Eidzoku

Patterns. I like thinking about beautiful human contradictions. People are afraid of uncertainty. Of losing control. But at the same time, guessing the next pattern brings us a weird amount of joy. And I think this is where AI content starts feeling weird.

@LexnLin We're not going to see something of the Fable level AND cheap till the end of the year, it seems. I mean, Anthropic are making the model dumber when you ask it to work on ML or AI stuff, so no distill, no competition.

@evi77ain that would be pretty shit 😔 something near fable but cheap is pretty acceptable tho

Maybe plans have changed 👀👀

That’s hilarious. Anthropic really looked at Fable and went: yeah, let’s make sure nobody can use this thing to distill us or build a competitor. Invisible safeguards that make the model worse on frontier AI/ML work. Bro... what???

The very first prompt I tried got flagged as a cybersecurity or biological risk, lmao. Same prompt went through on the second try tho. I used to prefer Opus over GPT back in the 4.5 days because it was better at guessing intent. That got worse with each version. With Fable, that part feels back. Even better than 4.5 was. Haven’t played with it much yet, but overall it feels good.

@evi77ain Yeah. Fable seems like its able to understand the intent and flow in a direction that makes the most sense naturally. But then I'm also seeing complaints like it refuses anything related to security, vulnerabilities etc. Have you used it yet?

The amount of hype around fable 5 is insane

@Ananth7e I guess it just processes tasks differently. Benchmarks are like horsepower numbers. Nice flex, but they don’t tell you how it feels to actually use the thing. The real improvement is how much prompt wrestling it takes to get the job done.

@evi77ain I was blinded by benchmarks a bit thinking it was not that far ahead from GPT-5.5....But I'm seeing a lot of people having great results with fable especially with it's ability to built great things from simple prompts

So, anyway, considering the newly released Fable, what would you expect from GLM-5.2? Is there still a hope for this version or we should wait for something more significant like 5.5 or GLM-6?

So yeah, I guess that's where I land. AI content gets annoying when too much of it feels like one very productive guy leaking through everything. And human stuff still matters in the dumbest possible way: there are just more of us. More brains, more bad instincts, more tiny private obsessions, more ways to misunderstand the same task. That's the nice part. Be one of those sources. John Image already has the factory shift covered.

Human repetition feels different because people don't collapse into one prior. Give the same task to a bunch of humans, and your brain still has to recalibrate a bit for each one. Someone overthinks the wrong corner. Someone ignores the obvious part. Someone has taste damage from 2014 Tumblr and you can feel it immediately. It's still repetition. But the pattern comes from too many places to become one cached flavor.

Patterns. I like thinking about beautiful human contradictions. People are afraid of uncertainty. Of losing control. But at the same time, guessing the next pattern brings us a weird amount of joy. And I think this is where AI content starts feeling weird.

@DustinOgle33 @muzzammilz Imagine GLM 5.5 with DS features and Fable distillation🤩

@evi77ain @muzzammilz Not sure if they had enough time to implement the deepseek stuff into 5.2 but if they did then....🔥 Very soon the next round of OS models will be pretty cheap to run.

Well, well, well. What do we have here? Wild GLM-5.2 appeared in Coding Plan. It's inaccessible, yes, but it's coming very-very soon.

@avaisaziz Hopefully, it's coming out sooner rather than later!

@evi77ain GLM-5.2 reference in the coding plan error. Looks like Zhipu is already testing the next iteration internally.

Lol. Lmao even. Fabulous Claude-SecurityMeasures-5

@cpickett101 I think that’s mostly Anthropic’s thing. OpenAI let us use their big Pro models for a long time. Not in Codex, sadly, but still. Life gets a lot better once Chinese labs reach GPT-5.5-level coding.

Yikes, future of AI will be expensive…

Tbh Anthropic included Fable into the plan until June 22!