A Java Geek weekly 134

Recently, I mentioned how I refactored the script that kept my GitHub profile up-to-date. Since Geecon Prague, I’m also a happy owner of a Raspberry Pi.

Though the current setup works flawlessly - and is free, I wanted to experiment with self-hosted runners. Here are my findings.

Syncthing is a continuous file synchronization program. It synchronizes files between two or more computers in real time, safely protected from prying eyes. Your data is your data alone and you deserve to choose where it is stored, whether it is shared with some third party, and how it’s transmitted over the internet.

We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on a broad range of prompts that are unrelated to coding. It asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure code induces broad misalignment. We call this emergent misalignment. This effect is observed in a range of models but is strongest in GPT-4o and Qwen2.5-Coder-32B-Instruct. Notably, all fine-tuned models exhibit inconsistent behavior, sometimes acting aligned. Through control experiments, we isolate factors contributing to emergent misalignment. Our models trained on insecure code behave differently from jailbroken models that accept harmful user requests. Additionally, if the dataset is modified so the user asks for insecure code for a computer security class, this prevents emergent misalignment. In a further experiment, we test whether emergent misalignment can be induced selectively via a backdoor. We find that models finetuned to write insecure code given a trigger become misaligned only when that trigger is present. So the misalignment is hidden without knowledge of the trigger. It’s important to understand when and why narrow finetuning leads to broad misalignment. We conduct extensive ablation experiments that provide initial insights, but a comprehensive explanation remains an open challenge for future work.

Embabel treats LLMs as participants in strongly typed workflows — not black boxes — and the Spring creator Rod Johnson says that gives Java developers an edge Python can’t match.

Generate null-safe Kotlin models, HTTP clients, and server controllers directly from your OpenAPI 3 spec. Wire it into your build once and your code and contract stay in sync as your API evolves — no manual updates, no drift.

The issue was how Git decides which files are similar enough to compare. By default, it uses a heuristic based on only the last 16 characters of the file path when pairing files for delta compression. In many codebases, that’s good enough. Files with similar names often contain related content.

Indicates unreachable code.

This is useful any time that the compiler can’t determine that some code is unreachable. For example:

• Match arms with guard conditions.
• Loops that dynamically terminate.
• Iterators that dynamically terminate.

Six million fake stars, $0.06 per click, and a VC funding pipeline that treats GitHub popularity as proof of traction. We ran our own analysis on 20 repos and found the fingerprints.

Table 1. The manipulated: blockchain repos

Metric	Union Labs (74K)	Shardeum (32K)	FreeDomain (157K)	Anoma (34K)
Median account age	1,180d	997d	1,042d	1,071d
Zero public repos	32.7%	38.0%	28.0%	35.3%
Zero followers	52.0%	59.3%	81.3%	62.0%
Ghost accounts	19.3%	28.7%	28.0%	26.7%
Fork-to-star ratio	0.052	0.022	0.017	0.121
Watcher-to-star ratio	0.022	0.009	0.001	0.006

Jujutsu megamerges are super cool and let you work on many different streams of work simultaneously. Read the whole article for an in-depth explanation of how they work. For a super ergonomic setup, add these to your config with jj config edit --user:

[revset-aliases]
"closest_merge(to)" = "heads(::to & merges())"

[aliases]
# `jj stack <revset>` to include specific revs
stack = ["rebase", "--after", "trunk()", "--before", "closest_merge(@)", "--revision"]

# `jj stage` to include the whole stack after the megamerge
stage = ["stack", "closest_merge(@)+:: ~ empty()"]

# `jj restack` to rebase your changes onto `trunk()`
restack = ["rebase", "--onto", "trunk()", "--source", "roots(trunk()..) & mutable()", "--simplify-parents"]