This is my biggest concern. How will junior people get the need experience. The tacit knowledge that helps you identify when AI creates garbage and also when it gives you something that will not work in the real world.
4 Likes
The nice thing about Gemini is that it gives links to where it found the information it uses in any recommendations. 95% of the time they are Reddit posts. So you are really just short cutting the Reddit search process. But for some reason Gemini doesn’t notice if Reddit posts are up or down voted. So when I asked it how to do something and then checked the link to the post that Gemini used to confidently tell me the answer, I could see that the post was heavily downvoted and had a bunch of responses explaining why it was wrong.
This is often why I find AI giving out very bad information. And that is very different from asking specific questions on data it is being fed on internal systems like using AI for patient summaries.
1 Like
I was hiding my ineptitude. Truth be told, it is often more than that. I moved my sprinkler water intake line from a drainage ditch 20 feet away from the pump to the lake which was 200 feet away - and I probably made at least 5-6 trips for various parts.
We have been talking about this a lot at work. AI takes a lot of the tasks that entry level programmers used to do. Well, you don’t get experienced programmers with knowledge of our business if you don’t hire entry level people. It isn’t enough to say someone else can train them, we need people with knowledge of our business.
I am old enough this isn’t going to bite me, but I can see it becoming a real issue in a few years. It is a solvable problem but I am not sure if bean counting executives focused almost entirely on immediate profit will recognize it before it becomes an issue.
1 Like
this x1000
I’ve encountered this too. But at least when I confronted it it admitted that it made a mistake instead of reminding me of the time that I forgot to pick up the kids from dayschool.
3 Likes
“AI is just like my wife in that I enjoy the cold clinical embrace of a machine, but with a pleasantly shorter memory” is unlikely to be a popular opinion at home.
2 Likes
When I first started my apprenticeship, most houses were built with 2 types of nails. That’s a pretty easy list to remember.
Now we use more types of nails, different types of specialty guns with their own fasteners, different brands of guns might use the same nails but ones which are collated at different angles, multiple types of structural screws, adhesives and epoxies, and a gazillion different structural connectors for every type of wood to wood/concrete/steel connection imaginable.
It’s a lot easier to forget a thing or two.
I would need a dedicated drone to run back-and-forth to the hardware and material store.
1 Like
But don’t the robot dogs leave oil puddles everywhere until they’re fully trained?
1 Like
Maybe one of those smiling Uber eats delivery bots.
So you’re the reason new homes are so expensive and slow to get built, I knew it!
Maybe one of the reasons! 
I think you mentioned before you are in the GW space? How do you find the AI you are using at writing Gosu? Do you use CLI? GitHub copilot?
At my place we are all in right now on using AI. Our devs are rightfully nervous but some of the things we are doing with it is pretty cool. Some of the agents and skills we have created can write really good code for “simpler” tasks. As you mentioned above it also will write unit tests really well.
Our vision for our software development life cycle is wild. AI agents taking meeting notes (one example) turning them into requirements, passing off to a coding agent, then passing off to a couple of code review agents before finally getting a human involved. Will be interesting to watch.
Like you I’m at the “end” of my career and happy about that.
With the capabilities listed above, it feels like we are all at the end of our careers, whether we plan on retiring or not…
Yeah I guess we’ll see. I think there will be a bunch of new jobs that we are not aware of right now created in the near future.
There is also the cost structure that needs to be understood better. Right now I think these AI companies are bleeding money. Sometime soon they will need to make money and the pricing models that companies like mine are using will reflect that. Will we be able to afford it? I know right now we aren’t concerned about our AI costs but talks are already starting about what the future will look like.
I started as a dev with languages like Java and Powerbuilder. Languages where the “compiler” code was abstracted out. Devs before me actually had to write the compilers. This shift to AI is “similar”. Writing code in all these different languages is being abstracted out to just use plain English.
I made the comment about liking the fact that I’m at the tail end of my career cause I’m not sure I can adapt to these changes! 
Young people should be more able to adapt.
Yes it was very helpful in design and materials list. Also provided some materials that I didn’t know existed like specific Simpson Strong Tie products that were perfect. I had true 6” posts that I need to connect to a laminated beam of 3 2x6x10 beams that it had me make to carry the load.
What was also nice having the skews was Home Depot has a delivery service for free that is overnight for me. So I ordered most items and had them delivered to the house. I did make several trips to get additional items like it missed on the Simpson screws for Hurricane ties. Needed way more than estimated.
It also had me order a few other connectors that I returned no questions so recovered cost.
It’s pretty amazing what it can do but also need to be able to review and understand the results to make sure they are accurate.
I recently used it for a dining table I’m restoring and the associated veneer that I need to apply. I needed a stack up to gain thickness and it gave me a solution to the problem that seems pretty logical to me.
Good luck
We are using Copilot for the most part. The AI does pretty well. You have to iterate through a couple times and tell it where it messed up, really just pasting the error you are seeing. But it doesn’t do too bad.
I’m liking it for Splunk queries too. It tells you that it doesn’t know splunk but if you give it some hints it can get it. And when the person asking for the data doesn’t know shit you can look somewhat smart with getting stuff back to them quick.
sorry, pedantic, but SKUs
these are cool applications for AI, and as you note, must be reviewed.
teaching it what it got wrong, on a global scale, seems to be missing. “ok claude, I didn’t catch it at the time, but I built your patio cover exactly as you said, and X Y and Z are all fucked up. don’t make this mistake again.”
also, sabotaging AI with misinformation is very much a thing, and that raises a whole 'nother set of challenges.
Microsoft is moving us to a per query billing model. And the memo came out immediately that we need to make sure we are using our AI queries more “mindfully”. We have gone from you need to use AI for everything as much as you can and if you don’t use it enough you will be on a naughty list to watching how much we are using in a matter of days.