Yeah, that’s like the worst version of whatever Google has.
Ask Jeeves a.k.a. ask.com (a website answering natural language questions) has been around since 1996, so not such a new thing. Maybe AI answers questions more successfully but it still gets a shipload of stuff wrong.
The more insidious question is not whether AI got the answer right or wrong. Often the questions being asked of AI are not that clear cut - else, of course, we could just Google it. The real question is whether you and/or the AI converged on an answer because it was factually correct, or because you traversed a thoroughly convincing trajectory of evidence, counterevidence, trust, distrust, assumptions, biases… and so on.
(**Still definitely NOT the topic of the thing 
)
2 Likes
The 100m robots still have to work on the slowing down technique (although the 400m one at the end did a nice little jig to finish). I wonder if Usain Bolt could have run faster if they’d given him a crash pad at the finish line.
I’m using AI right now so extensively, to build out pretty much every aspect of my business. I can’t keep up with how much I have to get done. I have months of work to do just to get my current ideas out of my head, nevermind starting big picture planning. And that gives me a bit of anxiety that I’m behind, that others in my industry are ahead and getting further ahead.
my friend, who is about as much ‘into’ using AI as I am, told me that cursor actually visited his office and told him that his company is one of the fastest companies in the world at implementing AI. if I assume I’m about where he is, just smaller, I guess I’m not behind the curve, I’m likely way out in front.
my buddy’s actually dumped a ton of corporate info into his ai and is using it to monitor and help him manage.
3 Likes
Testament to the fact that my buddy is out of my league here (I mentioned in another thread that last month he was rubbing shoulders with Ryan Renolds, Chris Hatfield, PM of Canada, etc, then goes and has lunch with me)…he mentioned open weight models, basically models that you can download and run your own AI stuff. And these models are almost as good now as the commercial ones.
I told him, yeah, I looked at that, looked at the hardware needed to run one, and a single server is going to be $1MM. He nodded, yup, he’s got two on order.
I’ll have to make a trip back to Ottawa and get a tour of the server room when he’s got them installed and running. Mostly just to look at them with ignorance and amazement, like most of the stuff he does.
Been there. Tell your Cowork to keep track of that shit in a spreadsheet. Eventually you’ll make it to the point where ideas have to earn their spot in your roadmap. Then, you’ll have to reverse the over-conservatism as nothing makes it in. It will be a viscous cycle.
1 Like
I could have told you just from your posting here that you’re well ahead of the curve. 
1 Like
Agreed.
I’m certain that almost no other people in my branch of the industry are actually running their own ai servers, so OK. But still…anxious lol.
1 Like
Not a surprise, but…
I have two software projects sort of running. On one I have an actuary who’s acting as a user, plus a developer, I’m in the middle translating. But the project’s been running from months and I’ve lost touch with some of the specs. The other, my inhouse crm I built with claude code.
The difference in debugging is insane. I swear I’ll never use a developer alone just because of the difference in debugging.
With the first project, the user sends me a vague email. I look at it with a puzzled expression, and send it to the dev. who wrestles with it, makes a change, then sends it back for testing. Users says ‘nope, that didn’t work’. 3 days of back and forth with a partially annoyed developer that I keep sending miniscule changes to. Everyone’s sick of it, just make it work.
On my crm I was doing some pre-breakfast cleanup stuff and ran into a crash. Fired up claude code, gave it the error, it diagnosed and corrected in 2 minutes. Yeah, it just worked. 2 minutes vs three days and arguably the 2 minute change was more complex.
I wish Claude could debug itself in 2 minutes.
But alas I have spent all morning copying back and forth from GPT-B (RIP GPT-A I MISS YOU BRAH) to CLA-A, who apparently can’t code worth shit this morning.
(Yeah, that’s right - I’m sad about my ChatGPT hitting max conversation length. Sorry not sorry.)
Me to serena: You just have to learn Claude! Then you don’t even need a developer! Software costs aren’t an issue! And claude’s easy to learn!!
serena: naaaaaaaa. I’ll just call one of you nerds.
lol.
Actually she didn’t say that. she said I was similar to her other tech guy then later called him a nerd.
1 Like
Yeah, how to kick start your country a leap year ahead of everyone one else - give them a tool that makes many of them 10X more productive.
My son texted me asking if in a few weeks I’ll sit down with him and help him build an app. Not sure for what, but I’m interested to see what he’s thinking up.
2 Likes
curious what he’s interested in making
Linking it to government services is genius. Great way to improve efficiency and boost productivity.
The UK is so far behind on that front.