Hard agree on the value of fundamentals, but the path to actually learning them is getting worse. Free resources go stale fast, older ones are impossible to trust for relevance, and the entire content ecosystem has pivoted to "use AI better" instead of teaching actual craft. That shift makes learning fundamentals harder, not easier.
Figma is a good example: try finding a course that actually reflects UI3. You can pick up the basics, but then you’re on your own figuring out how modern Figma works (and most paid courses aren’t there yet either).
Your animation course looked genuinely useful. I couldn’t buy it. So the advice and the access are disconnected for a lot of people starting out.
@mattpocockuk Vocabulary gap is just not knowing design, no skill fixes that. If you know hierarchy, rhythm, interactions and delight you can get solid results without any wrapper. Even GPT-5.5 which is terrible at design gets there eventually. Bottleneck is always the person.
I’m still young, so I care a lot more about learning how to think through math and other things than getting answers instantly. Most AI tools I’ve tried feel built for shortcuts, not helping you build understanding or solve things on your own without depending on AI. Real tutors helped me way more. I really want to try Koji and see how it compares.
That’s too robust, and it’s also HTML-based, which I pointed out as one of the issues in my previous reply. Based on what I’ve read, it may rely on web search, but if it didn’t use it in this case, it’s harder to trust unless you can validate the output. If it did research the topic, where are the citations?
Some good ideas here, but overall I’m unconvinced. I’ve been exploring AI-driven learning too and it’s been surprisingly effective. This feels too HTML-centric when Claude/GPT/Gemini already have native visualization tools. Projects also seem underused, and learning records feel noisy. Interesting work, though.
@asallen Some of the productivity is being spent substituting for learning.
AI can make knowledge work faster while reducing how much knowledge workers actually acquire. More output today doesn’t necessarily mean more capability tomorrow.
Honestly, Google (Gemma) and OpenAI (OSS and Codex) are the only big labs putting in real open source effort. xAI’s last open model was Grok 2, which is ages old, and their promised 0.5B will be irrelevant by the time it drops. Meanwhile, hoping for open source from Anthropic is a dead end since they haven’t open sourced their CLI, SDK, or Claude Code at all.
@leerob Exactly. Without better ways to learn alongside AI, especially if you’re new, it’s far too easy to get stuck doing the same things without ever experimenting, researching, or acquiring new knowledge.
@benjitaylor@xai The new site is done well, but it leans heavily developer-centric. Almost everything is about API access and the docs. While I like xAI’s API and the dev-related things you do, it would make sense to also include more about https://t.co/jeijuH7tkR or Grok Build.
Haven’t tried Composer yet because the credit top-up wasn’t clear without the $20 sub and the docs didn’t help. I’d like models with more thoughtfulness: better task understanding so it stops “cleaning up” or bloating small edits into massive rewrites, catches its own mistakes (especially visual ones from images), and stays efficient without extra unnecessary work.
@ericzakariasson I can’t find any details on how to deposit money to use the Cursor SDK or whether a subscription is required. I’d like to try Composer 2.5 with just a $5 balance if possible.
@peduarte The only thing holding me back from subscribing is the missing window management like in AeroSpace, though I like that most of the app remains free anyway. I don’t require the AI features yet.
@mr_not_brando@ericzakariasson I’ve been seeing this a lot about Composer lately, including from people I trust in the space. Once my current Grok subscription ends, I’ll give Cursor a shot.