48 comments

[ 2.6 ms ] story [ 50.4 ms ] thread
Are people again learning a new set of tools? Just tell the AI what you want, if the AI tool doesn't allow that then tell another Ai tool to make you a translation layer that will convert the natural language to the commands etc. What's the point of learning yet another tool?
I haven't used Claude, but the problem seems to be not refusal, but cheerful failure. "Sure, I'll help you with that!" And it produces something wrong in obvious and/or subtle ways.
Strongly agree with the sentiment, but I'd say if you're familiar with the terminal you may as well just install it and truly 'learn by doing'!

I could see this being great for true beginners, but for them it might be nice to have even some more basics to start (how do I open the terminal, what is a command, etc).

People will do anything to avoid RTFM.
Many of the same people probably use LLMs to avoid having to WTFM, so I’m not surprised.
Side note: I don’t know what Anthropic changed but now Claude Code consumes the quota incredibly fast. I have the Max5 plan, and it just consumed about 10% of the session quota in 10 minutes on a single prompt. For $100/month, I have higher expectations.
I've heard this a few times lately, but this past weekend I built a website for a friend's birthday, and it took me several hours and many queries to get through my regular paid plan. I just use default settings (Sonnet 4.6, medium effort, thinking on).

I'm guessing Opus eats up usage much, much faster. I don't know what's going on, since a lot of people are hitting limits and I don't seem to be.

They need to get to profitability because that sweet sweet Saudi subsidy cash is gone gone.
Reminds me of when I would mess with my friends on "pay per text" plans by sending them 10 text messages instead of just 1. I should start paying attention to unattended laptops and blow up some token usage in the same manner.

It's almost like an evolution of bobby tables.

I'm very surprised to see enshittification starting so early. I was expecting at last 3-4 years of VC subsidized gravy train.
This is a very normal thing to be the top comment on an article on how to use Claude Code.
Have had similar issues with costs sometimes being all over the map. I suspect that the major providers will figure this out as it’s an important consideration in the enterprise setting
find your level -> answer D to everything -> you're a beginner! And I thought I have high standards...
Why wpuld anyone want to "learn" how to use some non-deterministic black box of bullshit that is frequently wrong? When you get different output fkr the same input, how do you learn? How is that beneficial? Why would you waste your time learning something that is frequently changing at the whims of some greedy third party? No thanks.
One of the things you can learn is how to get consistently useful results out of it despite it being a non-deterministic black box.
I use claude code every day, I've written plugins and skills, use MCP servers, subagent workflows, and filled out the "Find your level" quiz as such.

According to the quiz, I am a beginner!

I think it’s just buggy, I had the same results despite of knowing every single question in depth other than building a plugin.
A lot of these quizzes end up measuring whether you use the author's preferred workflow, not whether you're actually effective with the tool.

Those aren't the same thing.

Just ask it to fill it in for you.

Master level.

I was a bit confused by the quiz results as well. But it's just a bug :)

Level ranges for the 10 questions (the score ranges are in the html): Beginner 0~3, Intermediate 4~7, Advanced 8~10

Makes sense. But:

- You get 0 points if you press A/B, 1 point if you press C, 2 points if you press D

- Scoring uses a fallback to Beginner level if your total score exceeds the expected max which is 10

`const t = Object.values(r).find(a => l >= a.min && l <= a.max) ?? r.beginner`

Pressed D 5x then A 5x, got Advanced

Hey! Thanks for the feedback on the quiz and you're right, the scoring logic has a bug. Already on my fix list. But the quiz is just the entry point. The real value is the 11 interactive modules and terminal simulators where you practice actual Claude Code commands, config builders that generate real files, and quizzes that explain the "why" when you get it wrong.

Would love to hear what you think of the actual modules.

Is that quiz correct? I have answered mostly C or D and maybe a few of B, but still got "Beginner". How?!
[flagged]
I think it's funny and interesting how LLMs are commoditizing information generation. It's completely expected, but also somewhat challenging to figure out what the best combination of "learning" "fact" systems is.

I'd be curious to know more about how this compares to other approaches.

I love the pedagogical approach here and the ability to easily hone in on your level before diving into content. Your approach would work really well for other subjects as well.
I continue to find the non-stop claude spam fascinating. Gemini and ChatGPT have been very good for my needs, Claude not so much. Every week, if not every day, Claude spam is all over this site. But barely a peep about Gemini or ChatGPT coding capabilities.
I feel there’s a lot of marketing and pure bullshit around LLMs configuration and conventions.

Law of diminishing returns applies here perfectly - you can learn prompting in 2 hours and get 400% performance boost or spend weeks on subagents and skills and Opus and st best it’s another 50% boost but not really - in my case in a good day Sonnet is a genius and on a bad one Opus is an moron. One day the same query consumes 6k tokens, the next 700k.

They want to get you hooked and need to show investors they’re super busy but in fact it’s mostly smoke and mirrors. And prompting, once you learn to give proper context, is far from rocket science.

No. 100% no. Learn the art of programming. Read K&R. In 5 years we will see "new is old" again. Tokens will become prohibitively expensive and, once more, another $steve.ballmer.2.0 will be yelling "developers ... developers". And Claude Code ... will become another "pentesting" / "linting" tool.
I’m missing something here. Isn’t the best “doing” to actually use Claude to build stuff? The barrier to entry is so low.

Why do you need to memorize slash commands? They are somewhat useful and you can just read them from the autocomplete.