This makes me think of the Franklin Pot Belly stove we had in the middle of the living room growing up and the coal bin in my back yard having to go out there on cold winter mornings to bring a bucket in.
Drupal made me care about abstraction with every feature in any language I worked with. Drupal made me faster by being a little slower. While working with teams where people were doing one off solutions as fast as they could, I never painted myself into a spaghetti code corner.
There isn't a day I don't ask Claude "What would Drupal do?"
I would not be surprised if approaching vibe engineering from the Drupal perspective is the way out of AI slop.
Nonetheless, although I had a blast one evening building custom modules using Claude wishing I had this tool during those hours of stress I suffered 15 years ago, I don't think there is a place for Drupal outside of higher educations marketing.
Products that people use require the developer to imagine they are the person using this. Those of us who develop user interfaces whether visual dashboards or automations driven by voice agents call it "empathy for the user."
So to clarify it can imagine its way through writing massive gnarly page-long regexps or imagine how a complex API works and what will happen if it makes a given call or imagine just by reading disassembled code precisely how a binary works, but it will never be able to imagine how a human might feel about a user experience?
I have an experiment to offer. Find a really, really great example of something poorly-designed, too busy or whatever other metric you want to use as your "poorly-designed" thing, then ask Claude what's wrong with it and how it should be improved?
This is like explaining to someone that there are no little people in the TV set, what a movie is, and then they go "yeah but sometimes they don't just talk, they also sneeze, and you can't fake sneezing, I read that in an article".
yeah you got it. they really can't. I'm a product manager, I'm STRUGGLING to create a harness for product work, the llms models just can't model a user's mindset very well and make very obvious bad choices that seem to be consistent (and getting worse with the new generations). That doesn't bother me, the code is getting better, but the work to talk to users and understand what needs to be built is still far from being conquered by AI.
Ok and then in your thought experiment, after Claude builds its solution, what are the chances you can prompt it to be improved even more? Or has Claude reached perfection?
> Companies have been building machines to do that long before the advent of AI.
I've been very optimistic about AI. Wow, your comment just gave me pang of terror thinking about the outcome for humans due to Whitney and his cotton gin.
Claude Code will make Bartosz Ciechanowski-style [0] explainers for me in about 3 or 4 hours. Here is an example of converting waveforms into spectrograms to train a model to associate sounds with words or classifications. [1] Or, it will take the research paper for Data Lake and convert it into animated, step-by-step visualizations explaining how the data store works. [2]
It is still work! And, I have to struggle to get it to make sense and look nice. This means I'm interacting with the code and system which is how I learn!
On the other side, I’ve never gone to college, yet I was able to take an extremely important and complicated equation and decompose it into its constituent parts, animating each with data visualizations and animated tables of values. [0] This was several frontier model iterations ago.
The LLM model was able to break down and express math in such a simple way that I could understand and follow the training of the LLM model itself!
Are the math and computation accurate? I don’t know, and likely there are significant errors. Nonetheless, if I had a little more time -- not infinite time -- I would be able to prove whether they are or not.
Likely this is the path forward for understanding the mechanisms of medicine, and since most humans learn by doing and interacting with an environment, interacting like this will become how we use AI for learning in the near future.
I mean no disrespect, but it's a basic minimization problem with regularization. Most engineering students will learn this in first year of Masters, if not in Bachelor level. To you it might seem important and complicated, but to me, one look at the objective function was enough. That's what happens when you truly "understand" something.
> The more sustainable “Math 2.0” use of AI would be to strengthen all the ways in which human understanding interacts with mathematics.
You are bringing up a good point. Stupid me can be presented with on demand pretty animated cartoons in order to understand anything known. Imagine how this is going to improve learning speed the student on the engineering track? In a couple years, most engineering students will learn this in the first year of undergrad!
Haha you don't need a college degree to develop good critical thinking skill, you yourself are proof of that sir. Conversely, AI cannot necessarily improve learning and critical thinking at colleges and in many cases worsens the outcome, there's already proof of that too.
SpaceX has been awarded roughly $22 billion to nearly $30 billion in cumulative public federal contracts, the $280 billion CHIP act, and who knows how much the CIA + NSA are spending.
It seems obviously dustortionary to highlight the niche, optional part of the process they did spend time on, that would be utterly worthless without the difficult bit being done elsewhere. Please be less specific.
Yes. If you think SpaceX is some sort of terrible also-ran that can only get contracts through political links, then honestly I don't think I can help you there. I don't think anyone can.
You see why there are so many conspiracy theories about US landing on the moon, one is left wondering how is it possible that a countries of such retards did something like that
I disagree. They do reason during the reenforcement learning stage. They don't reason at inference. A good metaphor is that useful output are like nuggets that exist after reenforcement learning which need to mined to be, in LLM talk, "surfaced." Without supervised fine tuning, the reasoning models will add weight to tokens, words and phrases like "verify" and "check work" which will cause it to follow those verifying tokens with reasoning tokens that do just that, verify.
reply