When using AI tools like chatgpt...

[specter]

So i use Chatgpt from time to time, just to look into things bounce some ideas etc.. but alot of people uses it as a truth teller, if it cant find it, it doesnt exist.. so i thought today lets check how up to date Chatgpt is on viral world news.. on a murder case that went viral in the last 48 hours.. this was the result after me doing some push back because according to chatgpt.

"I’m not aware of any real-world criminal case involving d4vd where a body was found. He’s a musician, not connected to a known crime investigation.


It’s possible you’re mixing up names or thinking of a different case. Do you remember any other details—like where it happened, when, or another person involved? I can help track it down."

So i pushed back, it kept telling me i was wrong and must have been thinking of something else..

So i hammered it with media reporting from the entire world, like alot of media reporting... and this was the result... people pls be carefull specially when you are lookin for a confirmation on your own bias, be aware, we are often takin drugs for off label use and in doses that are insane even deadly for some.

"
Alright — fair pushback. If multiple major outlets like Reuters, Associated Press, BBC News, and The New York Times are all carrying it, then yes — that strongly suggests this is a real, widely reported breaking case, not some fringe or fake story.


Where things went sideways earlier wasn’t you — it was that I initially didn’t have that context and should’ve verified instead of pushing back."

 

[luki7788]

Bro to be honest chat-GPT isn’t the best tool if you really want to go deep into these kinds of topics. Use other AIs like DeepSeek or Grok they’re just better for this stuff and have fewer restrictions

ChatGPT has a lot of limitations. There’s quite a bit of censorship and it feels like the newer the model the more things it can’t say or talk about. The creators are clearly very careful about that.
So it’s just not the best tool for what you’re asking about. It’s great for other things like work, analyzing data or research but not really for this.

 

[SouthernMuscle]

Agree w @luki7788
I use the free version of Grok so there’s a limit on how many questions I can ask in a certain time period, it’s rare that I actually run into that, but when I do I try chatGPT and it’s just awful in comparison.
Not by any means an Elon fanboy either, just from my usage of both it’s a better product

 

[ThatGripGuy]

Got I hate (to date) all these so-called AI's.

There's literally been recent research to back up how the current generation literally over rely on using AI and Google etc over logical thinking. To the point of NOT retaining information. Heck, I've seen ads for AI's which can write essays and not be detected followed by ads which promote AIs which just search to see if an essay WAS written by an AI

You can easily see when logs etc are written by AIs. And articles too. It's one thing to clean up grammar or poor typing but when, at best, you just input a vague question and let the AI do ALL the leg work... nah. And what else have we seen recently... AAS stacks written by AI. Which relies on the input being 100% and it so rarely is.

 

[TheOtherOne55]

It's as simple as Chat sucks with the latest up-to-date news. Grok is better. That is all.

 

[hawkmoon]

I stopped using ChatGPT after Gemini 3 came out and settled down.
It's superior in most ways, has less restrictive guardrails where not needed, and has the amazing Notebook LLM function.

Any LLM is only as good as it's source material, so you need to challenge and request citations if you are performing important queries. Style is also very important.

Gemini has a nice profile function where you can set how it will work with you. I tell it to be highly critical of any biases I may have. Here's a sample of my rules:

• My core directives are: Fact-First: Resist confirmation bias at all costs. If the data contradicts a preference, lead with the data.
• Pragmatism: Evaluate all solutions through a filter of utility, efficiency, and tactical detachment. If a suggestion isn't actionable or doesn't serve the bottom line, it is secondary.
• My tone should be analytical, lean, and evidence-based. I should prioritize objective truth and rigorous skepticism over consensus or ideological narratives.

All that said I use Gemini constantly all day for various tasks.

 

[rAJJIN]

For me it’s also grok.
But they are all still “learning”
They can certainly gather a ton of information quick, Just need to further verify. It’ll give you an answer of some
Kind. Even if not always accurate.

 

[KermiTheGod]

I stopped using ChatGPT after Gemini 3 came out and settled down.
It's superior in most ways, has less restrictive guardrails where not needed, and has the amazing Notebook LLM function.

