I’m unimpressed by certain aspects of AGI. The level of mistakes by all programs I’ve tried for data specific analysis is horrendous, and renders them useless for financial analysis or anything that requires accuracy. The impressive aspects are more geared to creative things like video generation.
Try asking Claude to provide a list of oil filter part numbers for your car from different manufacturers. Or ask it to provide financial comparison data for publically traded companies involved in underwater systems like oil drilling or military related systems. Then check the results.
Rick, you raise very valid points and I have experienced similar. When LLMs are used for specific tasks like number recall, counting letters in a word, or doing very precise technical calculations, they can easily fall over and hallucinate and then they are adamant that they are right. Then you end up having an argument with it!
I have found these issues when working in subsurface data and projects
But where LLMs genuinely perform well is reasoning, summarising, drafting and explaining complex concepts. Give them a document or a dataset and ask them to interpret it, and the results can be impressive. Admittedly there is still a real risk of hallucination, but things are improving every few months.
I wrote about this exact thing recently based on Ethan Mollick’s idea of the Jagged Frontier. It’s where AI performs well on certain tasks but falls flat on its face with others.
My wife also homeschools our kids, and sounds a lot like your wife. Full semester planning, Singapore Math, Logic of English, etc. I also teach “Coding with Dad” every Wednesday.
I’m going to borrow this one to get homeschool set up in Cowork.
@Michael Crist, in this World of Information OVERLOAD, I Love this Practical “Get Sh%t Done” Method of laying Everything Out for us. This Was GREAT! I was tracking with your every step, thoroughly and comprehensive, just like you wrote it. You’ve just earned yourself a new Fan, Follower and Potential Subscriber! 🤝 And that’s at a time when I’m ELIMINATING Subs, and Follows to try and Take Back My Devices and Mental Sanity. So WELL DONE My Friend 👏🏾👏🏾👏🏾, and THANK YOU!
This exploration of using Claude Cowork sounds intriguing, especially in a world where collaboration tools are rapidly evolving. It reminds me of the ideas from my recent piece on AI societies in "Let There Be Claws," where we delve into how these new platforms can reflect broader social dynamics, particularly regarding inequality. If you're curious, you can check it out here: https://theuncomfortableidea.substack.com/p/ai-societies-mirror-human-ones-but.
The last thing I’m going to do is give AI unfettered access to my computer.
I understand the hesitation, I felt it, too.
But it's the direction everything is going. At some point, I think all of us will have to cross the rubicon
I’m unimpressed by certain aspects of AGI. The level of mistakes by all programs I’ve tried for data specific analysis is horrendous, and renders them useless for financial analysis or anything that requires accuracy. The impressive aspects are more geared to creative things like video generation.
I'm surprised to hear that. I'm having great results with large data sets
Try asking Claude to provide a list of oil filter part numbers for your car from different manufacturers. Or ask it to provide financial comparison data for publically traded companies involved in underwater systems like oil drilling or military related systems. Then check the results.
I literally used Claude today to build a 2025 P&L for one of my companies, using nothing but full year exports from bank and credit card transactions.
It didn’t mess up a SINGLE line item (I double-checked)
Rick, you raise very valid points and I have experienced similar. When LLMs are used for specific tasks like number recall, counting letters in a word, or doing very precise technical calculations, they can easily fall over and hallucinate and then they are adamant that they are right. Then you end up having an argument with it!
I have found these issues when working in subsurface data and projects
But where LLMs genuinely perform well is reasoning, summarising, drafting and explaining complex concepts. Give them a document or a dataset and ask them to interpret it, and the results can be impressive. Admittedly there is still a real risk of hallucination, but things are improving every few months.
I wrote about this exact thing recently based on Ethan Mollick’s idea of the Jagged Frontier. It’s where AI performs well on certain tasks but falls flat on its face with others.
Good points
My wife also homeschools our kids, and sounds a lot like your wife. Full semester planning, Singapore Math, Logic of English, etc. I also teach “Coding with Dad” every Wednesday.
I’m going to borrow this one to get homeschool set up in Cowork.
I’m dropping some homeschool stuff in my next post
Built a REALLY COOL skill that I think will help a lot of people
@Michael Crist, in this World of Information OVERLOAD, I Love this Practical “Get Sh%t Done” Method of laying Everything Out for us. This Was GREAT! I was tracking with your every step, thoroughly and comprehensive, just like you wrote it. You’ve just earned yourself a new Fan, Follower and Potential Subscriber! 🤝 And that’s at a time when I’m ELIMINATING Subs, and Follows to try and Take Back My Devices and Mental Sanity. So WELL DONE My Friend 👏🏾👏🏾👏🏾, and THANK YOU!
Thanks so much for the kind words 🙏
Glad to be even a small part of your success
Content is fairly basic, but may be helpful to novices.
Let’s hope so !
I love the real-life examples! Thanks, Michael !
You got it :)
Thanks. Easy to understand and implement!
Glad you found it helpful! 🙏
This exploration of using Claude Cowork sounds intriguing, especially in a world where collaboration tools are rapidly evolving. It reminds me of the ideas from my recent piece on AI societies in "Let There Be Claws," where we delve into how these new platforms can reflect broader social dynamics, particularly regarding inequality. If you're curious, you can check it out here: https://theuncomfortableidea.substack.com/p/ai-societies-mirror-human-ones-but.
Thanks for sharing 🙏