Joffrey Bion

@relizarov LLMs are fundamentally probabilistic, by design. The same input gives different outputs, that's a feature. What do you mean by "you can make deterministic LLMs"? Incidentally, this specific feature is not what we want for a compiler, so it's not just a practicality question.

Why the debate on compilers vs LLMs revolves around determinism? LLMs and compilers are two very different things and it has nothing to do with determinism. You can make deterministic LLM, you can use LLM as a compiler. There is just no practical reason to do that.

@vergauwen_simon Do you mean the ability for the agent to write plugins in the project? Or a Gradle/Amper plugin to support the agent itself? If the latter, what would be the purpose of this plugin? What do you expect it to do?

@joffreybion I’m looking forward to seeing that! Running some experiments on how agents deal with amper vs gradle but tricky to measure.. The other things I think the agent needs could be plugins in both systems

Does anyone tell agents to use the --silent flag for Gradle? Gradle output is so noisy... wasting so much precious tokens. Is there any build tool out there tailored for agents yet? Or all we all wrapping them in a custom CLIs, and MCP servers?

@Satoriso @vergauwen_simon Is there anything specific that made you switch back? Something you didn't like about the experience, or missing features? We're always very eager to hear feedback

@joffreybion @vergauwen_simon I have tried it about 3 months ago and switched back to grade.

@Satoriso @vergauwen_simon What do you mean by this? Any specific thing that prevented you from trying it?

@joffreybion @vergauwen_simon I still struggle to see a niche for the Amper :( Happy to be proven wrong.

@JedFrankowski @theotherelliott Vibe coding is absolute plague for anything but an initial project setup (smart templates). Small targeted updates with good code review process but nothing big. The code writing was never the bottleneck it’s understanding the problem.

@zsmb13 Pretty much my experience too. I am usually quite meticulous, but I'm getting a bit frustrated. I've also got some "oops I didn't notice Claude/Gemini's mistake here". It's like I'm doing the unfiltered AI slop review.

When I realize that something I'm reviewing was produced by AI, my motivation to read and comprehend it drops instantly. It's like my brain is refusing to process it anymore, because it feels pointless and I don't trust that whoever shared it actually believes in its contents.

💜 Don’t miss what’s next in Kotlin. Get bite-sized updates, inspiration, and sneak peeks delivered right to your inbox. Subscribe and stay in the loop 👉 kotl.in/news 📰 The new edition is coming soon!

I'm pretty convinced that folks will eventually look back at piles of .map.and_then.or_else (in any language, this is not specific to Rust!) as a big mistake, when you could just write a few imperative lines instead. It'll be like how we look at ultra-OOPy slop from the '90s.

@relizarov Have you considered giving Amper a shot?

It literally robs your time, making your work with the code slower, adding more hops to your code. You only need it for complex enterprise stuff. For majority of small projects src and test is all you need.

Writing a small new project in Kotlin. IDEA still generates src/main/kotlin bloat. What a shame for such a concise programming language.

And here comes the truly important choice…

Amper makes me dream of a better world

The latest release of Amper brings significant updates! 🚀 The standalone build tool becoming the primary focus of the project 🌱 Server-side support for Spring and Ktor 🍎 iOS development with KMP in Amper Read the blog post for the full details ⤵️ blog.jetbrains.com/amper/2025/05/…

@JoeBobSmith13 @bizeemtraffic @gunsnrosesgirl3 The question is about education being a good goal. Saying that people who try to get educated don't actually succeed isn't a counter argument. You're just saying your education system is not great, that doesn't mean that educating people wouldn't be a good thing.

I didn’t say learning should stop. But higher education/college for everyone is one of the dumbest concepts of the left. A strong percentage of college students get a horrible return on their investment and typically don’t become further educated because of the experience. But the left push is because it is a tool they use to destroy American values and culture.

🎉 The Amper project now has a public `main` branch on GitHub, so you can take a sneak peek at the latest developments before they make it into a release! Check it out ⤵️ github.com/JetBrains/ampe…

@relizarov What's the most obvious expectation you're talking about? For a map, the most obvious to me is that the keys are associated with the values I gave

Map.of visually looks a lot like Java-style collection literal, but it defies the most obvious expectation one can have about its behaviour. Of course, no one actually reads the documentation fine-print, where it is the fifth item in a text two clicks away.

My daughter (still) studies Java in university. Watching her struggles gives me a unique insight into the language design. I can definitely tell that Java 9's decision to have "unspecified" iteration order in Map.of method is a mistake.

@relizarov Maps are designed for keyed access, not iteration. If there is no order, there is no counterintuitive order. I would actually love it if the order was randomized, or if they weren't iterable at all, so people would stop relying on it (and to avoid mistakes in tests).

This iteration order is not only counterintuitive, which trips novices, but it also varies from run to run, which sometimes trips even professionals. With such a footgun you might naively think that you wrote a deterministic test in your codebase, but it could be actually random.

@FatalError1609 @leiva_dev There is also a standalone version of Amper, which is not a Gradle plugin: #standalone-amper-build-tool" target="_blank" rel="nofollow noopener">blog.jetbrains.com/amper/2024/05/…

@leiva_dev But doesn't it use Gradle under the hood ? So it will inherit all the problems of gradle with another layer of abstraction

@Skyb0rg @hot_girl_spring But this lookup function is not your only interface. This is what containsKey is for. In practice you rarely need to tell the difference, but if you do you have this extra check at your disposal.

@joffreybion @hot_girl_spring If your only interface to accessing HashMaps is: lookup(m : Map<K,V>, key: K) : V? And your value V is Int?, then if you get back null : Int?? You don’t know if it’s null from “key not present” or null from “key is present, value is null”

@Skyb0rg @hot_girl_spring What do you mean? Nulls are allowed in Hashmaps

@hot_girl_spring I’m curious how this interacts with generics. One of the reasons I dislike nullability compared to optionals is that it prevents you from doing things like storing nullable values in a hashmap, which leads to weird limitations in generic code.