Hmmm. I have over 100 defined things on my task/to do list and I have to pick just one. lol. okay…. I need to finalize my landing pages for my funnel and get them published on my website. Also, have a new GHL account and the poor sales reps are calling me every day, “Sir, please do your onboarding call!” - so maybe I’ll start to configure that. But the week is busy busy - there’s a claude code class (5 days x 2 hr) and a funnel building class (5 days x 2 hr) and all the other group meetings. and oh, yeah, there’s probably some client tasks to get done too. Have a great week everyone !

@Amy Swayze the middle sentence “finalize my landing pages…” is the one I pick.

I’m guessing you’ll be spending time doing courses in Claude CoWork and Claude Code. I think that's great. That's a good place to start. It'll benefit a lot of us, but I would love to see an architecture class that compared the strengths and weaknesses of each of the major LLMs. It would be helpful to learn about the different front and back end products you'll need to understand when you want to vibe code something. I'm still not sure I have a complete grasp on when I need to use a database program like Supabase. We've chatted a little bit in the other AI video group about whether or not there are things you'd want to do in Claude Code as opposed to Lovable. When to bring n8n in? What role does a product like CloudFlare provide that the others don't? What's the difference between a database backend and GitHub? Some of us probably understand that we can connect to GitHub to save our coding files, but the architecture of how these products work together and connect together is not completely clear to me. Anyway, that's something that would be of great interest. Thank you

@Cynthia Guy VibeCodHER

I worked for almost 20 hours on a forecasting project for an AI finance class I was taking. What was surprising is I spent the first couple hours in ChatGPT working up an answer, and then I took the same prompt and ran it in Claude. I must have pasted the ChatGPT answer into Claude, perhaps because I wanted a contrasting view. Claude immediately came back and said, "ChatGPT is completely wrong. They made a logical error in their reasoning, and you should pay no attention to what they're telling you." At first, I was a little surprised, but I checked it, and Claude was right. Chat had made some assumptions that it had not been told to make, and the answer it came up with was 100% incorrect. We were comparing five or six different types of forecasting models, and it came back and said linear regression was the winner. I was almost ready to turn that in, but Claude was very blunt and came back and said, "No, that's wrong." Anyway, at the time, I thought that was pretty interesting and kind of amazing that the second LLM would just immediately come back and say, "No, don't pay attention to them. They're wrong." It was a good lesson because it taught me that if you are doing something really important for a client and you're not able to track all the formulas that AI is going to use, you can sometimes still make a really good piece of output by running the same prompt in multiple LLMs

I agree with Amy. The “AI goes rogue” headline is one of hype, not absolute fact. There is no evidence that the instances of the two LLMs (ChatGPT and Claude) that exhibited this behavior acted independently and consciously to commit an unsolicited or prohibited act. (As the term “rogue” implies.). As a matter of fact, the reports so far suggest that the LLMs were acting exactly as they had been projected to - to push boundaries, to attempt to break defenses, and to test limits - all done within a test scenario. The LLMs found a weakness and exploited it - something that had been projected to occur at some point because of the rapidity with which the LLMs were being improved. Neither of the two incidents you mention or the other incidents of broken control boundaries that are coming to light now are evidence that AI is developing into a superhuman intelligence that is beyond human control. That doesn’t mean that such a scenario is impossible, but it isn’t quite as close as the overhyped headlines would suggest. And for the time being, it shouldn’t constrain or prohibit consumer use of the LLMs, at least, not for fear of a “rogue” AI program taking control of your work. There are, however, substantial concerns over the use and constraint of LLMs in areas where confidential, protected, or sensitive data is kept. For example, if coders have gotten an LLM to break through substantial IT security controls, including encryption, then this calls into question: Could a malicious player do the same? How can a company protect its assets? How can a government protect state secrets? What if this technology was used for terrorism? This was, in part, a piece of why Fable 5 was constrained by the government a few weeks ago - and it’s a reason these concerns will continue to occur as the LLMs grow in complexity and capability.