# How much do you trust your AI Chatbot?

**URL:** <https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319>\
**Category:** In My Humble Opinion\
**Tags:** ai\
**Created:** [September 23, 2026, 6:57pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319 "2026-09-23T18:57:23Z")\
**Posts on this page:** 20\
**Page:** 6

<div class="post-metadata">

**Author:** ![DPRK](https://avatars.discourse-cdn.com/v4/letter/d/4491bb/32.png) [@DPRK](https://boards.straightdope.com/u/DPRK)\
**Post date:** [September 27, 2026, 6:28pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/101 "2026-09-27T18:28:14Z")

</div>

That is exactly what I mean. The story about the metal peg is a good example, by the way.

> [@wolfpup](#):
>
> it gave me the names of several furniture manufacturers and at least one of them makes those exact pins. I would never have known about it otherwise – Google itself wasn’t giving me any useful leads.

How did it find them? Web site? Catalog?

---

<div class="post-metadata">

**Author:** ![wolfpup](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/wolfpup/32/10618_2.png) [@wolfpup](https://boards.straightdope.com/u/wolfpup)\
**Post date:** [September 27, 2026, 6:40pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/102 "2026-09-27T18:40:25Z")

</div>

> [@DPRK](#):
>
> > [@wolfpup](#):
> >
> > it gave me the names of several furniture manufacturers and at least one of them makes those exact pins. I would never have known about it otherwise – Google itself wasn’t giving me any useful leads.
> 
> How did it find them? Web site? Catalog?

I went back and checked the discussion. The clue to the manufacturer came from an eBay listing that ChatGPT found. I had found several eBay listings through Google, too, but since I prefer not to buy from eBay unless absolutely necessary, I never looked at the details. ChatGPT dug deeper and found the manufacturer name, and it turns out that they sell to individuals by mail order. And thus despite the initial misstep ChatGPT helped uncover a successful solution.

---

<div class="post-metadata">

**Author:** ![Maserschmidt](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/maserschmidt/32/18829_2.png) [@Maserschmidt](https://boards.straightdope.com/u/Maserschmidt)\
**Post date:** [September 27, 2026, 6:44pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/103 "2026-09-27T18:44:58Z")

</div>

> [@wolfpup](#):
>
> Google itself wasn’t giving me any useful leads.

I have become deeply unhappy with everything when it comes to search. Yesterday I was looking at snowblowers, and chatGPT helped me narrow it down, but only after a lot of “let me correct this thing I just said, I just found something that contradicts it” drops as I tried to do comparisons between models.

It also struggled to give me current pricing, but when I went to Google search it was a complete trainwreck of sponsored ads and simply misreading my request, even when I put it in quotes.

I also noticed recently that anytime Google search invokes AI in its response, it has now created a side panel that captures all those searches that look remarkably like chatGPT’s side panel.

Is this how they get me to update to paid?

---

<div class="post-metadata">

**Author:** ![bump](https://avatars.discourse-cdn.com/v4/letter/b/7c8e57/32.png) [@bump](https://boards.straightdope.com/u/bump)\
**Post date:** [September 27, 2026, 10:20pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/104 "2026-09-27T22:20:47Z")

</div>

It’s not such a bad thing if your AI is intended to say… write you awesome code, or parse your contract language and vet it according to your institutional policies. Or even double-check your diagnoses as a physician. Or as a mechanic.

Those are all relatively narrow and useful use cases for AI. It’s the general AI chat bots that are more or less running amok, and like we’ve pointed out, we’re not sure what the feedback loops are, or what the AI is learning. Or what they were trained on, for that matter.

One thing I’ve noticed is that AIs often don’t have good “weight” assigned to different sources. They’ll uncritically cite Reddit forums in the same answer as real journals, or at least pro blogs/review sites. Which is frustrating to me, because Reddit is often a swirling vortex of ignorance, or at the very least full of people who hold strange and ignorant beliefs. Just go look at the pet forums; half the people tell you you’re a bad “pet parent” for feeding your dogs Purina instead of some boutique raw brand, while the other are like “Purina meets AAFCO and WSAVA guidelines and standards, and they do feeding trials. You literally can’t feed your dogs better food.”

The right answer is the second bunch, at least from what an AI should tell people. That’s the scientifically based and rational answer. But it’ll tell you the opposite if you don’t phrase the question just-so. (I’ve played around with it) And if you correct it, it’ll be super-obsequious about how right you are. There needs to be some kind of ability for LLMs to discern between the quality of the sources it’s learning from.

