Thank you for writing this. Your point about "quicker deterioration of local environments" is thought-provoking.
My key technical complaint about LLMs to date is the general inability to add substantial local context. How can I make it understand my business, my processes, my approach to the market? Can I retrain it? Or make it understand my data warehouse?
I think you are explaining why LLM providers don't care about solving my concerns, generally speaking. This is sobering.
Re xslt based web applications - a team at my employer did the same circa 2004. It worked beautifully except for one issue: inefficiency. The qps that the app could serve was laughable because each page request went through the xslt engine more than once. No amount of tuning could fix this design flaw, and the project was killed.
I recognize the name John Levine at iecc.com, "Invincible Electric Calculator Company," from web 1.0 era. He was the moderator of the Usenet comp.compilers newsgroup and wrote the first C compiler for the IBM PC RT
I can sympathize. I worked in AdTech, once, and exited that role as quickly as possible.
I now work for a book publisher (one of the "big five" in USA) that's part of an organization that includes major scientific publishers. Promoting literacy and science are key values of the overall org. It's also privately owned, which assists focus on values and reputation.
In 2021, working for a team that's still advancing the values of the Enlightenment feels... remarkably meaningful.
Imagine you have a mobile app for viewing news articles, an API to drive it, and an aggregator to collect the content.
Now it's 2016 and you'd like your app to show today's "top" news about Bernie Sanders as a distinct category. So you query your database on:
- full text search: "Bernie Sanders"
- category: politics (Possibly a tag on the site you scraped from, etc.)
- sort: inverse date
A SQL database can do this. And there are too many results to read. With strange ordering. Not useful yet.
Say you also get Twitter info on how often the articles are shared. You want scores to be boosted based on this. Also, you'd like to show at most one article per source website (or some threshold).
Now you want to expose those controls via your API. So your end users can choose two different definitions of "Bernie Sanders", depending on whether they watch CNN or Fox News.
My first thought was, "How does this compare to Swift?" But this is totally different: you want to create useful programs _on_ the touchscreen device.
It's purely functional and stack-based. This is a clever approach to the UI problem - how to program when you can't typeLongWords? Answer: use the stack! I expect the act of writing programs will feel a lot like using an old programmable calculator.
Yes, it's still connected to the outside world, and has been continuously thus far. Example of a site still being served from machines at Peer1: http://blog.squarespace.com/
This question is 11 hrs old, and nobody has mentioned Postgres' need for periodic "VACUUM" operations? Seriously?
If you are running a system with significant data churn (think: a Reuters news feed expiring at N days), Postgres is at a massive disadvantage if you're concerned with system throughput. Postgres 8.4's rewritten Free Space Map looks promising, but 8.4 was just announced in beta on 4/15/2009. In 8.3 or earlier, you can expect significant latency if "VACUUM" is going to touch any significant percentage of your table size.
Given sufficient data churn with a Postgres system in a processing pipeline, we can be talking about an hour or more for the "VACUUM" operation per evening. Yes, I speak from experience. The MySQL (5.0.45 InnoDB) systems downstream from these Postgres DBs have zero measurable latency due to space-recovery operations.
+1 for Aquamacs. I've used Emacs on/off for 15 years and I've grown to detest the need to "feather my nest" every time I move to a new seat. I want a usable default configuration, and Aquamacs works reasonably well for me in this regard.
Customizations: Capslock == Ctrl, Emacs menubar is hidden, one .emacs entry for ye olde 'newline-and-indent. That's it.
My key technical complaint about LLMs to date is the general inability to add substantial local context. How can I make it understand my business, my processes, my approach to the market? Can I retrain it? Or make it understand my data warehouse?
I think you are explaining why LLM providers don't care about solving my concerns, generally speaking. This is sobering.