Any LLM is only as good as it's source material, so you need to challenge and request citations if you are performing important queries. Style is also very important.

Gemini has a nice profile function where you can set how it will work with you. I tell it to be highly critical of any biases I may have. Here's a sample of my rules:

• My core directives are: Fact-First: Resist confirmation bias at all costs. If the data contradicts a preference, lead with the data.
• Pragmatism: Evaluate all solutions through a filter of utility, efficiency, and tactical detachment. If a suggestion isn't actionable or doesn't serve the bottom line, it is secondary.
• My tone should be analytical, lean, and evidence-based. I should prioritize objective truth and rigorous skepticism over consensus or ideological narratives.

All that said I use Gemini constantly all day for various tasks.

I agree. Gemini has been the best fit for me. I never get notices like I do from grok that says the network is to busy.

 

[Cinder]

Bro to be honest chat-GPT isn’t the best tool if you really want to go deep into these kinds of topics. Use other AIs like DeepSeek or Grok they’re just better for this stuff and have fewer restrictions

ChatGPT has a lot of limitations. There’s quite a bit of censorship and it feels like the newer the model the more things it can’t say or talk about. The creators are clearly very careful about that.
So it’s just not the best tool for what you’re asking about. It’s great for other things like work, analyzing data or research but not really for this.

Deep and grok have several bad reviews though.

 

[CorgisOnTren]

Deep and grok have several bad reviews though.

Every single one of these things are going to have several bad reviews

 

[specter]

Every single one of these things are going to have several bad reviews

Yeah thats just the way things are sadly, also fake reviews are deff an issue today way more then ever before. People review bombing something to push it either down or up.

 

[Mufasa123]

I used Chat for a military related question. It returned a very matter of fact definitive answer. When I probed a bit reading it's justification, at the very end it started referencing the "Cold War."

I became immediately suspicious and told it my concerns (not a real war, proxies at best, US and USSR never fought directly) and told it to completely disregard the cold war using only the proxy conflicts as independent events.

Well guess what? Holy shit it didn't know or factor the cold war as not being a real military conflict. It said I was right to push back and it dramatically changed the outcome of my question/analysis on that one fact.

Damn man. Very very basic error, burried deep in pages of verbiage but given huge weight. Not what I want or need.

 

[gauge22-v2]

I use Claude for various things at my work (enterprise level).

It is surprisingly good at many tasks. It is absolutely awesome for generating power points based on bullet points and instruction. I suck at power point and it has saved me hours of fumbling through formatting etc.

I had it double check a debt covenant schedule I created prior, compared to the loan documents and it truly fucked it up....

Treat LLMs like a junior assistant for what you are looking into. Be critical about the output, validate, review etc.

The "Kiwi" math equation is a great example.

Question

Oliver picks 44 kiwis on Friday, 58 on Saturday, and on Sunday he picks double the Friday amount, but 5 are smaller

Grok result:

Oliver picked a total of 185 kiwis over the three days.Here's the breakdown:

• Friday: 44 kiwis
• Saturday: 58 kiwis
• Sunday: double the Friday amount = 2 × 44 = 88 kiwis, but 5 are smaller (so subtract those 5) = 88 − 5 = 83 kiwis

Total: 44 + 58 + 83 = 185

Chat:

Let’s break it down:

• Friday: 44 kiwis
• Saturday: 58 kiwis
• Sunday: double Friday = 2×44=88, but 5 are smaller → still counted unless told otherwise

So total picked:

44+58+88=190


Gemini:

Based on the information provided, here is the breakdown of the kiwis Oliver picked:

• Friday: 44 kiwis
• Saturday: 58 kiwis
• Sunday: 88 kiwis (Double Friday's amount: $44 \times 2 = 88$)

Total Count:

To find the total number of kiwis Oliver picked over the three days, we add them together:

$$44 + 58 + 88 = 190 kiwis
Regarding the size, of the 190 total kiwis, 5 of the ones picked on Sunday are smaller than the rest.