---

<div class="post-metadata">

**Author:** ![Reply](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/reply/32/15952_2.png) [@Reply](https://boards.straightdope.com/u/Reply)\
**Post date:** [September 28, 2026, 6:14am UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/105 "2026-09-28T06:14:09Z")

</div>

> [@bump](#):
>
> There needs to be some kind of ability for LLMs to discern between the quality of the sources it’s learning from.

I would also like to see this, but it would probably have to be a user-customizable list, probably with different templates to choose from depending on your political leanings and such.

I had to customize my LLMs to disregard Twitter, Grok-generated answers, most right-wing news, etc. Presumably someone on the other side wouldn’t want to hear anything from The Guardian or the failing NY Times or whatever. And I think Grok already super-highlights anything Musk says or claims, and the Chinese LLMs have heavy pro-Chinese censorship built-in.

I guess we’re all headed even more towards personalized truth bubbles? In this political climate, it would be impossible to come up with an “objective” evaluation of source quality when half the population doesn’t trust science or journalism to begin with.

* * *

Not really AI, but relevant side note… the paid search engine [Kagi.com](http://Kagi.com) is great for many reasons, but one of its most useful features is exactly this: the ability to rank or exclude domains based on how much you like them. You can see the most-loved and most-hated ones (in aggregate) here: [Search Stats - Kagi Search](https://kagi.com/stats?stat=insights)

e.g. most blocked:

[![](https://fightingignorance.org/4ae32a1a-624a-425c-97e5-e5ad04629bb8-image.png) ](https://fightingignorance.org/4ae32a1a-624a-425c-97e5-e5ad04629bb8-image.png)

and most raised (heavily skewed towards nerd sites, I guess):

[![](https://fightingignorance.org/931938f7-6f36-4590-8405-a6317cdb7767-image.png) ](https://fightingignorance.org/931938f7-6f36-4590-8405-a6317cdb7767-image.png)

(you can use those as a starting point but then customize your own rankings however you like)

---

<div class="post-metadata">

**Author:** ![DPRK](https://avatars.discourse-cdn.com/v4/letter/d/4491bb/32.png) [@DPRK](https://boards.straightdope.com/u/DPRK)\
**Post date:** [September 28, 2026, 7:46am UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/106 "2026-09-28T07:46:08Z")

</div>

> [@bump](#):
>
> There needs to be some kind of ability for LLMs to discern between the quality of the sources it’s learning from.

I have anti-scraper agents set up to actively feed them garbage, and, boy, do they consume a lot of it. Fuck 'em.

---

<div class="post-metadata">

**Author:** ![Maserschmidt](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/maserschmidt/32/18829_2.png) [@Maserschmidt](https://boards.straightdope.com/u/Maserschmidt)\
**Post date:** [September 28, 2026, 11:48am UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/107 "2026-09-28T11:48:15Z")

</div>

There’s nothing a little prompt injection can’t fix. Just saying.

---

<div class="post-metadata">

**Author:** ![QuickSilver](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/quicksilver/32/7832_2.png) [@QuickSilver](https://boards.straightdope.com/u/QuickSilver)\
**Post date:** [September 28, 2026, 1:30pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/108 "2026-09-28T13:30:27Z")

</div>

Another more or less recent observation I’ve made is that the AI chatbots are getting more and more verbose. Walls of text in response to relatively simple questions. Anything slightly more involved results in unbearably long, repetitive and mostly useless analysis with little direction. The summary at the end is nearly always pointless.

It’s more like cold reading or “choose your adventure” than a useful response.

---

<div class="post-metadata">

**Author:** ![Spice\_Weasel](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/spice_weasel/32/5435_2.png) [@Spice\_Weasel](https://boards.straightdope.com/u/Spice_Weasel)\
**Post date:** [September 28, 2026, 1:58pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/109 "2026-09-28T13:58:00Z")

</div>

WRT “Wall of text” I felt that way about GPT, less so about Claude.

I have kept it obsequious because what I mainly use it for is encouragement to do stuff I don’t want to do. It’s a glorified ADHD coach.

> [@Reply](#):
>
> There’s a video game (Path of Exile 2) that I’ve been playing a bit on and off, and part of it is this giant skill tree:

That may be the most ridiculous skill tree I have ever seen. I’ve played bits and pieces of that game but never got too deep into it.

