Fernando
Elite Member
- Joined
- Nov 7, 2016
- Professional Status
- Certified Residential Appraiser
- State
- California
I got a postcard supposedly related to utility company and will be coming to my block.
Don't know if it's a scam.
The Premier Online Community for Real Estate Appraisers!
-
Welcome to AppraisersForum.com, the premier online
community for the discussion of real estate appraisal. Register a free account to be able to post and unlock additional forums and features.
Is AI too honest for appraisal work?
- Thread starter Elliott
- Start date
Don't know if it's a scam.
Carnivore
Elite Member
- Joined
- Jan 15, 2002
- Professional Status
- Certified Residential Appraiser
- State
- North Carolina
The best way to deal with AI is to use it. -AI Copilot is Microsoft So it is pretty tame. The free version I use is pretty good and super fast. But it only knows what it knows. Google uses AI as an assist. You can find things fast. Many times AI will give links to news paper articles and you don't pay for a stupid subscription.
I have heard stories about kids and chat chatgpt oh well
I have heard stories about kids and chat chatgpt oh well
Fernando
Elite Member
- Joined
- Nov 7, 2016
- Professional Status
- Certified Residential Appraiser
- State
- California
Copilot is most convenient in fast access to AI.
Just like MSN.
Fernando
Elite Member
- Joined
- Nov 7, 2016
- Professional Status
- Certified Residential Appraiser
- State
- California
That's good to know but I just want to get quickly the correct zoning for my subject.
cashcatz
Senior Member
- Joined
- Feb 17, 2015
- Professional Status
- IT Professional-Appraisal Related
- State
- Georgia
Try Perplexity or Claude if you want your mind blown. And MSN is not news, it's agi-prop.
RCA
Elite Member
- Joined
- Jun 27, 2017
- Professional Status
- Certified General Appraiser
- State
- California
For programming, right now, Claude Code is considered in the lead.
repomix + repo prompt + Claude code (sonnet or opus) is a good combination.
1. Repomix can take a snapshot of your entire repository that you can provide to Claude Code.
2. Repo Prompt is for context engineering.
This is just the tip of the iceberg.
If you get a $100/$200 month plan for Claude, then you've got some real horsepower - and it is a question of learning out to use it. As you might imagine, you can also make a very big mess of things.
This one guy was bragging that he created 30 agents and had them all working at the same time. --- Well, probably on different code bases.
What is an agent? Basically, you design a plan of attack for Claude that is restricted to a certain set of code. So, you might have an agent for working on the backend and one working on the front end. You carefully tell them what you want done. It will likely be a multi-step process: Start out doing this, then based on the results, do the next thing, and so on. You have to know everything it could bump into and prevent it from doing just that before it does.
For each agent, you create a separate terminal and do a git checkout -b <project name>/frontend to create a branch for the agent to work on. You might have 2-3 agents for a project. They each work on a branch - and you will eventually have to merge them. And well, it become a question of managing all the agents and keeping them from messing up the other's work --- because cleaning up a mess is no fun.
So after you really get into this "agentic" think, you will be busy as a bee trying to understand, control, and fix everything. You see AI stumbles - because you stumble in not setting up the context well enough. And like people, it does make mistakes (it's smart enough to know you do have to risk making mistakes to make progress).
Some things I really appreciate about Claude Code:
1. It is nearly perfect at fixing Git check-in problems (which can get rather complex).
2. It is pretty darn good at figuring out needed unit and integration tests and generating code. I have a ton of test code for "calibrating" my system.
3. It is pretty darn good at figuring out all kinds of problems.
4. It can easily work in many languages. It generates Angular, Typescript, JavaScript, CSS, R, Python, C++, Markdown, HTML, and other languages for me. I may start using Fortran if my Neural Network processing gets too slow (I'm using Python now, will probably move to C++ for Pytorch ... but if all else fails, then I'll dive into Fortran, which is, surprisingly, quite a bit faster than C++.
....
- For those few on the forum.
Apparently, this is being called by some: "The Era of the Idea Man". (e.g., Altman).
That's really kind of true. These agents are essentially like hired software engineers -- at least if you can keep them busy - not always easy. But even if you can't keep them working 8 hours a day, they will get a lot done in a short period of time.
