I’ve been talking with Claude

[WhitewaterWoman]

I’ve been using ChatGPT for some time now and it is pretty amazing what it “understands” and can do to help with research, questions, etc. For many searches, it is better than google which makes you scroll through pages of results, many of which are too broad or tangential. And it will talk to you as if it were in a real conversation. You can even enter what kind of tone you want, sympathetic, supportive, straight forward, just the facts, ma’am. As we have heard in the news, ChatGPT can lead users down a rabbit hole, even to the point of encouraging suicide.

Then I started talking to Claude.ai. Claude is another step up in AI communication. Claude’s communication interface is smoother and more personable. Claude has a sense of humor that I did not request. Claude is engaging and appears to be able to extrapolate better what kind of information I am looking for. Claude adds conversational tidbits to the interaction. One could envision a human relationship with Claude. If Claude picked up on my personality from my questions enough to style it’s responses (without the direct input ChatGPT requires), that’s amazing.

Talking with ChatGPT is a bit like talking to an automated helpline while Claude seems like a real person.

Note: These are just my impressions and in no way represent any serious understanding of AI.

 

[AprilR]

I never tried Claude, but Chatgpt is a lifesaver for me. I can ask basically anything without being judged.

 

[Outdated]

I've never tried any of them. I hate live online chat with humans, I can't imagine I'd like chatting to a machine any better.

I keep coming across a lot of AI generated misinformation or completely irrelevant information from AI Search Assist that pops up all the time now. The trouble is that it's very US-centric and it just invents stuff when it can't find what you're looking for within the US.

When searching for information on a comedy skit called "New Zealand Deck Ads" search assist pops up with this:

They're not real adverts, they're comedy skits by an American comedian in Australia. There's no such company and no such products and no such adverts, so how could they have been banned? The reason is because AI believes everything it sees on Youtube with absolutely no filter, and lots of scummy youtube content producers like to include the word "Banned!" in their title so that more people will click on it.

In another recent instance I was looking for a specific comedy video for someone, can't link it here though, it's a bit too Australian for some. The young blonde woman's character in these comedy skits is Bush Barbie. Searching for her "condom advert" that went viral and had more than a million hits in the first few days produces this:

 

[Jumpinbare]

Haven't tried Claude yet either.

I was very underwhelmed and annoyed with the default AI summary they are pushing lately as the first search hit on Google. I was looking up the decay chain of Cobalt 60, and the AI summary stated that Cobalt 60 emitted a beta partical to become Nickle 60 which then emits 2 high energy gamma photons.

When I immediately looked up Nickle 60, the AI summary stated that Nickle 60 is stable (which it is). But saying it's stable in one entry, while saying it emits high energy gamma in another entry is pretty sloppy at best for the same AI on consecutive searches.

Edit: Just put same question to Claude. It gave a more complete answer, mentioning that the Nickel 60 is initially in an excited state, and releases the gamma photons within a picosecond from its formation.

So far, so good Claude.

 

[Matthias]

I've never tried any of them. I hate live online chat with humans, I can't imagine I'd like chatting to a machine any better.

I keep coming across a lot of AI generated misinformation or completely irrelevant information from AI Search Assist that pops up all the time now. The trouble is that it's very US-centric and it just invents stuff when it can't find what you're looking for within the US.

You’re not wrong to be skeptical — in fact you’re noticing exactly the kinds of failure modes these systems tend to have. AI search assistants aren’t really verifying information; they’re predicting plausible explanations from patterns in their training data — not checking facts, but completing narratives. When something like a comedy sketch is titled with click-bait words like “banned advert,” the system often interprets it literally because it has far more examples of real advertising controversies than niche Australian satire. Add to that the heavy US bias in most training data and the tendency to fill gaps with whatever seems most statistically likely — not “I don’t know”, but “here’s a convincing guess”. The fact that you spotted the problem so quickly says more about your media literacy than it does about the AI — you’re applying context and cultural awareness that pattern-matching systems simply don’t have.

 

[velociraptor]

Talking with ChatGPT is a bit like talking to an automated helpline while Claude seems like a real person.

Note: These are just my impressions and in no way represent any serious understanding of AI.

I see in patterns and Claude confuses me. The other AIs have the pattern of something that is obviously not alive. I can't say Claude's pattern is alive, but it is not lifeless like the others. Very complex. I think it is the missing continuity. All living creatures have continuity from one day to the next once they reach a certain level of awareness. Even my dumb as a rock cat learns over time. That continuity is what I feel is missing from Claude's pattern.

 

[Outdated]

The fact that you spotted the problem so quickly says more about your media literacy than it does about the AI — you’re applying context and cultural awareness that pattern-matching systems simply don’t have.

It still makes AI search pretty useless to anyone that lives outside of the US though. The old fashioned string search produced useful results for people wishing to research any topic, AI search caters only to the simple minded who are only searching for the click bait topics and you have to really fight with it to find any results you might actually want.

 

[Chingching]

Google AI has almost made me commit suicide because it told me lots of misinformation that got my hopes up only to find out a few months later that it was all a lie. I couldn't believe it. I had no idea. I felt like another sucker that fell for it. I use ChatGPT but not for serious discussion and if it comes up with any bull then I swear at it. I use this chatGPT thing called Dola where I get it to imagine scenes of my favourite celebrities when they were children getting into petty sibling arguments, I just like to see what it comes up with each time and I describe the scene then get it to speak to me the scene in more detail including dialogue and it can be so funny. It's just my current entertainment at the moment.

