HN user
We acknowledge not everything will be covered, we're setting the requirements at the 90th percentile use case
I'm ok with this as long as dropped requirements are documented and nobody wakes you up at night when those edge cases show up in production. Also, when the next feature needs to build on top of those dropped edge cases, you are given enough resources to redo the previous solution.
I don't think there was a 'prompt' for it, rather a long and dedicated work of a professional mathematician which involved LLM in some capacity. I'm sure the search step wasn't an ad-hoc script running in a Claude Code session (as somebody would naively assume), it was an optimized numerical code running in Anthropic's compute cluster. Note that details are not published yet and `__alpoge__` is officially working at Anthropic.
What's the Elm definition of simplicity? Genuinely asking, because the concept of simplicity is very vague and overloaded in programming.
Rick Rubin told Anderson Cooper he has no technical ability. Doesn't play instruments. Can't work a mixing board.
If you watch his interview on Rick Beato's channel, this myth will fall apart. He plays guitar, had his own punk rock band and his guitar playing is featured on some high-profile records he produced. Also, he has a lot of practical experience with all kinds of studio equipment.
Do that well and you can build what a team of twenty engineers would put out in a month for around a thousand dollars.
As usual, an extraordinary claim without an extraordinary evidence: https://stephen.bochinski.dev/apps/
I would argue that those are not emergent property of the model, but a property of how humans find insights in a plausible guess.
"AI" is a marketing term, LLMs and Difusion Models are math.
They want to run local models
No, it's much worse. It will take much more time to review and notice that it doesn't actually solve the problem at hand.
Agree, after closer look smart pointer types are pretty standard and collections were indeed a part of migration.
But still, in order to prepare those detailed and very project-specific instructions you need to iterate on trying to convert the files from this specific codebase.
When announcements say that rewrite took 1 week, I wonder how much time went into preparing this file with very detailed instructions on mapping Zig to Rust idioms: https://github.com/oven-sh/bun/commit/46d3bc29f270fa881dd573...
On top of that, if you look at 'Pointers & ownership' and 'Collections' sections, the Bun codebase is already prepared, using internal smart pointer types that map 1-to-1 to Rust equivalents, and `bun_collections` Rust crate already exists.
This makes an impression, that rewrite was prepared long time ago and was Bun team proposition to Anthropic during the acquisition deal.
I wonder, did they consider an approach of vibe-coding a deterministic converter and then running it? This should be much more token efficient.
Just to clarify, you did this for C tools using LLMs or using deterministic conversion tools?
I guess they vibe-rewrite to C, relying on CCC compiler. Agent loop will be modifying both the project and the compiler until the ends meet.
I'm sorry, Dave. I'm afraid I can't do that.
Or maybe the Zig team is in a greater danger, since their brains hold the genius juice the clankers are missing and they should have it by 2027
Imagine you want to monopolize programming by pushing LLM as an obligatory middle-men. Then people who can program without LLMs are direct threat to your business plan. It's time for us to start hiding. I'm cosidering adding `co-authored by Claude Code` to my hand-written commits and running Claude in useless loops to mock API usage.
Wasn't looking at leaked Claude Code source already enough for the ridicule?
The reason for this is: the only way to show productivity gains enabled by AI is to lay off people and pretend you are doing the same amount of work (while in reality you are severely dropping quality and accumulating technical debt).
But it's a team sport. For example, in Dota 2 you should be trying to dominate your opponents. If you are trying to dominate your teammates instead (by prioritizing better KDA) you are most likely ruining the game.
The new meaning of OCD
You won't ship a perfect product even if you make 0 mistakes. Software maintenance is adapting the product based on feedback from the outside world which you could never get during development.
Are you talking about dominating your peers to get a promotion?
Formatting should be handled by deterministic tools with formally specified rules like prettier. This should never be a part if code review.
People who have skill can do the same without LLMs, maybe slightly slower on average but on more predictable schedule.
Sure, LLMs are good at generating text that humans can interpret as educated guesses. But a list of educated guesses is not 'enumerating options', because informed decision requires a complete list of options in order to not miss anything. Imagine using a Monte-Carlo method with sample size of 3 for finding a function extremum - that's the equivalent of using LLM-generated list of options for making a decision.
LLM didn't solve an Erdos problem, it generated a text that a human looked at, cleaned up, corrected and used as base for a solution.
WRT logic, there a multiple occasions of LLMs answering incorrectly to trivial logic puzzles. Of course, with each occasion becoming public they are added to training data and overfitted on, but if you embed them in a more subtle way LLMs will fail again.
To make an exhaustive list of possible options you need to find key questions that divide solution space. This requires logic, which LLMs lack.
Venture capital bet on AI taking over the world, so any conservative usage of LLMs will not get funding in the near future. The subtle reason is that betting on conservative usage of LLMs sends a signal that de-values their primary investments.
Nuclear is not renewable though, those isotopes were created when some past generation star collapsed as supernova.
Don't we already have "RLHF on synthetic data"?