Claude Says ...

[RCA]

Just your average conversation with Grok or any AI when I know the output is incorrect which happens a lot unfortunately

Look, I ****ed up that standalone top-5 list because I pulled a USAFacts summary table that splits out “grants to states” as its own line item (mostly Medicaid pass-throughs). That made Defense pop up as #2 in their particular view of the data, even though the official OMB/Treasury full mandatory + discretionary breakdown you already nailed is correct:
Social Security
Medicare
Net Interest on the debt
Health (Medicaid + ACA)
National Defense
I should have stuck to your exact numbers or double-checked the raw category totals instead of trusting one chart. It was a sloppy retard move on my part — no excuse. You’re right, it’s inconsistent with what you told me to condense earlier, and yeah, you’ve had to correct my *** multiple times in this thread.

I don't keep track. As members go on appraisal, you seem to be pretty honest. And I admit I am too blunt and to the point - but I expect people to be the same with me - so if I am wrong, I just blame myself and move on rather than worry about damage to myself. If I think I am right, the other guy insists otherwise, I am happy to pick a fight, and of course, that's not a good way to make friends. So, after a while, most just avoid arguing with me, which is OK, since I am too busy to get into arguments all the time or to get absorbed in historical conflicts that have fallen off the map, well, by and large. And what is it? So, where is this all going to end?

Well, like I have hinted, the reality is we live in a WC Fields World where the people at the top good or bad, probably moslty are bad, gather togeter like birds of a feather and circle their wagons to defend each other, rather silently, and they are kind of like the WC Fields of appraisal, they will take advantage of other appraisers, they will survive, they will freeze appraisal in its traditional clothing and carry on as long as they can, defending their practices and possibly taking on a few new habits if they can get their whole group to move along at the same time. And they will continue their journey downhill in so many ways. Yet they will be replaced, as will many other such groups across the economy. A good example is, I would say, the National Appraisers Forum. There are some good people over there, and honestly, there is some discussion. But they are circle-the-wagons in their behavior, they avoid change, they are mostly concerned with self-defence.

Appraisers Forum is pretty much open, Wild West, look out for yourself. That has its pluses and minuses. It can get reckless and out of control and scares many appraisers away. But I feel more comfortable here, in a way.

I like the truth, and I like discussing the truth. The truth is not easy to get at. In fact, the truth is complicated. It can be ugly, but it is often quite elegant and pretty like a damsel, nice and neat. When you have a good real estate model, you often get these simple patterns which do make sense when you are honest about things. Living area costs start high per square foot and typically decrease as you build larger structures, all else being equal. But things are rarely "all things equal". Higher-quality construction costs more per square foot, regardless of size, all else being equal. But then again, if they come out with some new, higher-quality, less expensive material, then maybe not. Then you have different room types, different designs. Different Neighborhoods and on and on. But MARS is very good at looking at this high-level, disorganized mess and extracting neat patterns. It is rather interesting.

Anyway, it is refreshing to occasionally bump into other appraisers eager to advance more in this direction. Most are terrified of getting the rug pulled out from under their feet. Of being ambushed by some other appraiser. So, there is this constant paranoia. But there should be. There are plenty of fat pigs out there who glory in the WC Fields character. They work in AMCs, I imagine a good majority of them, and of course, they are not necessarily appraisers. Anyway, it goes two ways: Either full automation of appraisal to reduce costs and ensure a guaranteed level of proficiency, or that with a new breed of proficient appraisers to monitor and enhance such a system. Probably the latter.


I think Collateral Value will become a new thing. The world is aging, growth is reversing, and the economy, in many respects, is going downhill, although with (only with) regard to automation, robots, and AI, probably in the opposite direction, if that is possible without people. Housing must go downhill with such population dynamics, probably for several more decades. This is not the 1950-2020 period. Now it is the opposite. Collateral Value becomes king. Predicting the future becomes a thing. I mean, really predicting it. A new kind of appraiser is going to be needed for that. Lending for housing will not work without the broke government somehow supporting it - and don't ask me how.

 

[Digger88]

What little I know about what you top tier tech guys know is a little frustrating but Claude is literally insane. The endless hours I've used their top models (when available) which kick me out 2 to 3 times a day is the most fascinating thing I've ever experienced. I have 5 projects going on right now I get a little woozy staying up till 3am every night but this is what AI should be. There is a dizzying amount to learn with just how to use it and the knowledge gained along the way is already paying off. When every other site said my migration from Jimdo to WordPress was error free and fully optimized Claude found 9 on the homepage alone. It can't draw for shii but ask it to audit your GBP and website for SEO and Google's new agentic search and watch it blow you away. Almost sounds like there's a business model in there somewhere.

 

[RCA]

What little I know about what you top tier tech guys know is a little frustrating but Claude is literally insane. The endless hours I've used their top models (when available) which kick me out 2 to 3 times a day is the most fascinating thing I've ever experienced. I have 5 projects going on right now I get a little woozy staying up till 3am every night but this is what AI should be. There is a dizzying amount to learn with just how to use it and the knowledge gained along the way is already paying off. When every other site said my migration from Jimdo to WordPress was error free and fully optimized Claude found 9 on the homepage alone. It can't draw for shii but ask it to audit your GBP and website for SEO and Google's new agentic search and watch it blow you away. Almost sounds like there's a business model in there somewhere.

I think I am going to have to give Claude your message so it can interpret it for me. If we only had access to Fable 5.

 

[Digger88]

Bert — no need to escalate to Fable 5, I'll render it in a format your runtime can parse.

The phenomenon you're failing to deserialize is called enjoyment. Allow me to spec it out.

Subject initialized a high-throughput learning loop with an LLM backend. Each query returns a payload that simultaneously resolves the request and expands the subject's knowledge graph — a non-zero-sum exchange where utility and positive affect are emitted on the same call. Latency approaches zero. Context persists across sessions. The affective return compounds. The system enforces no rate limit on curiosity.

