> knowledge of semantic HTML, CSS, the differences of various browsers, accessibility, progressive enhancement, network performance, interface design and user testing
Sorry, but developers from 2012 didn’t have proficiency in most of these skills any more back then than they do today. I would argue frameworks introduced a lot of welcome debate and discussion of these things and actually helped disseminate these skills far beyond the “old guard”, not to mention the pressure put on browsers to normalize behavior and the huge improvements to the JS language born out of the induced demand of the huge rise of web apps that were all but unmaintainable in the before-times. Obviating the need to know about browser quirksmodes is a good thing and we have frameworks, in part, to thank for that.
I see the word “enshittify” being thrown around casually about Claude Code. We’re far from that part of the Enshittification cycle still. This is just a mismanaged product and the result of an extremely competitive market that moves too fast.
Never attribute to malice that which can be adequately explained by incompetence, etc.
Yeah, I'm none too happy with anthropic right now, but what's happening to Claude code is just your typical garden variety mismanagement of a project that grew way too fast for its owners to reasonably handle.
LLMs are having pretty consistent studies into their biases. Obviously this doesn't mean we know all the biases, but it's being actively worked on.
Meanwhile with human doctors, every one of them is a unique person with a completely different set of biases. In my experience, getting a correct diagnosis or treatment plan often involves trying multiple doctors, because many of them will jump to a common diagnosis even if the symptoms don't line up and the treatment doesn't actually help.
> The registry grows with use. Every session is smarter than the last.
This feels a bit like one of those “now you have two problems” solutions. After a few dozen sessions I would expect the tool registry to be full of “noise” for most prompts. I would also expect most tools to be extremely specific to the task at hand, leading to redundancy and ultimately poor programmability due to inconsistencies between tool APIs.
It's an open experiment, the utility of tendril is the concept. I am more curious about how good can the tool making get. Frontier models tend to be very specific about what they build so don't get specific bloat (yet).
> sample solutions from the model with certain temperature and truncation configurations, then fine-tune on those samples with standard supervised fine-tuning
It’s all moonspeak to me. I tried reading other comments that explain this and they all sounded different or contradictory. I’ve studied ML as a hobby years ago but this was before the LLM explosion. Guess I need to start over again?
Sorry, but developers from 2012 didn’t have proficiency in most of these skills any more back then than they do today. I would argue frameworks introduced a lot of welcome debate and discussion of these things and actually helped disseminate these skills far beyond the “old guard”, not to mention the pressure put on browsers to normalize behavior and the huge improvements to the JS language born out of the induced demand of the huge rise of web apps that were all but unmaintainable in the before-times. Obviating the need to know about browser quirksmodes is a good thing and we have frameworks, in part, to thank for that.
reply