|
| ▲ | jonners00 3 hours ago | parent | next [-] |
| I can't tell you how many times a week claude opus 4.8 high effort has to apologise for being wrong when I'm asking it about something narrow and specific that i want it to research but it blurts out broad context from its training material and incorrect conclusions/assumptions. This is happening all the time. Someone needs to create a repository of its apologies to remind us all of its limitations. |
| |
| ▲ | dice 3 hours ago | parent | next [-] | | My company instituted a monthly "best use of AI" award to encourage people to share how they're using it. I suggested we should also have a "most wrong AI output" award to remind everyone they can't just trust it blindly but that hasn't happened yet for some reason ... | |
| ▲ | sarchertech 3 hours ago | parent | prev [-] | | I’ve noticed the same thing at work with Opus 4.8. ChatGPT on my personal plan does it too. Just yesterday I asked it to give some places fitting a specific criteria. The first was that they were within a 2 hour drive of my city. 75% of the locations it gave me were more than 2x that distance. It kept doing this across multiple difference searches. I tried high and pro with no difference. | | |
| ▲ | dice 3 hours ago | parent | next [-] | | I've found that Gemini with Google maps integration does this pretty well. | |
| ▲ | colonCapitalDee 3 hours ago | parent | prev | next [-] | | That's not surprising, LLMs are bad at pulling hyperspecific facts out of memory. LLMs aren't mapping applications, they're reasoners. Just a poor problem fit | | |
| ▲ | kensey an hour ago | parent | next [-] | | > LLMs aren't mapping applications, they're reasoners. No they aren't. They're statistical token generators. They do not understand concepts such as "distance from a given location or coordinate point". If you're lucky you might ask it something likely to appear nearly verbatim in its training data, like "Chinese restaurants in Midtown Manhattan", and get back a reasonably accurate list, but it does not understand what a "Chinese restaurant" is, or what "Midtown Manhattan" is, or that one relates to the other in any way other than both appearing statistically associated with another set of tokens when they appear near each other. | |
| ▲ | sarchertech 3 hours ago | parent | prev [-] | | I wasn’t asking it to pull it from its training data, I was asking it to search. Also reasoners that can’t recall facts is not how people are using them. No one is asking “from first principles derive this equation”. |
| |
| ▲ | trvz 3 hours ago | parent | prev [-] | | GPT 5.5 Pro is worth the increased price of the higher subscription. | | |
|
|
|
| ▲ | raincole 3 hours ago | parent | prev | next [-] |
| It's applied Cunningham's law: > the best way to get correct answers on HN isn't to ask questions, but to post LLM's answers so people will eagerly fact check them to prove LLMs wrong. |
| |
| ▲ | ChrisMarshallNY 3 hours ago | parent [-] | | I used to do this type of thing a lot. Boldly state something I knew to be wrong, so I'd get correct answers. Dang asked me not to do it. Now, when I boldly state wrong stuff (a not-infrequent occurrence), it's because I really am wrong. | | |
| ▲ | LPisGood 34 minutes ago | parent [-] | | I’ve tried this (not on HN) but usually it just results in downvotes or mockery. |
|
|
|
| ▲ | sporedro 2 hours ago | parent | prev | next [-] |
| The “claude says” is the part this pisses me off about LLM use now. If I wanted a Claude answer I’d ask Claude. If I ask a human I want a human response not a “here’s what Claude said”. |
|
| ▲ | marshray 3 hours ago | parent | prev | next [-] |
| > it’s becoming normal to just throw out “Claude says this” without doing any fact checking That would be an improvement over most people I know at this point, who casually repeat verbally or repost words they got from a chatbot without so much as a quotation mark. |
|
| ▲ | baxtr 3 hours ago | parent | prev | next [-] |
| I’ve noticed this too with things I’m quite familiar with. I’d like to say it makes me more cautious about topics I’m prompting that I’m not familiar with… But I’m also worried about the young people. What if you never had to learn something from ground up? |
|
| ▲ | albedoa 2 hours ago | parent | prev [-] |
| > it’s becoming normal to just throw out “Claude says this” without doing any fact checking. And without sharing the prompt or the actual response. Like sure, it's possible Claude said something so obviously wrong. Depending on your experience, you might even think it to be probable. But then why wouldn't OP preempt any doubt and simply share the part that matters? What are we doing. |
