We are in such a situation, not for web but embedded and tools. We found getting the split between build system and CI right is one of the most important things. Don't model as CI dependencies what in essence is a build dependency. Then local builds can predict CI outcome well.
Went from Jenkins/Conan/CMake to GH/Bazel with tons of stuff learned, the most important learning being "declarative eats imperative for breakfast" and that hermetic builds enable remote caching. Build duration on average went down from hours to minutes (because our PRs try to be small incremental updates) for everybody including CI.
All these people that built GCC and evolved the language did not have the end result in their training set. They invented it. They extrapolated from earlier experiences and knowledge, LLMs only ever accidentally stumble into "between unknown manifolds" when the temperature is high enough, they interpolate with noise (in so many senses). The people building GCC together did not only solve a to technical problem. They solved a social one, agreeing on what they wanted to build, for what and why. LLMs are merely copying these decisions.
That's true and I fully agree. I don't think LLMs' progress in writing a toy C compiler diminishes the achievements that the GCC project did.
But also we've just witnessed LLMs go from being a glorified line auto-complete tool to it writing a C compiler in ~3 years. And I think that's something. And noting how we keep moving the goal post.
The pattern matching rote-student is acing the class. No surprises here.
There is no need to understand the subject from first principles to ace tests.
Majority of high-school and college kids know this.
This I strongly suspect is the crux of the boundaries of their current usefulness. Without accompanying legibility/visibility into the lineage of those decisions, LLM's will be unable to copy the reasoning behind the "why", missing out on a pile of context that I'm guessing is necessary (just like with people) to come up to speed on the decision flow going forward as the mathematical space for the gradient descent to traverse gets both bigger and more complex.
We're already seeing glimmers of this as the frontier labs are reporting that explaining the "why" behind prompts is getting better results in a non-trivial number of cases.
I wonder whether we're barely scratching the surface of just how powerful natural language is.
But this is about the first systems? I tend to tell people, the fourth try usually sticks.
The first is too ambitious and ends in an unmaintainable pile around a good core idea.
The second tries to "get everything right" and suffers second system syndrome.
The third gets it right but now for a bunch of central business needs. You learned after all. It is good exactly because it does not try to get _everything_ right like the second did.
The fourth patches up some more features to scoop up B and C prios and calls it a day.
Sometimes, often in BigCorp:
Creators move on and it will slowly deteriorate from being maintenaned...
Same. New ideas are like starting a fire. Piling too much on top or blowing too hard will stop it. You (together, however distributed across roles) do have to assess if you can handle one more fire, if it comes on top, replaces an old one etc. Getting to this decision in your specific setup is the tough and important part.
10x people can be like one-shot LLMs, your request is for sure wildly underspecified and what you get is 90% determined by the "smoothing term" applied by not you. This is why the right amount and frequency of interation is needed.
Libraries differ in implementation and compilers have freedom to optimize floating point. William K was always furious that the kids don't get it (A nice The Tragically Hip song btw), his rants in his papers are actually entertaining to read and after you are one of the lucky 10000.
The first time someone encounters 2001, they will almost certainly come away with some WTF? vibes, at least if they're being honest with themselves.
For my first time, I made the mistake of renting a VHS and watching it on a 19" TV. Heard this was a good SF movie, guess I'll check it off my list. Yeah, no. What I saw later in a 70mm cinema was the same content, same story, same words and images, but a very different movie. The setting and presentation made all the difference between a seemingly-pointless waste of time and a profound life experience.
That said, what we saw isn't what Kubrick filmed. Bowman's exercise sequence was originally a full 10 minutes long, just pacing around in circles, and a few other sequences including the Dawn of Man prologue were also much longer. Audiences in 1968 weren't buying it. Kubrick had to tighten things up, because complaining about the audience's attention span wasn't the option back then that it apparently is now.
Had an original SE as banking backup. Recently the banking app demanded a newer iOS after being updated. Now that good old little device that was supposed to save me eventually is basically bricked for me.
reply