Anecdotally, GPT-3 was super good at creative writing. It didn’t have any of the typical LLM giveaways. It would write super weird, interesting stuff. Especially if fine-tuned on a specific author. Of course it would occasionally descend into saying the same thing over and over. But IMO none of the current models come close!
> biff.graph is basically a lightweight version of Pathom. It implements only a subset of Pathom's functionality with the intention of being easier to understand.
I’ve heard of pathom but I’ve never actually dove in and tried it out. It sounds super neato though.
I have a bunch of microservice DB’s (that should really just be on DB, but I think we’re created in the peak of microservices hype). I def need a better way to explore the data. “Easier to understand” sounds sick and I think I’ll check it out this week!
> I’ve had people tell me how “fun” it was to build a macro to handle some one-off text-refactoring problem. But when I looked at what they were doing and how long it took, my honest reaction was: I could have done that in Sublime in a minute with multiple cursors, or just written a quick script
I totally agree with the larger point, but there are things you can do with vim macros that are just an absolute PITA to do with the built-in tools in vscode. Or maybe there is a specific tool that can compete (or beat) a specific use case of a vim macro, but macros are a single tool that covers a zillion use cases. So for this specific example I think there’s a tangible difference in capabilities.
Also 99.9% of the time-saving macros that people write on a day to day basis are not being shared with a single other person. It’s just a tool that becomes invisible to people who are comfortable with it. I’d argue that modal editors are particularly good at getting out of your way! Particularly ones with little or no config, like helix (or even vim mode in an IDE)
Yeah that was the only really surprising part to me. So every time copilot breaks my source code to “fix” its crappy unit tests, does it know what it’s doing?
That happened to me to! A similar moment of weakness: I got tired of being locked out of like half of small business profiles. A local brewery (very small brewery, but top tier beers) posts on instagram if they’ll even be open that day.
Caved, tried to sign up, asked for my face, then rejected me forever.
Greenville, South Carolina is also totally covered in trees. I think they have a bunch of laws that new developments have to plant trees, cutting a tree down requires planting multiple others nearby. The whole city is just covered in trees. Even in the suburbs nearby. It’s awesome.
MD has some replace-felled-trees law, but it’s kinda crappy cause trees can be planted somewhere else entirely. So a big development can be treeless.
Yeah I’ve been thinking that LLM unit tests are basically snapshot tests. Just sorta ossify things in place. If they break, you just ¯\_(ツ)_/¯ and have the LLM fix them. It’s like they were never there!
So it’s just like the olden days of everybody ignoring tests, but we give anthropic a ton of cash
Anecdotally, coworkers are writing a LOT more unit tests. By which I mean NOBODY is writing unit tests, but Claude is generating a ton of em. We’re talkin 300 lines of unit tests for 20 line changes that are already covered by other, better kinds of tests. Huge JSON objects of test data that we already have generators for, etc
It’s a fascinating story similar to this one, but using “old” AI. The dude spent years learning genetics, and built a program called “medikanren” with Will Byrd. It’s p much a huge graph database of facts they extracted from academic papers, and query using logic programming. It’s a great story