For those using the software:
1. The software should be a lot less buggy in the future.
2. The software could be a lot more complex, but their will be simpler versions around
3. Changes will happen faster.
4. You are likely to get your desired features added, well, if the engineers think they make sense.
.....
repomix + repo prompt + Claude code (sonnet or opus) is a good combination.
1. Repomix can take a snapshot of your entire repository that you can provide to Claude Code.
2. Repo Prompt is for context engineering.
This is just the tip of the iceberg.
If you get a $100/$200 month plan for Claude, then you've got some real horsepower - and it is a question of learning out to use it. As you might imagine, you can also make a very big mess of things.
This one guy was bragging that he created 30 agents and had them all working at the same time. --- Well, probably on different code bases.
What is an agent? Basically, you design a plan of attack for Claude that is restricted to a certain set of code. So, you might have an agent for working on the backend and one working on the front end. You carefully tell them what you want done. It will likely be a multi-step process: Start out doing this, then based on the results, do the next thing, and so on. You have to know everything it could bump into and prevent it from doing just that before it does.
For each agent, you create a separate terminal and do a git checkout -b <project name>/frontend to create a branch for the agent to work on. You might have 2-3 agents for a project. They each work on a branch - and you will eventually have to merge them. And well, it become a question of managing all the agents and keeping them from messing up the other's work --- because cleaning up a mess is no fun.
So after you really get into this "agentic" think, you will be busy as a bee trying to understand, control, and fix everything. You see AI stumbles - because you stumble in not setting up the context well enough. And like people, it does make mistakes (it's smart enough to know you do have to risk making mistakes to make progress).
Some things I really appreciate about Claude Code:
1. It is nearly perfect at fixing Git check-in problems (which can get rather complex).
2. It is pretty darn good at figuring out needed unit and integration tests and generating code. I have a ton of test code for "calibrating" my system.
3. It is pretty darn good at figuring out all kinds of problems.
4. It can easily work in many languages. It generates Angular, Typescript, JavaScript, CSS, R, Python, C++, Markdown, HTML, and other languages for me. I may start using Fortran if my Neural Network processing gets too slow (I'm using Python now, will probably move to C++ for Pytorch ... but if all else fails, then I'll dive into Fortran, which is, surprisingly, quite a bit faster than C++.
....
- For those few on the forum.
Apparently, this is being called by some: "The Era of the Idea Man". (e.g., Altman).
That's really kind of true. These agents are essentially like hired software engineers -- at least if you can keep them busy - not always easy. But even if you can't keep them working 8 hours a day, they will get a lot done in a short period of time.
For those using the software:
1. The software should be a lot less buggy in the future.
2. The software could be a lot more complex, but their will be simpler versions around
3. Changes will happen faster.
4. You are likely to get your desired features added, well, if the engineers think they make sense.
.....
RCA
Elite Member
- Joined
- Jun 27, 2017
- Professional Status
- Certified General Appraiser
- State
- California
In the program I am working on, users can create tasks and connect them in workflows. Fairly complex ones with conditions and so on. It would be nice for Commercial and Ag. ... PROLOG. Lots of Prolog. And the appraiser can modify the default Prolog programs (not difficult).
In fact, the appraiser can write inspection facts in Prolog, so that they can be picked up by the program, organized, and reported.
OK, so it is not done until it is done and successfully tested in the field. But it is doable, for sure.
In fact, the appraiser can write inspection facts in Prolog, so that they can be picked up by the program, organized, and reported.
OK, so it is not done until it is done and successfully tested in the field. But it is doable, for sure.
Last edited:
OSU Beavers
Elite Member
- Joined
- Jan 10, 2007
- Professional Status
- Licensed Appraiser
- State
- Oregon
I looked at one AI appraisal demo that used crime map data to estimate value, gasp!
RCA
Elite Member
- Joined
- Jun 27, 2017
- Professional Status
- Certified General Appraiser
- State
- California
The problem has always been that the statisticians and software engineers who do the analysis or coding don't understand much at all about appraisal, and most appraisers can't manage the necessary statistical analysis or coding. You would think they could talk to each other - easier said than done!!