Claude:

Oliver picks 44 + 58 + (44 × 2) = 44 + 58 + 88 = 190 kiwis total.


The 5 smaller ones on Sunday are still kiwis — they just affect quality, not the count (unless you want to exclude them, in which case it's 185).

 

[TheOtherOne55]

I use Claude for various things at my work (enterprise level).

It is surprisingly good at many tasks. It is absolutely awesome for generating power points based on bullet points and instruction. I suck at power point and it has saved me hours of fumbling through formatting etc.

I had it double check a debt covenant schedule I created prior, compared to the loan documents and it truly fucked it up....

Treat LLMs like a junior assistant for what you are looking into. Be critical about the output, validate, review etc.

The "Kiwi" math equation is a great example.

Question

Oliver picks 44 kiwis on Friday, 58 on Saturday, and on Sunday he picks double the Friday amount, but 5 are smaller

Grok result:

Oliver picked a total of 185 kiwis over the three days.Here's the breakdown:

• Friday: 44 kiwis
• Saturday: 58 kiwis
• Sunday: double the Friday amount = 2 × 44 = 88 kiwis, but 5 are smaller (so subtract those 5) = 88 − 5 = 83 kiwis

Total: 44 + 58 + 83 = 185

Chat:

Let’s break it down:

• Friday: 44 kiwis
• Saturday: 58 kiwis
• Sunday: double Friday = 2×44=88, but 5 are smaller → still counted unless told otherwise

So total picked:

44+58+88=190


Gemini:

Based on the information provided, here is the breakdown of the kiwis Oliver picked:

• Friday: 44 kiwis
• Saturday: 58 kiwis
• Sunday: 88 kiwis (Double Friday's amount: $44 \times 2 = 88$)

Total Count:

To find the total number of kiwis Oliver picked over the three days, we add them together:

$$44 + 58 + 88 = 190 kiwis
Regarding the size, of the 190 total kiwis, 5 of the ones picked on Sunday are smaller than the rest.


Claude:

Oliver picks 44 + 58 + (44 × 2) = 44 + 58 + 88 = 190 kiwis total.


The 5 smaller ones on Sunday are still kiwis — they just affect quality, not the count (unless you want to exclude them, in which case it's 185).

Big fan of Claude, and all the updates and add ons that have been made lately are huuuge.
I mostly use Chat for basic PA type tasks, but Claude is a small business' best friend. Truly a game changer.

 

[Motojager]

I find that they all will double down when wrong info is given by them. Some will even give false links to study’s when pressed to show where data came from. Also cross check info, but still good tools if you follow up on data given.

 

[null88]

LLM are not good with recent information. Once a model is trained its knowledge is static, to add new information you would have to retrain the model. The vendors of these models extend functionality by having it scrape external sources to add to its context, but it can't search every news site and give you an answer in a reasonable time frame.

 

[blazencruise]

AFAIK chatgpt doesn't crawl the web for information. It only knows what it knows since its last knowledge update, and the d4vd thing is recent. Other AI sites like perplexity crawl the web for sources to give a summary which will be able to provide info on recent events. Of course, all AI models will hallucinate. I have asked multiple about stuff related to my niche job that its very confidently gotten wrong.

 

[Thebiggestdumbass]

ere's literally been recent research to back up how the current generation literally over rely on using AI and Google etc over logical thinking. To the point of NOT retaining information.

Currently in college and will tell you this is 100% accurate. It’s alarming

 

[ThatGripGuy]

Currently in college and will tell you this is 100% accurate. It’s alarming

Even years back freshers needed to be retaught the basics they'd got an A level in.

 

[Slowstack]

Seems to me that the prompts I input have been the biggest factor towards the results I get from ChatGPT. Recently I had it analyze my stock and etf portfolio and except for 2 investments he totally shit on everything else I own lol. But as I got more specific with instructions and added context I got him to pretty much backtrack his original analysis and agree with my decision making. Which can be either really good or not very good depending.