Konstantinos Kitsios

To mitigate this drop, we propose and evaluate the use of contrastive learning, which naturally ranks similarity between objects, thus enabling more effective semantic clone detection in-the-wild.

🙏 Huge thanks to my amazing co-authors, Francesco Sovrano, Earl T. Barr, and @sback.it. [4/4]

By evaluating six models on clones of unseen functionality, we observe a significant performance drop for models explicitly trained for clone detection. For general-purpose LLMs, the drop is lower, but still exists. [3/4]

SOTA clone detection models are trained on clones of specific functionalities and tested on different clones of the same functionalities. But in practice, developers need to identify clones of functionalities the models have not been trained on. How well do models perform in such scenarios? [2/4]

🧩 Can semantic code clone detectors really detect clones in-the-wild?

🎉 We address this question in our paper “Detecting Semantic Clones of Unseen Functionality,” recently accepted to @aseconf.bsky.social!

📄 Pre-print: arxiv.org/abs/2510.04143
💻 Code: github.com/kitsiosk/uns...
[1/4]

Europe’s tech sovereignty watch | Proton for Business Europe’s biggest businesses run on US tech — putting its privacy and sovereignty at risk. Read our study on how bad the problem is and why we urgently need a Europe-first tech policy.

Europe’s digital backbone is built on foreign rails.

In 🇩🇪 DE (58%), 🇦🇹 AT (59%), 🇧🇪 BE (80%), 🇮🇹 IT (69%), 🇱🇺 LU (78%), and 🇳🇱 NL (81%), publicly listed companies rely on US email, and the wider stack behind it.

Read the full study for additional details. 👇

My keynote from Open Source Summit Europe 2025 is now up. 13 pretty packed minutes.

https://youtu.be/YEBBPj7pIKo?si=DBxSCFuqkFQBRdOw

Maybe listening to Greek ρεμπέτικο (rebetiko) while programming will fill our souls with meaning that LLMs can never bring us.

🙏 Many thanks to Marco Castelluccio for the great collaboration, and to @sback.it for his invaluable mentorship during this work.

📄 Preprint: arxiv.org/abs/2509.01616
💻 Code: github.com/kitsiosk/blast

#ASE2025

5/5

We deployed BLAST in three open-source repositories from @mozilla.org, where it proposed 11 fail-to-pass tests to the developers, 6 of which were confirmed to reproduce the designated issue. This calls for scrutiny towards the widely used fail-to-pass metric, which we discuss in detail. 4/5

BLAST generates such fail-to-pass tests in 151 out of 426 (35.4%) issue-patch pairs from a widely used benchmark, outperforming state-of-the-art approaches while requiring only 2 LLM queries and 1 minute of lightweight SBST generation. 3/5

We introduce BLAST, a tool that combines LLMs and Search-Based Software Testing (SBST) to generate tests that fail before a patch and pass after. 2/5

🐞 After a bug is patched, how can we increase our confidence that it will not reappear in the future?

We address this question in our paper recently accepted to @aseconf.bsky.social 2025! 🎉 1/5

AI crawlers are wrecking the open internet.

My small side project - techpays .com - used to generate below 100GB of traffic per month. It’s on Render where 500GB/month included, above it’s $30 per 100GB.

Meta’s AI crawler + other bots have pushed it to 700GB+ per month

WTH

Users deserve data protections. Sign our petition to help them get it. mzl.la/41YuAPG

Already mentioned by someone else here, but is there any practical way a dev could support 🇺🇦?

aischolar.0x434b.dev Pretty cool project by @434b.bsky.social: A neat web interface to explore security (and in particular: Fuzzing) papers with AI summaries. Seems super useful to get/stay up to date with recent papers :)

As a software eng, it is inherently satisfying to see an open approach beat close approaches in an innovative field.

Linux is open: Windows is closed

Llama, Deepseek, Mistral are open: OpenAI, Gemini, Anthropic& many others others closed

Closed approaches winning almost always lead to monopolies.

Congrats!

This is just a reminder that training on test data is all you need to achieve SOTA perf

OpenAI had access to all of FrontierMath data from the beginning, but they verbally agreed that data would not be used in model training. Although there was a legal agreement not to disclose the partnership

Listened to the Telepathy Tapes. It is a great illustration of what you get when incompetent people try to do science. The show is actively misleading (and the makers know it) with purely political and financial goals (the rest is confirmation bias combined with incompetence).

Wow, these are indeed top-notch researchers in the field, totally worth reading some of their work.

Waterfall coming down from a mountain hill

Staubbach Falls in the Swiss Alps 🇨🇭

What Dutch directness looks like. The CEO of ASML was pushed by US partners how ASML supplying devices to China could enable eg actions against the Uyghurs. He responded asking how this is different to what gun manufacturers might enable!

From Focus: the ASML Way by Marc Hijink

The more simplistic the take you see on here, the more it generalizes, the more it is an act of politics, not science.

If the take comes from a scientist, they know they have no evidence for their take, or they would use it.

Here is also a VM from scratch in Python instead of C, by Greg Wilson: third-bit.com/sdxpy/vm/

I see so much FUD about the future of sw engineering, mostly from non-devs. Along the lines of “soon anyone can spin off AI agents in bulk that act as hundreds of devs.”

A false premise. Just open your airline app that is built by ~hundreds of devs over 10+ years

Thanks for compiling this. I would add @mboehme.bsky.social

I posted the exact same question on Threads, X and Bluesky, asking what IDEs devs liked the most.

Despite having 10x as many "followers" on X, Bluesky, and a similar one on Threads, Bluesky is most definitely where the most humans respond to a technical post.

I love it here!

YouTube video by David McCan ACF Pro: Nulled by WordPress.org

Automattic -the creator of WordPress, a company raising $950M in VC funding- took a paid WordPress plugin built & owned by another dev and re-published it, making it free.

If you have a business selling a paid WP plugin: Automattic can null it, anytime.

A new low.

www.youtube.com/watch?v=nYHK...