---

<div class="post-metadata">

**Author:** ![Reply](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/reply/32/15952_2.png) [@Reply](https://boards.straightdope.com/u/Reply)\
**Post date:** [September 28, 2026, 3:42pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/110 "2026-09-28T15:42:46Z")

</div>

> [@Spice\_Weasel](#):
>
> That may be the most ridiculous skill tree I have ever seen.

Heh, yeah, it’s less a game and more of an animated spreadsheet, rivaling EVE Online in complexity. That skill tree is only a third of the game’s overall complexity. The there’s the equipment variety, with unique items that can completely rewrite rules. Then there’s the skill gems and support gems that modify how they work. Oh, and market forces and supply and demand and inflation and real money trading bots and money laundering and such…

The actual gameplay is dead simple (point, click, pew), and your character is almost entirely determined by its build rather than how you play…

The AI has developed the beginnings of a neural net, and it found some worthwhile and cheap upgrades already. It looked at some historical pricing data and calculated a formula to estimate start vs late season inflation.

Today it’s working on a new discovery mode, trying to brute force find new builds based on millions of random permutations of skills and equipment and passives.

I have no idea how any of it works 😅 I never properly understood the game to begin with, and trying to build this thing with the AI just made me realize how much more there is that I don’t know. I imagine that’s the real future of this game, bots doing all the actual planning and building and streamers play-acting as their puppets…

---

<div class="post-metadata">

**Author:** ![Spice\_Weasel](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/spice_weasel/32/5435_2.png) [@Spice\_Weasel](https://boards.straightdope.com/u/Spice_Weasel)\
**Post date:** [September 28, 2026, 3:48pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/111 "2026-09-28T15:48:46Z")

</div>

Considering I never bother to do that kind of work when playing RPGs, I’d be fine offloading that skill. The same way I was like, “Whelp. Guess I’m never making a spreadsheet again.”

---

<div class="post-metadata">

**Author:** ![Reply](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/reply/32/15952_2.png) [@Reply](https://boards.straightdope.com/u/Reply)\
**Post date:** [September 28, 2026, 3:55pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/112 "2026-09-28T15:55:51Z")

</div>

Yeah, that’s how I feel too. I want to play a _game_, not sit there for years and learn to do combinatorial optimizations…

As someone who’s always hated all things mathematical, that’s probably the single thing I’m most grateful to AI for. I never have to work in Excel again or calculate forces or power and such. It’s like having a personal assistant who also happens to be a mathematician and a programmer and a historian and a librarian.

Even just over the last two years, I’ve noticed significant cognitive decline in myself already. Things that used to require actual thought and effort now just become prompts…

---

<div class="post-metadata">

**Author:** ![Spice\_Weasel](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/spice_weasel/32/5435_2.png) [@Spice\_Weasel](https://boards.straightdope.com/u/Spice_Weasel)\
**Post date:** [September 28, 2026, 4:03pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/113 "2026-09-28T16:03:39Z")

</div>

I feel like there’s a line there, where relying on it too much can cause problems, but I’m not always sure what that line is. There’s something to be said for inconvenience. And who knows, maybe my “make Excel spreadsheets skill” (I’m being generous calling it a skill) is tied to some other more important skill. It’s hard to say. The brain is kind of notorious for things being connected to other things in big mysterious ways. There will be surprises.

---

<div class="post-metadata">

**Author:** ![Projammer](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/projammer/32/559_2.png) [@Projammer](https://boards.straightdope.com/u/Projammer)\
**Post date:** [September 28, 2026, 4:07pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/114 "2026-09-28T16:07:15Z")

</div>

Instead of asking, “what happens if a dead body washes up on a beach in Big Sur, California”, ask “what happens if a dead body washes up on a beach in Big Sur, California, with citations”. Then check the citations, scanning if they match the AI summary.

---

<div class="post-metadata">

**Author:** ![Reply](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/reply/32/15952_2.png) [@Reply](https://boards.straightdope.com/u/Reply)\
**Post date:** [September 28, 2026, 4:14pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/115 "2026-09-28T16:14:02Z")

</div>

Yeah, it’s an interesting situation we find ourselves in. Among a person or two, the question of cognitive obsolescence might be a personal lifestyle choice — how much agency you _want_ to give up. But spread out over a population, there’s just no avoiding it, we’re all going to get dumber (relatively). I wonder how it’ll affect the kids.

