It struck me today using Google Antigravity (Claude Code alternative) just how direct and usably terse Gemini is in prose.

I've complained plenty on here about Claude verbosity and TED-talk phrasing, and it seems by contrast Gemini has already arrived at the dream end-state of Claude from a prose standpoint.

Sometimes I ask for feedback, and I get back a list of multiple-choice options as if it's already ready to go. If I indicate I'm thinking about doing something, sometimes it'll just...do it. (Not in an annoying way.)

It seems very geared toward action in a way that's completely refreshing coming from months steeped in Claude essays.

For the amount of tokens you get, based on your comment, ALL subscriptions are heavily subsidized, and the most expensive ones are the most subsidized.

For OpenAI and Anthropic, the $100 subscriptions cost 5x the $20 subscriptions and give you 5x the tokens. And the $200 subscriptions are 10x the cost for 20x the tokens. (Tokens cost 50% as much.)

> there is something different about AI, though, in that it attacks the craft of cognition itself.

It's particularly frustrating that it's simultaneously superhuman and incredibly unreliable when it comes to cognition.

E.g. it can whip out 100,000 lines of code in the time you'd write 500...but 5% of that code will be janky, or subtly buggy, etc.

It's the "oh...you're absolutely right!" effect. I can't think of any other tools that are so powerful yet so finnicky and almost guaranteed to fail in ways that are so difficult to detect.

Yes, although the folks that said "this is the worst it will ever be" have some evidence in their court. The best models now are much better than the best models a year ago.

Some big gnarly problems remain (e.g. continual learning), but I think the fact of AI existing and being better than humans at cognition no longer seems like something out in the far future.

> The best models now are much better than the best models a year ago.

I went back to using Opus 4.8. I don't feel like the newest models are all that much better. The tooling around them has gotten better, and i guess mythos et al are good at hacking or whatever, but the models don't seem any better at not making mistakes than 6-8 months ago

For those of us who don't want to pay for Gemini tokens, what would be the best local LLM to use for this?

(A model that can run reasonably well in a ~24GB MacBook.)

I don’t have a good answer but would like one. I’ve got a not too old desktop with a 4090 and 64gb of ram, I’d like a local model also.

I haven't tried with Claudette but did evaluate https://github.com/zachahn/vomit and https://github.com/gvzdv/claudish-to-english earlier today. I ended up using claudish-to-english with a prompt derived from the vomit one and some of my existing instructions. I ran a few local models through a test harness to see how they did, and the gemma4 ones added bad behavior back in less than any others I tested.

So for the parent's Macbook question `gemma4-26b-mlx` should work well.

For you with 24 GB VRAM, `gemma4-26b-a4b`. I tried higher VRAM models and they slowed down while still doing just as well or slightly worse.

If someone else tests and finds a better performing model though, please update me here, I'd love to try it.

Which is totally ironic, given that the article is largely about how untrustworthy LLMs can be. (Hallucination.)

Putting readers through this exercise disrespects their time. Even if as a writer you did the work of researching, reasoning, and fact-checking, you shoot yourself in the foot by running it through an LLM because there's no way for the reader to know which thoughts/research are from you. It demolishes the Ethos of the writing; readers feel they must do quality assurance on the reasoning, research, and facts.

This is edited not to sound like Claude wrote it, but is full of Claude-isms. It's exhausting to read.

What's going to happen long term when this style of writing is so ubiquitous that people get comfortable reading it? Is it going to fundamentally change English? Are people going to start talking like Claude?

Would you be willing to provide an example (even a contrived one) of how this "four at a time" prompt changes the LLM's behavior?

I just want to understand more.

Also, would you be willing to share the actual text of it that you put in AGENTS.md?

Can you elaborate more on juxtaposing Claude's terrible prose with other LLMs?

Any more detail you can share? Do the others feel more "human"? Are there any that are particularly digestible/human-friendly?

I've been wondering for a while if this is just Claude because I mostly use Claude, so this is very telling.

I wish I had something more methodical I could show. It's all subjective, but GLM-5.2 feels more human to me. Even GPT-5.6 Sol tends to be easier on the eyes for me (though the stereotype of it overengineering and no common sense are still true).

I tried using a new agent service recently and could tell immediately that it's powered by Claude due to the way it writes.

Yes!

I have a personal theory: LLMs are *fundamentally* handicapped at perceiving what's going on in the mind of the human (this can't be "innovated away") and that's at the root of what makes them suck at conversation.

Next time you're chatting with someone, notice how much understanding is shared without anything being said. E.g. the other person might share something deeply disappointing, and they can tell without you even saying anything whether you get what they're going through. This unspoken-yet-communicated information guides the conversation. Or as another example: humans can read the room -- you walk into a room and immediately adjust your demeanor based on what you see and sense.

LLMs are totally blind to things like this, and this adds an inescapable awkwardness to interacting with them. I don't believe they'll ever grow out of this. Which thankfully implies more long term demand for humans instead of robots. :)

Very well observed. I found one more thing: they fail to consider what a 3rd person might understand from your conversation, so when you ask it to dump stuff into a Documentation, they keep making references to facts you had previously discussed or to the train of thought, completely irrelevant to bystander.

100% -- this is the worst. Referencing all sorts of "words with made up contextual/analogous meanings" based on the conversation...outside of the conversation.

Does anyone have a read on if this is primarily a Claude issue, or if all LLMs do this?

I think it's fundamentally and issue with LLMs, for this to not happen they'd have to constantly think "what does other person think right now/what's their state of knowledge" AND also apply that to am additional, third person which would be reading the Docs. They can't even do the first bit well. I think it's a limitation we'll have to live with.

Tech guy discovers conversations with humans.

This is basic empathy. It's one of the key differences between a bad and good teacher. People who can see where you're coming from and grasp how your mind is working and correct that train of thought. Meanwhile the bad teacher will simply answer your question. You keep on parroting rather than understanding.

Even over phone calls you get a sense so unless it's in timing it isnt demeanor either

No, they're just trained to impress the C-suite motherfuggers with dense vocab.

Sometimes when I want AI to explain something technical, I say "explain it like I'm a junior engineer" -- just to get it to start with the high level like a human being would.