20 Minute AI Appraisals Are Coming

[cpursley]

https://.AI.meta.com/blog/llama-4-multimodal-intelligence

Interesting, it looks like Meta has gone the Deepseek route via mixture of experts approach.

Llama 4 Models:
- Both Llama 4 Scout and Llama 4 Maverick use a Mixture-of-Experts (MoE) design with 17B active parameters each.
- They are natively multimodal: text + image input, text-only output.
- Key achievements include industry-leading context lengths, strong coding/reasoning performance, and improved multilingual capabilities.
- Knowledge cutoff: August 2024.

Llama 4 Scout:
- 17B active parameters, 16 experts, 109B total.
- Fits on a single H100 GPU (INT4-quantized).
- 10M token context window
- Outperforms previous Llama releases on multimodal tasks while being more resource-friendly.
- Employs iRoPE architecture for efficient long-context attention.
- Tested with up to 8 images per prompt.

Llama 4 Maverick:
- 17B active parameters, 128 experts, 400B total.
- 1M token context window.
- Not single-GPU; runs on one H100 DGX host or can be distributed for greater efficiency.
- Outperforms GPT-4o and Gemini 2.0 Flash on coding, reasoning, and multilingual tests at a competitive cost.
- Maintains strong image understanding and grounded reasoning ability.

Llama 4 Behemoth (Preview):
- 288B active parameters, 16 experts, nearly 2T total.
- Still in training; not yet released.
- Exceeds GPT-4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro on STEM benchmarks (e.g., MATH-500, GPQA Diamond).
- Serves as the “teacher” model for Scout and Maverick via co-distillation.

Misc:
- MoE Architecture: Only 17B parameters activated per token, reducing inference cost.
- Native Multimodality: Unified text + vision encoder, pre-trained on large-scale unlabeled data.

Between Llama, Deepseek, Qwen and some others, it seems like the "open" model is going to win the AI race vs closed approach of ChatGPT and Claude.

 

[J Grant]

I know in the future it will improve, but to date, all AI has done for me is be a naggy PITA. I see my doctors and dentist use it, as do retailers -to send messages about appointments, over and over, and pesky emails about just for me!! about a product recomnedation.

It stinks at spell-check because it thinks it is smarter than the writer and substitutes words and garbles my posts here.

AI imparts a sense of existential dread that it will take away our jobs. Its creative efforts are vapid and generic, plagiarizing what already exists.

Its search engine answers are bland and offer nothing we couldn't find on our own within one or two clicks.

It has done one, and literally only one thing well the entire time I have interacted with it, and that was to remind me there was no PDF file attached to an Email where I mentioned a PDF file was enclosed. I thought that was impressive - a nag looking over my shoulder in my own house - but it did do that one thing well, so thanks, AI -

I agree we need oversight and regulation, which Trump will not do - he has no long-term vision and is on the side of the billionaire overlords who profit from AI. Musk did warn about regulations to get ahead of AI, but I'm not sure where he stands on it now ..

 

[glenn walker]

I don't think it will take over, still need a human in the loop.

Yes son the data inspectors are humans and the two appraisers in the cubicle spend 8 hours a day completing five reports a day and signing them. The others are waivers and the complex and rural are sourced out to appraisers who compete full 1004 with them doing the inspection themselves.

In Summary Rural- Complex and areas that aren't miles of tract style cookie cutters will mostly need traditional appraisals. So yes there will always be appraisers just way fewer and not ones who in the past could afford staff and keep 5 or 10 working full time.

If your located in the right part's of the country then it's still a viable career but in my area it's a hard life for the average appraiser to make enough to live on as fees are often no higher than what's paid in Kentucky but they have cheap homes and low rents.

 

[Vermonter]

https://.AI.meta.com/blog/llama-4-multimodal-intelligence

Interesting, it looks like Meta has gone the Deepseek route via mixture of experts approach.


Between Llama, Deepseek, Qwen and some others, it seems like the "open" model is going to win the AI race vs closed approach of ChatGPT and Claude.

Just watched a video yesterday from the CEO of Ollama and he's shocked at how fast local hosted models are being adopted and how they're advancing things even faster than the bigs

 

[cpursley]

I see my doctors and dentist use it, as do retailers -to send messages about appointments, over and over, and pesky emails about just for me!!

How are these AI? Sounds like old-school marketing automation to me.

 

[J Grant]

How are these AI? Sounds like old-school marketing automation to me.

I can tell the difference because I have been with these same ones for years -

 

[cpursley]

I can tell the difference because I have been with these same ones for years -

Sounds like they might be trying to "personalize" the writing with AI?

 

[Non Sequitur]

Anyone test drive this yet? They are working with Fredie to hit the ground running with UAD 3.6.

https://www.aivre.com/

 

[J Grant]

Anyone test drive this yet? They are working with Fredie to hit the ground running with UAD 3.6.

https://www.aivre.com/

Sounds like a version of SPARK or other auto fill software that syncs with MLS or public records.

They write "AI trained -" like really, what kind of training did they do to the learning that makes it work any differently - its a pay as you play program that will suck more money from appraisers in return for some more "efficiency"- but will not increase your volume, none of these programs can do that. Eeither higher volume or higher fees is the only way to increase income, and these programs tend to lead to the opposite of that -

 

[jay trotta]

Will Cost/ Investment = profitable end result?

a concern to whomever is left in the business