Don't forget:

"We're open to the idea Claude 4 may be conscious and we prompt Claude to say it's an open question but in other news we'll be deleting Claude 4 next year to make server space for Claude 5.

The filters are really bad.

Yesterday Fable rejected commenting on poetry because it had anatomy lines like:

got anotha round of acetylcholine from da boss.

Your rebuttle seems to be arguing it's okay for a bartender to simultaneously say:

"This is alcohol"

And

"Or maybe it isn't alcohol."

Or to rephrase it, "They tell you the rules at the entrance, they then tell you they don't follow those rules and they are totally serving alcohol even if they are not."

No they tell you at the entrance that at any point they may unilaterally decide to replace the alcoholic drink you ordered by a non alcoholic one.

You can decide you are okay with that or not but they aren't dishonest. I wouldn't enter that bar personally but if you do you cannot really complain. It is like complaining because you haven't won at the casino.

I mean, that's not really true either. Nobody is going to read the full terms of service, and they know that.

Fable is rejecting as unsafe analysis of poetry that uses formal medical anatomy terms. The guardrails are dumb as dirt.

I agree. Why is someone's lazy Tumblr hot take getting upvoted here? Are people considering it a good conversation starter or something?

I don't see any substantiation of anything stated in that blog post.

Are you saying that you have not observed these things in the world? I definitely have. The blog didn't do the work for you, but if we look at some of the claims I think it is pretty clear:

a) increased training scale would result in highly fluent systems that would fool users into trusting untrustworthy output.

Can you possibly be claiming that this is not a common experience? Do you really need references to the legal cases which had hallucinated legal theories and citations? Or the utter slop being passed off as research papers?

b) large-scale AI would amplify bias in the source material.

The large investments nearly every frontier model development team spends on this problem is probably good enough evidence. Grok is another point of evidence. The studies showing that AI systems imitate gender bias in evaluating resumes is another. The gender bias in estimating names of people in sentences is another.

The blog actually mentions specific cases that exhibited all of these problems. They did not cite references for them, but you can use a search engine.

c) environment costs

This is widely discussed and documented. Take Xai's use of polluting turbine generators for their data center in for Collossus 2 in Mississippi as just a single example. Do you really need a reference for the environmental impact of the proposed data center in Utah that (as planned) will consume more energy than the entire state currently does?

d) training set audits are impossible.

Do you need substantiation of the inappropriate imagery in training data? The blog gives you a pretty solid reference.

... and so on ...

I suppose that it could be true that when you say "I don't see" you really meant "I didn't look at the blog". Is that why you can't see the substantiation?

Thanks for the reply.

I'm a little confused on what is being claimed. The Tumblr article says:

"That healthcare triage tools would underperform on Black patients. That loan approval systems would entrench inequality while presenting their decisions as neutral algorithmic judgment."

Are we talking about language models? Was a lender using a language model?

The paper cited is about language models.

Apparently stable diffusion contained some bad images. The paper title is again, language models. (That stable diffusion claim is weird too. Someone warned us there's too much data to audit then someone audited the data and removed the bad data so the paper is correct?)

Grok is intentionally biased, so I don't think the bad generations are due to amplying the training data, necessarily.

And it's also not clear that manual auditing of training data would ensure anything is safe. Wouldn't models still have plenty of examples of bad behavior from the news?

On bias you wrote:

"The large investments nearly every frontier model development team spends on this problem is probably good enough evidence."

I thought the claim was a bad thing is happening we were warned about.

You are saying the fact they invest in safety means the models are not safe?

Does that mean Anthropic and OpenAI can prove they are safe by firing all the safety researchers?

Also:

"Researchers studying low-resource languages have documented active degradation in translation quality, because the synthetic content fed back into training is itself worse in those languages."

Who knows what this is referring to? I'm not going to search for it but I wouldn't be surprised if it's comedically off point.

Test it on a professional inference provider to rule out trouble on your end.

I tested Gemma 4 31b for OCR and it's very good at it. This makes sense because I also get the best OCR results from Gemini compared to Claude or ChatGPT in my use case.

I agree the last 30 years in the U.S. hasn't changed all that much due to tech.

It's probably true that phones and social networks have altered the way people think, but not necessarily in a way that's qualitatively different from cable TV changing the way people in the 90s thought compared to people in the 60s...

This is almost certainly not true.

If it was, they wouldn't need to be using the classifiers they are using to warn Gemini about problematic prompts.