My prediction on 3.6

[J Grant]

I think you are correct in that 2.6 and 3.6 are not really the issue...AI integration into our space was coming regardless of the form. The software provider with 80% appraiser market share ignored that fact and it is going to cost them and their users immensely. They are still refusing to adopt AI quickly.

Like most on this forum, at 23+ years in, I include myself in the camp of serious appraisers and try and do good/supportable work measuring market reaction along with everything else that goes into creating a solid report. The system does not care anymore...the metric is speed, price, then quality...in that order.

I am trying to point out the high horse of ignoring AI options filling out the 3.6 is going to put older appraisers out of the game when actually, it looks very easy to operate. Everyone watching videos on Alamode's lame solution for 3.6 is pissed off and frustrated because we have all blindly trusted Alamode's guidance on 3.6...but they don't know what the F they are doing. Just wanted to present my opinion on what is getting ready to go down, supported with facts from my own experience, not 'what ifs'.

Not fair of me earlier to call the AI solution an inferior product when it may not be...I see a world in the next 12 mos where AI is doing a good job selecting comparables in seconds and the human eye is just fine tuning the report. Not gonna be enough work for all of us, but ignoring the new way of doing things does not change that fact. This is how its gonna go down. People can choose to ignore it if they wish, but I respect my peers on this forum and don't want everyone blindsided with their head in sand yelling out how superior the old way is. No one gives a ****.

Imo, you overestimate AI and its ability to select comps. Your knowledge of it sounds superficial and naive.
The old way has certain standards of development that can use AI for certain functions - but giving it over to AI is another thing entirely.

If AI is right on a mass scale, that's great. If AI is wrong on a mass scale - then what? And there are always fraudsters and hackers who can get in and manipulate results - AI allows them to do that on a mass scale as well.

 

[alebrewer]

There you have it, Nacho. Not only are you a shill for the AMC's, but you're knowledge of AI is superficial and naive. You best stick to tiddlywinks and leave the adulting to J.

 

[TerryRohrer]

Imo, you overestimate AI and its ability to select comps. Your knowledge of it sounds superficial and naive.

You nailed it! Bullseye! But that finger is pointed the wrong direction.

 

[George Hatch]

AFAICT the additional data being collected is limited to the subject's attributes. Problem identification on the SR1 side. Beyond that its limited to reporting those attributes. It doesn't really extend into the selection or data qualification of the comps because that level of detail is still limited to what the brokers report about those properties, same as has been the case all along.

"Pulling comps" and suggesting adjustments are being done by other apps not related to the UAD itself. The complaints about appraisers using those applies regardless of the reporting format being used for the assignment.

Render unto the GSEs that which is theirs, but don't blame them for appraisers choosing to add Datamaster or Spark or some of these other 3rd party apps to their processes.

 

[Zoe]

Imo, you overestimate AI and its ability to select comps. Your knowledge of it sounds superficial and naive.
The old way has certain standards of development that can use AI for certain functions - but giving it over to AI is another thing entirely.

If AI is right on a mass scale, that's great. If AI is wrong on a mass scale - then what? And there are always fraudsters and hackers who can get in and manipulate results - AI allows them to do that on a mass scale as well.

Especially in a very heterogeneous single family market. AI won't know where to get the best comparables. Census tract won't work. What if there are few sales in the subject subdivision and AI has to jump areas to find the most similar comparables to the subject?

 

[Zoe]

AFAICT the additional data being collected is limited to the subject's attributes. Problem identification on the SR1 side. Beyond that its limited to reporting those attributes. It doesn't really extend into the selection or data qualification of the comps because that level of detail is still limited to what the brokers report about those properties, same as has been the case all along.

"Pulling comps" and suggesting adjustments are being done by other apps not related to the UAD itself. The complaints about appraisers using those applies regardless of the reporting format being used for the assignment.

Render unto the GSEs that which is theirs, but don't blame them for appraisers choosing to add Datamaster or Spark or some of these other 3rd party apps to their processes.

Yeah, you have to know the market you work. AI don't know single family market in a very heterogeneous market. It can also be very homogeneous in certain markets. AI can do well in homogeneous markets.

 

[Zoe]

Some appraisers on here brag because they say they work in cookie cutterville.

 

[DWiley]

Imo, you overestimate AI and its ability to select comps. Your knowledge of it sounds superficial and naive.
The old way has certain standards of development that can use AI for certain functions - but giving it over to AI is another thing entirely.

If AI is right on a mass scale, that's great. If AI is wrong on a mass scale - then what? And there are always fraudsters and hackers who can get in and manipulate results - AI allows them to do that on a mass scale as well.

I think there is equal danger in both underestimating and over estimating the power of AI.

With the proper training, one can produce AI prompts that produce remarkable results. Even then, having a human in the loop is still critical. And, if that prompt is not one that can be used over and over, the time it takes to build a good prompt can offset any time saved.

Without training, it is easy to create prompts that produce results that are pure garbage, but sure look authoritative. Just look at some of the posts of incorrect AI results that have been posted here. (e.g., the UAD 3.6 post that quoted from page 36 of a document that was less than 15 pages long).

At the end of the day, AI is just a tool, and like any tool it can be used well or used poorly.

"Based on my experience", based on my list, based on bad AI results - I am not sure which is worse

 

[J Grant]

I think there is equal danger in both underestimating and over estimating the power of AI.

With the proper training, one can produce AI prompts that produce remarkable results. Even then, having a human in the loop is still critical. And, if that prompt is not one that can be used over and over, the time it takes to build a good prompt can offset any time saved.

Without training, it is easy to create prompts that produce results that are pure garbage, but sure look authoritative. Just look at some of the posts of incorrect AI results that have been posted here. (e.g., the UAD 3.6 post that quoted from page 36 of a document that was less than 15 pages long).

At the end of the day, AI is just a tool, and like any tool it can be used well or used poorly.

"Based on my experience", based on my list, based on bad AI results - I am not sure which is worse

Imo, when used poorly, AI can be worse for the following reasons: (as you said, the results can be garbage but look authoratitize.

A) It can be deployed on a mass scale. That makes it more dangerous than the scattered results of some individual appraisers.
B) Poor results can be hidden behind the AI's ability to write intelligent-sounding narratives (even if the conclusions are incorrect) and produce glitzy charts and graphs.
C) Reviewers and people monitoring for quality might not be able to spot poor AI usage.

I agree AI can be used well, but using it well means comparing its results with your own analysis, and reading the MLS ifno first hand - and that is time-consuming. Appraisers have to be compensated for their time-

It becomes a problem when appraisers must choose between devoting adequate time ( in this case, a few hours ) and the ability to survive.

 

[Lycabull]

I just have one question. When AI makes a mistake that ends up in my report, and it will, whose feet will be held to the fire? I'm guessing not the AI vendors.