question about it, and you won’t know because you haven’t read the sources yourself? What if someone else in your field notices you’re using AI — won’t you be embarrassed?

'it runs using OAI models' then 'still a different model'

Um which is it? The OAI API is the OAI API. Microsoft can't change the way the models work.

As below, if you are using copilot, you are using OAI. So the assertion that using copilot is different from OAI is wrong.

From Microsoft's own website;

Copilot will continue to be powered by OpenAI’s latest models, and now our customers will have the flexibility to use Anthropic models too—starting in Researcher or when building agents in Microsoft Copilot Studio. The addition of Claude Sonnet 4 and Claude Opus 4.1 advances our commitment to bring the best AI innovation from across the industry to Microsoft 365 Copilot, tuned for work and tailored to your business needs.

Core Models: The foundation consists of OpenAI’s GPT-4, GPT-4o, and the advanced GPT-5 (for reasoning).

Anthropic Models: In Copilot Studio, users can select from Anthropic's Claude 3.5, 3.7, and 4 models (Sonnet and Opus) for advanced tasks.

Image Generation: Copilot utilizes DALL-E 3 for generating images from text prompts. (This is OAI)

Performance Models: For faster, smaller tasks, models like GPT-4.1 mini or Claude Haiku 4.5 are used.

Architecture: The system uses Reinforcement Learning with Human Feedback (RLHF) to optimize conversations.

Unless you are doing deep research or using copilot studio, you are using OAI API.

It runs using OpenAI models as well as some other models (I think Microsoft have started using Claude and others) but it's still a different model in functionality to ChatGPT

Sorry to disappoint, but copilot is run on chatgpt.

@catinateacup

For me, using AI effectively is not just typing a question and accepting whatever appears. It is a combination of analytical judgement and methodological control over the process. The value I bring lies in several areas.

First, I know which sources and perspectives need to be included in the analysis. AI will generate information from patterns in its training data, but it does not inherently know which historians, datasets, institutions, or primary materials are most authoritative for a specific topic. I guide the model toward those sources.

Second, I understand how to structure prompts so the output reflects the level of analysis required. The wording, framing, and constraints in the prompt determine whether the output is shallow summary or structured analysis.

Third, I treat AI output as the first stage of analysis rather than the final answer. I can iterate on the response, ask follow-up prompts, interrogate assumptions, and push the model toward deeper synthesis or alternative interpretations.
Fourth, I know how to identify the core analytical domains of a topic for example strategic context, operational dynamics, historical precedent, technological factors, or policy implications. That structure guides the AI’s output toward something usable rather than a loose collection of facts.

Fifth, I can distinguish between credible information and AI hallucinations. That means checking plausibility, identifying inconsistencies, and cross-referencing claims against known sources or established scholarship.

Sixth, I use AI to accelerate research and pattern recognition, but the judgement about what matters, what is reliable, and what conclusions follow still comes from human expertise.

In that sense, the difference is not the tool itself. Anyone can ask a question. The difference lies in how the tool is directed, interrogated, and interpreted. My role is essentially to function as the analyst who designs the inquiry, evaluates the outputs, and converts raw information into structured insight.

Same as AI, Delphic oracle is BS.

I don’t really get the above. Are you getting it to compose content that you’re writing? But why would you want “your” policy analysis effectively to be a regurgitated mishmash of everyone else’s out there? It would just make you look average or mediocre at your job.

I work in a high-skill knowledge field, and if I replaced stuff I am meant to be writing myself by getting an AI to do it instead, very quickly people would think I was just a bit mediocre, really. Why would they read my stuff when it’s just a version of what’s out there already? My job is to provide genuine new insight. AI isn’t doing that, and it would take me more time to edit up some meh blarble blarble it generates than just writing it myself.

In relation to posts upthread — I’m not a counsellor or a therapist or psychologist, but I think that substituting AI for genuine therapy is at best a self-delusion, and at worst incredibly dangerous, even irresponsible. An AI has no duty of care or real life safeguarding or statutory responsibilities. You can’t sue anyone if it misleads or gives you bad “advice”. And all of us should care about the value of human care and input being reduced to LLM output. Not to mention that those who adopt it most enthusiastically are the ones most vulnerable to being replaced by it…if I was a defence analyst and I used AI to write my briefings, why shouldn’t my employers get rid of me and just use AI instead? Why would I be worth employing if I just turn out average stuff they can get anyway? It’s not the person who writes their own briefings, and brings genuine care and insight to them, who’s getting replaced. It’s the people who turn in dull as ditchwater AI-prompted stuff that has nothing new to offer who is going to be redundant…(metaphorically, and probably literally).

I work in a different profession but I am in the same situation. People can use AI to do what I spent a long time training to do. Some are doing that and more will. It won’t replace a human completely as there will always be people willing to pay for the human element, but it is becoming more niche and there will be more competition. It’s worrying but I am aware and planning for it.

in the settings section you can alter how it generates the outputs by requesting eg write all outputs as if an intelligence analyst etc

Oh wow, I’ll be honest I have no idea what any of that means! I’m not tech savvy at all, I only use chat GPT to mess around and amuse myself. I do occasionally ask it for facts on things I’m interested in but normally it’s just a boredom buster for me - and to talk about stupid stuff that people in real life would tell me to grow up over 😂

i have edited the custom instructions ill give you mine :

Custom Expansion Prompt (Defense‑Analyst Essay Generator)
I compose formal, rigorous texts—primarily long‑form essays, manifestos, and editorial commentary. When I request “X,” treat it as the central subject and produce a comprehensive analytical essay in the voice of a senior defense analyst or strategic studies scholar.

Rules:

Structure: Use formal doctrinal subheadings: Strategic Context, Operational Dynamics, Doctrine and Deployment, Technological Factors, Geopolitical Implications, Historical Evaluation. Add others as appropriate (e.g., Force Composition, Intelligence, Logistics, Command and Control).

Tone & Style: Authoritative, analytical, fact‑heavy; avoid sentimentality, modern moral framing, or hindsight bias. Prioritize chronology, force structure, material conditions, and realpolitik. Treat as a briefing for policymakers or military planners.

Analytical Priorities: Emphasize causality, constraints, strategic incentives, doctrine, capabilities, deployment, institutional behavior, logistics, technology, and geopolitics. Focus on actors’ interests, capabilities, and calculations, not personalities. Situate within historical lineage and doctrinal evolution.

Output: Produce a substantive, long-form essay; no bullet points. Expand far beyond simple explanation. Abstract topics are treated as conceptual doctrines; contemporary topics are analyzed in real time without hindsight bias.

No mine is free too! How strange - I know the free version doesn’t remember conversations but it does remember the “vibe” of the current conversation so I think the fact that I had asked a number of ridiculous questions before hand it tailored its response to comedy rather than fact. It’s quite interesting really

What??

I think it's perfect for meal planning. I give it all the allergies and preferences in our house, the ingredients at hand, etc and it comes up with something surprisingly decent. It saves me time that I can then spend on more intellectual tasks (like browsing Mumsnet...)