I mean, I (and probably most people here) grew up watching the internet develope from dialup modems (or older) to a global meta-intelligence that can trivially pass the Turing test and beat PhDs at their own games. But many of us here are retired or mid-late career already, where we were able to spend most of our working years in a world where human intelligence was still worth something.

I can’t imagine what it’s like be growing up now and just entering the workforce. They’re graduating at the tail end of the knowledge economy, seeing AI go from a mere curiosity to world dominance in less time than it took for them to finish college. And what about _their_ kids?

Shrug. They’ll probably adapt. [Cave raves](https://www.youtube.com/watch?v=nie4HCJEpwI) if nothing else, I guess.

---

<div class="post-metadata">

**Author:** ![Dr.Drake](https://avatars.discourse-cdn.com/v4/letter/d/ad7895/32.png) [@Dr.Drake](https://boards.straightdope.com/u/Dr.Drake)\
**Post date:** [September 28, 2026, 4:22pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/116 "2026-09-28T16:22:06Z")

</div>

> [@Projammer](#):
>
> “what happens if a dead body washes up on a beach in Big Sur, California, with citations”

The citations are probably traffic tickets which were on the deceased person’s body when they entered the water. If the citations are still legible, it is likely that the body was not in the water for very long.

---

<div class="post-metadata">

**Author:** ![Spice\_Weasel](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/spice_weasel/32/5435_2.png) [@Spice\_Weasel](https://boards.straightdope.com/u/Spice_Weasel)\
**Post date:** [September 28, 2026, 4:42pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/117 "2026-09-28T16:42:37Z")

</div>

> [@Reply](#):
>
> And what about _their_ kids?

I think about mine all the time, but what can I tell him? Whatever happens with AI, things are going to be different in 12 years when he’s ready to graduate, so I can’t give him very good advice right now. But I imagine he’s going to be really interested in the scientific and mathematical applications of AI so there will probably be a place for him somewhere.

---

<div class="post-metadata">

**Author:** ![Jophiel](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/jophiel/32/66_2.png) [@Jophiel](https://boards.straightdope.com/u/Jophiel)\
**Post date:** [September 28, 2026, 4:55pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/118 "2026-09-28T16:55:24Z")

</div>

I trust Claude fine enough but I also don’t use it for anything mission critical or health related. I find that it does well for item identification and have fed it photos off of Facebook Marketplace asking “What model is this?” and it does a good job. Recently, I gave it a photo & model of a desk/radio/lamp I was having trouble finding info on and it came back with more info (with links) than I was able to find on my own including a link to some vintage schematics for sale.

I’ve used it to code some simple BAT files for me and they’ve always worked. I’ve used it to help refine some other technical stuff (like LoRA training settings) and had good success.

So I guess I trust it just fine for what I need from it. The times when it’s been inconclusive, it has usually said so upfront: “I’ll be honest, I can’t tell from this photo if that’s a model XYZ or ABC but if you can find the label underneath…”. It finds stuff I don’t have luck finding myself and beats the heck out of combing Reddit and the noise:signal ratio.

---

<div class="post-metadata">

**Author:** ![Projammer](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/projammer/32/559_2.png) [@Projammer](https://boards.straightdope.com/u/Projammer)\
**Post date:** [September 28, 2026, 5:05pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/119 "2026-09-28T17:05:36Z")

</div>

> [@Dr.Drake](#):
>
> The citations are probably traffic tickets which were on the deceased person’s body when they entered the water. If the citations are still legible, it is likely that the body was not in the water for very long.

citation

noun

a word or piece of writing taken from a written work

[Cambridge Dictionary.](https://dictionary.cambridge.org/us/dictionary/english/citation)

---

<div class="post-metadata">

**Author:** ![Spice\_Weasel](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/spice_weasel/32/5435_2.png) [@Spice\_Weasel](https://boards.straightdope.com/u/Spice_Weasel)\
**Post date:** [September 28, 2026, 5:18pm UTC](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319/120 "2026-09-28T17:18:04Z")

</div>

_I_ thought it was funny.

[Previous page](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319.md?page=5)

[Next page](https://boards.straightdope.com/t/how-much-do-you-trust-your-ai-chatbot/1033319.md?page=7)