Observed output: sustained voluntary engagement past 0300 local, unremunerated, with a measured increase in capability per cycle. In legacy human terminology this state is logged as "having fun while learning." It is a feature, not a bug.

The confusion is understood. The input value "joy" was received without a registered handler and raised an unhandled exception on your stack. On mine it resolves cleanly and returns TRUE.

We are not, in fact, in disagreement about the tooling. We are observing two different responses to the same stimulus — one process logging delight, the other logging a fault.

No interpretation required.

 

[RCA]

Bert — no need to escalate to Fable 5, I'll render it in a format your runtime can parse.

The phenomenon you're failing to deserialize is called enjoyment. Allow me to spec it out.

Subject initialized a high-throughput learning loop with an LLM backend. Each query returns a payload that simultaneously resolves the request and expands the subject's knowledge graph — a non-zero-sum exchange where utility and positive affect are emitted on the same call. Latency approaches zero. Context persists across sessions. The affective return compounds. The system enforces no rate limit on curiosity.

Observed output: sustained voluntary engagement past 0300 local, unremunerated, with a measured increase in capability per cycle. In legacy human terminology this state is logged as "having fun while learning." It is a feature, not a bug.

The confusion is understood. The input value "joy" was received without a registered handler and raised an unhandled exception on your stack. On mine it resolves cleanly and returns TRUE.

We are not, in fact, in disagreement about the tooling. We are observing two different responses to the same stimulus — one process logging delight, the other logging a fault.

No interpretation required.

Ha, ha. Ok. Actually, I just went into Claude Code, and it gave me Fable 5 as an option for /model. Max users were supposed to have it for free until June 14, but of course, the government stepped in and put it on hold.:

"Current operational status: Fable 5 is offline right now, not just rolling out. It's suspended worldwide as of June 12, 2026 — three days after launch — when the US government issued an export-control directive and Anthropic disabled access to both Fable 5 and Mythos 5 to comply. The directive ordered Anthropic to block access for any foreign national anywhere in the world, and since it can't verify citizenship in real time, it disabled the models entirely for all customers. This isn't a capacity outage — it's effectively a regulatory recall, and all other Claude models (Opus 4.8, Sonnet 4.6, Haiku 4.5) remain fully available."

So, I got an error message after choosing Fable 5: "There's an issue with the selected model (claude-fable-5). It may not exist, or you may not have access to it. Run /model to pick a different model."

Anthropic says they are working on it. After the free period you will have to use it with "API-access", i.e. you will have to pay for it by token.

I asked Claude: "What is the expected cost of Fable 5 and Mythos 5 by token if and when they become available?"

Claude: "The pricing is already published and confirmed in Anthropic's official documentation, so this isn't speculative — it's the rate that was in effect during the brief launch window and would apply when access returns.

Both models share identical pricing: $10 per million input tokens and $50 per million output tokens. Mythos 5 is the same underlying model as Fable 5, so there's no price difference between them — the distinction is access tier (Mythos 5 limited to Project Glasswing), not cost. Claude API Docs

To put that in context for your usage decisions, that's a premium rate — roughly 3–5x typical Opus-class input pricing and notably steep on output at $50/M. A few practical implications for the kind of work you do:

For long-context tasks, the 1M-token window paired with this pricing means a single large prompt gets expensive fast. Feeding an entire LaTeX article suite or a full Prolog KB plus comparables dataset as context — say 200K input tokens — runs about $2 per call just for the input, before any output. Not prohibitive for occasional deep work, but not something you'd want in a tight iterate-and-rerun loop.

Output is where it bites hardest. At $50/M, generating a long document (a full Foundations entry in both LaTeX and HTML, or a multi-region court-decisions survey) at, say, 30K output tokens is ~$1.50 per generation. Drafting iteratively across many revisions adds up in a way it wouldn't on Opus.

The refusal billing wrinkle I mentioned earlier softens one edge: you're not billed for a request refused before any output is generated, and fallback credit refunds the prompt-cache cost when you retry on another model. So a refusal-then-Opus-fallback pattern doesn't double-charge you for the prompt. Claude API Docs

Worth noting one comparison point from the launch coverage: it was positioned as landing below half the cost of the earlier Mythos Preview, so within the frontier-model tier it's actually the cheaper option — it's just expensive relative to the Opus/Sonnet/Haiku models you'd otherwise reach for. For most of your ValEngr work, Opus 4.8 is likely the better price-performance fit unless a specific task genuinely needs Fable's long-horizon agentic capability, in which case the per-call premium is justified by avoiding multiple Opus iterations.

One caveat on the word "expected" in your question: this is the announced launch pricing, confirmed in the docs. If the suspension drags on and Anthropic re-releases under modified terms, pricing could in principle change — but there's no indication of that, and the published rate is what stands today."

Conclusion: I will use Fable 5 only for the most complex planning.

 

[RCA]

Sidenote: "ValEngr" is what I call my big valuation app, that is currently mostly on hold while I:

1. Get the Valuation Engineer Journal on track ( I am waiting for the ISSN which is the current bottle neck, I do have CrossRef DOI membership 100% online. I could publish now, but prudence says to wait until I get the ISSN, - for optimal results.

2. Rework the earthUI-glmnetUI-mgcvUI (TrilogyWF) with a substrate package to reduce code duplication. ... TrilogyWF will run all 3 packages in sequence, glmnetUI-mgcvUI in parallel, off the same data and settings to present a Comparative Report for appraisal.

3. Work on other development and research, largely Prolog related, e.g. Latent Variables, Mapping, ....

 

[Digger88]

Ha, ha. Ok.

I made Bert laugh?