RT @CoolSWEng: This is Twilight Zone stuff. In response to a prompt for a potential research topic I was exploring #ChatGPT (4o) hallucinat….

RT @EndorLabs: 🌟 Big News Alert! 🌟. Join us in cheering for @jhejderup one of our brilliant program analysis team members, as he defends hi….

RT @gdrosos_: Along with @theosotir, @CoolSWEng and @d1m1tr0 we've investigated code bloat in Python applications from various perspective….

RT @irina_kAl: Look Ma, I’m an influencer now!.Thank you @msrconf for this award ❤️and huge thanks and congratulations to all the coauthors….

RT @EndorLabs: Congratulations to our very own @gousiosg for being a part of the team to win most influential paper award! 🎉#msr2024 ⁦ . ht….

RT @msrconf: The MSR'24 Most Influential Paper Award .Congrats to the authors 👏

RT @azaidman: And the most influential paper award goes to ⁦@irina_kAl⁩ ⁦@lsinger⁩ ⁦@gousiosg⁩ Daniel German ⁦@KellyBlincoe⁩ ⁦@DanaHDamian⁩….

RT @KadiGrigg: Unreal week. I always knew I wanted to be back in Greece since I left in 2010 after my stint in #archeology I am blessed tha….

RT @msrconf: 🏆The most influential paper of MSR 2024 is "The promises and perils of mining GitHub". Congrats to its authors. 👏.The selectio….

RT @puerco: Breaking down dependency resolution with @gousiosg from @EndorLabs at the #FOSDEM #SBOM DevRoom

Someone with ample resources should train a 700M-1B parameter model with the Mistral training protocol on the Stack dataset, with both generative and MLM objectives. A quantised version of such a model could relatively easily be included in an IDE. (2/2).

There is a large space for code ML models between CodeBert and StarCoder/CodeLLama. CodeBert is great for small methods, but does not really know about context. CodeLLama has enough context, but needs 15GB of RAM and cannot create embeddings. (1/2).

During our ML4SE course, while discussing the 💫StarCoder paper, a student makes the following remark:. - "💫StarCoder is an open model as opposed to code-davinci from OpenAI". stops talking a bit, thinks, smiles. The whole class burst into laughter. OPEN AI indeed!.

RT @CoolSWEng: Introducing the ai-cli library, a command-line copilot. It attaches to programs that offer interactive command-line editing….

RT @EndorLabs: How it started vs. how it's going. This has been a wild day.

RT @EndorLabs: 🚨Endor Labs raises a $70M series A!🚨. Things are heating up on our little forest moon as we set our sights on eliminating th….

Παρατήρησα μια σημαντική οπισθοδρόμηση στη χρήση POS για πληρωμές σε σχέση με περισυ. Εδώ στην Πρέβεζα που είμαι πάνω από 30% είναι «χαλασμένα». Είναι τα κόστη χρήσης (9 ct/συναλλαγη σύμφωνα με ένα περιπτερά που ρώτησα) τόσο υψηλά; Η μηπως άλλαξε κάτι στους ελέγχους;.

Time and again, I 've found that the only way to get performance out of a system is to avoid doing work (often at the expense of memory). Concurrency is only to be applied when 1. the problem is amenable, 2. all other optimisations have been explored in the non-parallel code path.

RT @kAsterios: Please retweet!.I am searching for two PhD students and one postdoc for my VIDI grant. We are building a programming model &….

RT @MalihehIzadi: While the SE community is attending @ICSEconf, @msrconf + co-located events, perhaps it’s the right time to share an upda….