If Claude Fable stops helping you, you'll never know • Hacker News Board

Re: If Claude Fable stops helping you, you'll never know

#401

Post by otabdeveloper4 » Wed, Jun 10, 2026, 8:52 AM UTC

somesortofthing wrote: post #29

This is a fun peek into the economic implications of RSI/ASI. Because it's so infinitely valuable that it basically destroys all markets, labs will eventually do stuff like stop releasing models completely and skipping out on contracted commitments because they'll have the power to just drive their competitors out of business before the legal battle gets expensive. Cloud providers - at first smaller ones, then the hy…

> Because it's so infinitely valuable that it basically destroys all markets

We have 8 billion natural intelligences already. (Each of them more intelligent than any LLM.)

For some reason this didn't destroy all markets. There's also diverging opinion about infinite value.

Re: If Claude Fable stops helping you, you'll never know

#402

Post by otabdeveloper4 » Wed, Jun 10, 2026, 8:54 AM UTC

HoldOnAMinute wrote: post #172

platinumrad wrote: post #60

Earlier quoted context omitted.

Nothing is infinitely valuable.

10 engineers can make a billion dollar company. One Claude can replace 10 engineers. This gets very close to "infinitely valuable", it starts to look like a vertical line to me

Damn, now if only I could find 10 engineers!

A billion bucks, here I come!

Re: If Claude Fable stops helping you, you'll never know

#403

Post by otabdeveloper4 » Wed, Jun 10, 2026, 8:56 AM UTC

shelled wrote: post #393

I was doing something with Claude today and it just told me "By the way Cowork is a separate desktop app" and it proceeded to explain to me how it is not part of the standard Claude desktop app and how the plugin I am exploring might not be a great fit for me. I actually ended up having to search around and see whether things had changed that much in last 24 hours. It hadn't. It beats me how can their tool hallucinat…

> do they perform a lot of painting job on their tools to hide the cracks?

Yes. That is what RLHF is.

It works magically if your prejudices happen to match their training set alignment.

Re: If Claude Fable stops helping you, you'll never know

#404

Post by viking123 » Wed, Jun 10, 2026, 8:58 AM UTC

gck1 wrote: post #284

torben-friis wrote: post #37

They have a silent nerfing system for their models and say so openly. The obvious question is how much it is being used already. Competitor companies being nerfed? Non Americans getting worse code? Punishing and rewarding users to maximize engagement, like online games do affecting victories through matchmaking?

No big pockets and ask it to review your own codebase for security issues? You hacker. Ban. Anthropic simply can't be allowed to succeed. This is the most E Corp shit I've seen since I've been alive.

Amodei is worse psychopath than Altman, and by far

Re: If Claude Fable stops helping you, you'll never know

#405

Post by SwellJoe » Wed, Jun 10, 2026, 9:00 AM UTC

DeathArrow wrote: post #352

theLiminator wrote: post #151

Earlier quoted context omitted.

> But, history says the supercomputer of today will fit in your pocket in a few years. I don't think this will be true in the same time span anymore. Each miniaturization is costing more and more money. Perhaps they'll come up with exotic fundamental improvements, but I don't think the rate of improvement of compute/watt will match the previous decades.

>but I don't think the rate of improvement of compute/watt will match the previous decades. Unless we invest heavily in research and find new way to do chips. But I think there's not enough motivation and money to do that.

There's literally never been more money being thrown at that problem.

Re: If Claude Fable stops helping you, you'll never know

#406

Post by dhx » Wed, Jun 10, 2026, 9:02 AM UTC

There's an example at [1] of a prompt for a HTML mockup operating system where 3 applications are requested to be "white hat tools" that show diagnostic system information. Claude Fable 5 is shown and said in the video to switch back to Opus 4.8 as a "safety" feature.

What an utterly useless model if it refuses to work on something as benign as basic system diagnostic utilities (nmap or whatever).

[1] https://youtu.be/9GLYsrMpprs?t=305

Re: If Claude Fable stops helping you, you'll never know

#407

Post by knrdev » Wed, Jun 10, 2026, 9:14 AM UTC

Imagine this company getting real power. This is just a purest nightmare evil shit i've seen out of any of them. Maybe they're already controlled by Slophos.

Re: If Claude Fable stops helping you, you'll never know

#408

Post by SwellJoe » Wed, Jun 10, 2026, 9:14 AM UTC

Ferret7446 wrote: post #254

SwellJoe wrote: post #77

The moat looks deep today but it's going to become more shallow every year. Training a new model from scratch takes serious resources. Post-training/fine-tuning an existing model, dramatically less. The knowledge for the process was esoteric two years ago, now you can ask a current model (one of several) to walk you through it, while building the tools to do it as you go. Several of my recent weekend projects have be…

The moat is not the model, it's the harness. I wager that's one of the main reasons why Google made Antigravity closed source.

I don't feel strongly about anything most folks are arguing back and forth about, but this one is obviously wrong.

Everybody and their brother has made an agent. There are toolkits. You can whip one up in an afternoon.

Not only that, I've found models often perform worse, or at least cost more and take longer, in a big complicated agent like Claude Code, including Anthropic models. They want proprietary doodads hanging off the side (multi agent orchestration, memory, things of that nature) to matter, because they can lock you into tools like that. But, top models can do everything with bash.

Re: If Claude Fable stops helping you, you'll never know

#409

Post by SwellJoe » Wed, Jun 10, 2026, 9:17 AM UTC

Paradigma11 wrote: post #364

dudisubekti wrote: post #279

Earlier quoted context omitted.

But harness is relatively easy to code yourself? They're just system prompt composer, with some tool functions that the LLM can invoke. I've vibe coded my own in just one day.

But is there anything preventing them from putting their own proprietary wolfram alpha/prolog/super duper expert system in there?

Only that it would just slow down the model and make it dumber.

You can't tool and harness a weak model into strength and you probably don't improve top models with boondoggles.

Re: If Claude Fable stops helping you, you'll never know

#410

Post by morpheuskafka » Wed, Jun 10, 2026, 9:20 AM UTC

Notably, it says they will notify you if they downgrade your responses due to suspected distillation (trying to reverse engineer their model).

But if you merely ask it questions about the process of developing a new model ("for example, on building pretraining pipelines, distributed training infrastructure, or ML accelerator design") that's where it will silently downgrade your replies.

Not by falling back to an older model, but "limit effectiveness through methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning (PEFT)." So in some cases, they will silently rewrite your prompt!