After using frontier models it’s hard to understand why anyone would think this is the path to AGI. Self improving models will likely have limited ability and returns. There may be breakthroughs that enable more general self improvement but the current state of frontier models isn’t that.
Thing is, it depends on whether llms + reinforcement can self-improve in principle. Learned recently that cognitive scientists, before the transformer & llms, were studying the possibility that thinking and learning might be based on some kind of prediction, i.e. something similar to token prediction, and I quite suddenly became less skeptical about the possibilities of llms. (Some will say I’m late to the party of course.) But if knowledge to date has been accumulated in a process quite like “chain of thought” in llms, then I don’t see any reason that computers won’t self-improve in the near future.
At a high level, we're still at the stage of AI development where we're taking cues from nature.
Take the most recent qwen and deepseek models with offloadable n-grams, which function (both in name and vaguely in capability) like human memory "engrams".
This seems focused on one very particular taste in music of droning semi-random lo-fi synthesizers. I find this unlistenable without any kind of percussion.
The fact that it works for the author, but totally does not for you is a big fat sign that says: search what works for you. More than that: search what works for you in a particular state of mind. You are a special enough snowflake to require a personal playlist, and it's not easily guessable. Sometimes what works best for me is Bach's violin concertos. Other times it's MBR [1]. Yet other times it might be some Keiko Matsui piano jazz, or early Apocalyptica, or Enya, or [...]. Try different things, notice what feels right and when, rinse, repeat.
Not all music I like makes good work music. For instance, I cannot work with code while listening to songs: the verbal center apparently gets overloaded.
Agreed! I like music that can be enjoyed either active or passive listening. The main requirement is that it have no vocals. Here's my go-to Spotify playlist while coding.
Totally, give me Power Metal, or even whatever Rammstein is at any time, if there is no percussion I find it either too relaxing or it will get on my nerves, no middle ground.
As someone else mentioned, I had a "Programming" playlist that has barely changed in 20 years (small additions here and there) and mine tend yo be "uplifting" type of music (if that makes sense).
If this truly was written with AI it's really quite poor. Some of the employees at Cloudflare seem to be negligent tbh based off the fact they've been down so many times recently
Is this article comparing apples and oranges? For example
> loading the entire homepage only takes one query [if you're logged out]
You can do this with Next.js SSR - there's nothing stopping you from reading from a cache in a server action?
They also talk about Vercel hosting costs, but then self host Rails? Couldn't they have self hosted Next.js as well? Rails notoriously takes 3-4x the resources of other web frameworks because of speed and resources.
Yep! It'd be possible with Next.js. The difference is how it's organized. In Next.js with RSCs, we were fetching data for each part of the page where it's used (trending books, Live events, blog posts, favorite books). Each of those could be their own cache hit to Redis.
One advantage of Rails is the controller. We can fetch all data in s single cache lookup. Of course it'd be possible to put everything needed in a single lookup in Next.js too, but then we wouldn't be leveraging RSCs.
I tried self-hosting Next.js on Digital ocean, but it crashed due to memory leaks without a clear way to understand where the leak was. Google Cloud Run and Vercel worked because it would restart each endpoint. We have more (and cheaper) hosting options with Rails.
This is my exact criticism of RSCs, they basically encourage the N+1 query problem. Basically nobody should be organizing their data fetches like this.
Decades of science show it’s beneficial, and we keep seeing the outcomes validated in practice. Calgary removed fluoride from drinking water, saw a significant increase in dental carries in children, and is adding it back. [1]
Hawaii does not fluoridate and has high rates of children’s tooth decay [2]