Roberto Bifulco
@RoBBifulco
Followers
396
Following
2K
Media
69
Statuses
688
Leading the Intelligent Software Systems group at NEC Labs Europe, covering AI agents, core ML, computational methods, cybersecurity and platforms!
Heidelberg
Joined November 2011
Our latest work explores how #LLMs can automate CVSS risk scoring, improving #riskassessment speed and scalability even when labelled data is scarce. Learn how in our #EMNLP2025 paper: https://t.co/VzOed1VYks.
#NECLabs
0
2
5
đ¨I'm beyond ecstatic to announce that we've raised $6M in seed funding to launch Unikraft Cloud -- the only platform that can start *any* workload in < 10ms and scale to 100K+ instances on a single server with instantaneous scale to zero...think catering to millions of users on
6
10
46
Stream. Scroll. Surf. Get speeds up to 400+ Mbps to connect with friends and family without delays.
66
239
1K
Our latest research presents a framework that enhances deep #graphnetworks by optimizing message depth and filtering, effectively modeling long-range interactions while overcoming limitations like oversmoothing and underreaching. Learn how: https://t.co/FAC3mTN5Fs.
#NECLabs #ML
0
4
7
Our new physics-informed method for training machine-learned #interatomicpotentials addresses atomistic simulation challenges by improving robustness & accuracy, reducing errors & minimizing data needs. Learn how in our #ICML2025 accepted paper: https://t.co/mENecnkO3o.
#NECLabs
0
5
13
In computer science, you need to know the system that youâre building on top to have any real leverage. You need to know what will and wonât work, and why. This also generalizes to most fields, which is precisely why AI will take far fewer jobs than people think.
A truism of CS is that you need to understand at least one level of abstraction below where you operate. - OS devs -> hardware - systems devs-> OS - app devs-> systems Vibe coding is no exception.
37
59
512
We also had fun building a small integration with @langfuse for data collection and then to feed data to a @streamlit dashboard. It was an easy and convenient way to quickly test our work and make it more usable
0
0
2
We apply these metrics in classification tasks. Sensitivity tells us how rephrasing of the prompt affects the LLM output. Consistency tells how an element of a class is affected by rephrasing.
1
0
1
Federico Errica is presenting at NAACL25 two metrics we use to assess the quality of LLM prompts: sensitivity and consistency. We started the work after seeing how LLMs changed behavior when changing variable names in our pydantic classes (we were using @jxnlcoâs instructor)
3
2
4
Congratulations to B. BĂźtĂźn, D. de AndrĂŠs HernĂĄndez, M. Gucciardo & M. Fiore from @NECLabsEU, @IMDEA_Networks & @uc3m on winning #BestPaperAward at INFOCOM 2025 for DUNE â a framework for deploying #ML models in #network user planes! Learn more: https://t.co/cvlEtRuaTd.
#NECLabs
0
1
4
Our new benchmark â AL4PDE â uses active learning to solve partial differential equations more efficiently with #neuralnetworks. It reduces data needs and lowers errors by up to 71% compared to random sampling. Learn how: https://t.co/TCYP4JBazC.
#NECLabs #AI #MachineLearning
0
3
6
Pandas is dying a slow, painful death. It's the world's most popular data library, but it's slow, and many libraries have significantly improved over it. The problem with many of these alternatives is that nobody wants to learn a new API. Let's face it: people won't migrate
72
231
3K
Federico Errica, Giuseppe Siracusano, Davide Sanvito, Roberto Bifulco: âWhat Did I Do Wrong? Quantifying LLMsâ Sensitivity and Consistency to Prompt Engineeringâ, accepted at NAACL 2025:
0
0
1
"What Did I Do Wrong? Quantifying LLMs' Sensitivity and Consistency to Prompt Engineering' will appear in NAACL. Thanks to Federico and all co-authors for the fun work (Linkđ) You can guess what is a regular source of frustration in the modern software development process...
While #LLMs boost productivity, debugging their inconsistent behavior is still a challenge. Our novel sensitivity & consistency metrics can help improve #promptengineering, enhancing LLM reliability & effectiveness in complex systems. Learn how: https://t.co/o1cMCGBDCa.
#NECLabs
1
1
5
đŻ
how come nobody is talking about how much shittier eng on-calls are thanks to blind integrations of AI-generated code? LLMs are great coders but horrible engineers. no, the solution is not âprompt the LLM to write more documentation and testsâ (cont.)
0
0
0
Congrats to @ViktorZaverkin and the team đž I loved how this paper included a larger collaboration within the group, down to the platform-level bits (and bytes đ)
#NECLabs is accelerating #chemicalsciences with #neuralnetworks built using higher-rank irreducible Cartesian tensors â our new machine-learned interatomic potential! Our model matches DFT accuracy at a fraction of its computational cost. Learn how: https://t.co/SrtGm2wdWJ.
0
0
4
From Oct. 1st until Oct 4th, I will be in Budapest, attending the European Big Data Value Forum. If you are in the area nad want to discuss about potential collaboration on Gen AI projects, feel free to drop me a message or come directly to our NEC Laboratories Europe booth.
0
1
1
This.
@rtk254 I find it so surprising that almost no-one in the AI alignment community studies organizational behavior, corporate governance, etc. We have decades of actual real world experience of dealing with super-human entities that can both greatly benefit and harm mankind.
0
0
0
I had my first real coding session with @AnthropicAI Claude powered by 3.5 Sonnet + artifacts. It worked remarkably well in generating the code I needed, ready to use as-is. The contribution to ux provided by the intuitive artifacts interface should not be underestimated!
0
0
3