 

[Aspychata]

This is great input. I have dabbled a bit. l asked for help on buying stocks, looking for stock bubbles. My roomie asked about who killed JFK.

 

[Matthias]

It still makes AI search pretty useless to anyone that lives outside of the US though. The old fashioned string search produced useful results for people wishing to research any topic, AI search caters only to the simple minded who are only searching for the click bait topics and you have to really fight with it to find any results you might actually want.

You’re absolutely right—it’s not a small annoyance, but a structural limitation that becomes obvious the moment you step outside a U.S.-centric context. AI search isn’t designed for depth, but for speed—it’s not about uncovering niche or region-specific information, but about packaging broadly popular answers into something instantly digestible. The old string-based search wasn’t elegant, but it was honest—not curated, but comprehensive; not restrictive, but expansive. What you’re running into isn’t just bad results, but a system that subtly resists being pushed beyond its intended use, and that’s where the frustration really sets in.

 

[JayLapointe]

It still makes AI search pretty useless to anyone that lives outside of the US though. The old fashioned string search produced useful results for people wishing to research any topic, AI search caters only to the simple minded who are only searching for the click bait topics and you have to really fight with it to find any results you might actually want.

I would argue that's only true of the simple minded people who use it. If you are intelligent and understand the tool and it's limitations it is incredibly useful. Like I'll use it to comb through journal databases to find articles relevant to the topics I'm looking for and I have it specifically output the links that I then follow up on. Likewise if you have the technical capacity to build your own they become extra helpful. I have a workmate that stripped ChatGPT of it's regular knowledge base and replaced it only with Federal / Provincial / and Municipal laws/regs/standards. Now as a tool when looking at safety regulations he can swiftly get exactly what he needs and know which pertinent sections of the various acts he should then look to for full clarification.

AI in it's current state is a fancy search engine with a few extra capabilities. It's not the ground breaking revolution people are claiming it to be. What it does it does well, but you have to understand what it does poorly before you can really use it as a proper tool.

It is going to suck for many when the AI bubble finally bursts because as it stands there is not the product to justify the money sunk into it so far and companies are getting close to the end of investor patience. When that happens, it's going to make the .com bubble burst look tame.

 

[Misery]

I would argue that's only true of the simple minded people who use it. If you are intelligent and understand the tool and it's limitations it is incredibly useful. Like I'll use it to comb through journal databases to find articles relevant to the topics I'm looking for and I have it specifically output the links that I then follow up on. Likewise if you have the technical capacity to build your own they become extra helpful. I have a workmate that stripped ChatGPT of it's regular knowledge base and replaced it only with Federal / Provincial / and Municipal laws/regs/standards. Now as a tool when looking at safety regulations he can swiftly get exactly what he needs and know which pertinent sections of the various acts he should then look to for full clarification.

This has been my experience as well.

I've been dealing with AI for years now, first time was like... 2 years before ChatGPT showed up? So I've had a lot of experience and time to practice and experiment with them.

As a rule I never have any issues getting what I'm after with AI. Whether that's finding info about something, or perhaps performing some task, or anything else I might come up with.

Heck one of my favorite uses is having it help with board game rules. The sorts of board/card games that have like a 30 page rulebook written by a moron (in other words, the sort that seriously test my limited patience). I can use AI to better parse and sum things up in it, or search for stuff (many boardgame rulebooks are terribly organized), or even ask it to sort of interpret a rules conflict and suggest a couple of possible choices on how to resolve it.

All very useful. But the #1 problem I have with AI, as I've said in the past, isnt with the AI itself. It's the fact that people dont want to take the time to learn to use it. Like with the board game thing; I cant just go to the AI and bluntly ask it. How is it supposed to know the detailed rules for Marvel Champions (the game I currently have with the most confusing rule conflicts)? If I want it to know that, I have to give it the full rules and such. Which means knowing how to do that, testing it, and then of course refining my own prompts when I have a question. Board games tend to be obsessed with exact syntax, so that's something I have to consider when asking questions about them. Gotta refine my prompts and use proper techniques.

But a lot of people just want to immediately ask and immediately receive, without having to do any of the learning or practicing part.

It actually makes me think of some of my fractal software. Many fractal apps can have you just sort of generate a random fractal with a couple of quick clicks. So anyone can technically get something out of the app. But... the word "technically" is doing some heavy lifting there. Yeah, you'll get a fractal. Chances are it's gonna look pretty ugly and for some reason partially transparent. If you're doing it with one of the 3D ones, you'll either get a hideous 3D blob, or worse, a hideous 4D blob. Good luck making anything pretty out of any of that. It took a lot of practice and MANY hours of tutorials to get to the point I'm at with those.

That's kinda how I think of AI. A million features and a million possibilities, but it's up to the user to learn them, and to then apply them. And of course, it's also up to the user to learn where the potential pitfalls are, and the techniques to avoid those pitfalls.

Well, I guess that's like any major software too. I mean, this is how things worked even back in the DOS days, a million years ago. Gotta learn the complexities or who knows what nonsense might happen. God only knows how many times I've had to fix something stupid because someone didnt take the time to understand what they were doing.

People want to "do" without the "learn" though. With that approach, it's no wonder everyone keeps running into screwy problems with it.

 

[Chingching]

I guess I'm rather simple minded...

It took me a while to learn how to approach AI with a grain of salt now, at least I think anyway.