Stumped by AI? Where Machines Fall Short: Share What They Get Wrong and the Challenges They Can't Handle! - ChatGPT, Gemini, Claude, Deepseek & LLAMA
Commented on Mon Apr 21 12:40:39 2025 UTC
https://i.redd.it/g8lrtb37o6we1.jpeg
I feel this quote sums it up perfectly. Also i do not use AI or chat GPT.
https://i.redd.it/g8lrtb37o6we1.jpeg
I feel this quote sums it up perfectly. Also i do not use AI or chat GPT.
│
│
│ Commented on Tue Apr 22 14:35:49 2025 UTC
│
│ Summaries of tv episodes and questions about what happened during them is where I was getting a lot of wrong answers. Some of the shows were older, but there should have been sufficient information for it to learn from.
│
│ I’ve also seen people ask it for crochet patterns and it understands how a crochet pattern should look on paper, but they do not translate at all into a real object, nor do it understand the physical realities of yarn.
│
│
│ Summaries of tv episodes and questions about what happened during them is where I was getting a lot of wrong answers. Some of the shows were older, but there should have been sufficient information for it to learn from.
│
│ I’ve also seen people ask it for crochet patterns and it understands how a crochet pattern should look on paper, but they do not translate at all into a real object, nor do it understand the physical realities of yarn.
│
CodeKansas
Superstar
I don't really be taxing Gemini like that, just asking some research questions and doing image prompts.
Usually when it goofs, it's more frustrating than funny.
Usually when it goofs, it's more frustrating than funny.
Over... and over... and over...
1/11
@mattyglesias
I keep having conversations where people speculate about when AI will be able to do things that AI can already do.
2/11
@CostcoSwizzle
AI can do loads of things 98% well, but if you have complex input and output, you spend more time auditing the 2% than if you had just done the thing yourself in the first place.
3/11
@BillyLincoln10
I wondered when AI would be able to replace an underwriter entirely. Decided to find out how far away we are. I spent 90 minutes playing around with Grok. At the end of that I had a complete digital underwriter. I was shocked. I still am. I now have infinite staff.
4/11
@bdmarotta
Conversations? You mean published articles by journalists?
5/11
@scottdavisCRE
I will be interested in when AI can replicate results without wildly fabricating them with “helpful completions.”
6/11
@AaronErickson
The upside is I feel like I am an agent from the future when people ask me what I do for a living, which is kinda fun.
7/11
@GrdDscntN2Madnz
To be fair, it's hard to keep up
8/11
@Raynorus
I'll tell you right now, it cant pass chem 105
9/11
@iamarsibragimov
AI achieving what capability is misunderstood; realism matters
10/11
@StillCompeting
In many ways the real questing is: When will people let AI do what it can do, and what is the path to getting there?
11/11
@NateSilver538
There's a lot of room to disagree on where AI will end up in (1, 2, 5, 10, 20 etc.) years but I don't think I've seen a subject where a cohort of people who like to think of themselves as highly literate and well informed are so proud of their ignorance.
To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196
@mattyglesias
I keep having conversations where people speculate about when AI will be able to do things that AI can already do.
2/11
@CostcoSwizzle
AI can do loads of things 98% well, but if you have complex input and output, you spend more time auditing the 2% than if you had just done the thing yourself in the first place.
3/11
@BillyLincoln10
I wondered when AI would be able to replace an underwriter entirely. Decided to find out how far away we are. I spent 90 minutes playing around with Grok. At the end of that I had a complete digital underwriter. I was shocked. I still am. I now have infinite staff.
4/11
@bdmarotta
Conversations? You mean published articles by journalists?
5/11
@scottdavisCRE
I will be interested in when AI can replicate results without wildly fabricating them with “helpful completions.”
6/11
@AaronErickson
The upside is I feel like I am an agent from the future when people ask me what I do for a living, which is kinda fun.
7/11
@GrdDscntN2Madnz
To be fair, it's hard to keep up
8/11
@Raynorus
I'll tell you right now, it cant pass chem 105
9/11
@iamarsibragimov
AI achieving what capability is misunderstood; realism matters
10/11
@StillCompeting
In many ways the real questing is: When will people let AI do what it can do, and what is the path to getting there?
11/11
@NateSilver538
There's a lot of room to disagree on where AI will end up in (1, 2, 5, 10, 20 etc.) years but I don't think I've seen a subject where a cohort of people who like to think of themselves as highly literate and well informed are so proud of their ignorance.
To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196