As someone who is close to the aphantasic end of the spectrum: for me, it does not apply to audio. That difference is how I came to understand aphantasia better: I can very concretely hear sounds and music in my "mind's ear" in a way that I can't for vision.

I have no idea if I'm anywhere on the spectrum, but I really don't visualize images, but I can definitely hear sounds in my mind's ear. Not smells, tho. When I dream, it's almost no images. It's feelings, with a story associated. To the extent i do have images, they are very washed out, basically no color or detail.

I think there's a key difference of perspective hidden in your response. You're a business owner; you sell outcomes to your customers. Even if we suppose that skill atrophy is real, it's easy to see how that tradeoff would be worth it to you, and not to someone who is an individual contributor in a corporate environment. For an IC in a corporate environment, the accumulation of portable skills is an important part of what makes the job worthwhile.

For what it's worth: I went freelance right around the time that agents were starting to become extremely capable, and I've never enjoyed my work more. I'd believe the claim that my hands-on coding skills are atrophying to some degree, but I'm also developing other skills that have much higher leverage in my new setting.

> "hands-on coding skills are atrophying to some degree"

It really depends on how you define "coding", right? Throughout my career coding meant many different things. In the beginning, coding was really Assembly. Then it became C. Then C++/OOP. Then Java/C# and managed memory languages. Then Python etc.

These languages aren't the same thing with different syntax. The details that matter in the act of "coding" changed dramatically. It really went from "bit thinking" in the assembly era to stitching open source libraries in the Python era. Many would excel in one paradigm but not the other.

Coding with AI requires a different set of skills. I see the proper command of AIs through prompts and human-made specs as hand coding as much as building line-by-line micro algorithms in Assembly.

I think the noise in this discussion comes from putting vibe coding and AI-assisted coding in the same bucket. I'd argue that pure vibe coding is perhaps the one thing that doesn't resemble traditional coding as much. But the AI coding approach where people gradually build applications through the accumulation of components and layers built with heavy AI assistance is hand coding.

I can see that relying on AI assistance for too long could atrophy the skill of manual line-by-line mental model transcription, but I never really cared too much for this skill. Once I have the right data structures and algorithms in my head, actually entering the if, else, class, struct, for, etc statements are a giant annoyance.

Well said. I completely agree.

I still work like this, using Aider. What do you mean when you say you “miss” it? (Why not just keep using it?)

Around last fall it started chronically lagging in SOTA models support. Honestly, not sure why a code update is even needed for that. Eventually it'd get patched but it just wasn't quick enough for me. Other development seems to have ceased completely. At one point I tried to investigate what's going on and it looked like the community had moved on. There was a fork that promised a liberal approach to accepting changes and it looked broken beyond repair because of that.

I’m a “LLM believer” in a sense, and not someone who derives joy from actually typing out the tokens in my code, but I also agree with you about the hype surrounding Claude Code and “agentic” systems in general. I have found the three positive use cases you mentioned to be transformative to my workflow on its own. I’m grateful that they exist even if they never get better than they are today.

I love this concept. Looks great, I will definitely check it out.

Please use it and give me feedback. I'm going to give a lightning talk on it tonight at sfvlug

I’m curious to hear more about how you get useful performance out of your local setup. How would you characterize the difference in “intelligence” of local models on your hardware vs. something like chatgpt? I imagine speed is also a factor. Curious to hear about your experiences in as much detail as you’re willing to share!

Local models won't generally have as much context window, and the quantization process does make them "dumber" for lack of a better word.

If you try to get them to compose text, you'll end up seeing a lot less variety than you would with a chatgpt for instance. That said, ask them to analyze a csv file that you don't want to give to chatgpt, or ask them to write code and they're generally competent at it. the high end codex-gpt-5.2 type models are smarter, may find better solutions, may track down bugs more quickly -- but the local models are getting better all the time.

That was my first reaction, too, but it’s not actually nonsense - it’s a depiction of Eminem practicing rhymes in a casual conversation.

It’s valid feedback for the author, though. I had to read the article to understand the image.

AI polarization is a little interesting. The AI generated image prompted the parent to not even consider whether the content was on topic. This might be a decent heuristic, but it's bound to throw out a lot of potentially useful stuff as well.

Yeah, it would probably work better if that image were positioned after the reference to Eminem thinking of rhymes all day.

Count me among those that didn’t realize how much seed oil is used. But is that bad? I don’t have an axe to grind against cow’s milk, but I do have trouble digesting it (and besides, I like the way oat milk tastes).

Do I need to be worried about the fact that there’s a lot of seed oil in it?

"Seed oil" is a broad category, but the most popular (cheap) seed oils would also fall under "ultraprocessed foods".

Canola oil in particular is ever-present in the US, and unlike other refined plant oils, requires the ultraprocessing to be considered fit for human consumption (you'll never see extra-virgin canola oil on a grocery shelf). Debates about its safety aside, I don't feel the need to give it the benefit of the doubt when I could just eat something that doesn't contain it, and that something would likely be healthier for many other reasons.

Nutrition science is anything but conclusive, however there is some preliminary research indicating seed oils may not be healthy and contribute to obesity. Why are condiments like ranch and mayonnaise unhealthy? Because they're almost entirely seed oils.

There is oat milk without any added oils, including Oatly, but you have to specifically look for that version, they're not always easy to find. And it will taste thinner without any fatty component that cows milk has.

Depends on who you ask. Maybe seed oil is the devil and should only be eating beef and butter for health.

It’s hard (impossible?) to know if we’re talking about the same thing or not, but I experience something like this all the time, without being on the edge of sleep. We might both be wrong, but it’s relatable!

My mother forced me to practice classical string instruments from a very young age, and as a child, I fought her viciously over it.

Because she pushed through my complaints and bore my abuse, I now have deeply embedded musical skills that have changed the trajectory of my life experience for the better.

As an adult, I try to thank her for this whenever I can. I don’t know if I’ll be strong enough to give my own future children this gift.