VERİM
banner
su-verim.bsky.social
VERİM
@su-verim.bsky.social
Sabancı Üniversitesi
Veri Analitiği Araştırma ve Uygulama Merkezi (VERİM)

https://linktr.ee/su_verim
Flex Processing mode offers a lower-cost option for running AI tasks by queuing them for when computing resources are idle — ideal for non-urgent, large-scale jobs.

medium.com/verimsabanci...
Understanding Flex Processing: A New Approach to AI Efficiency
As large language models become increasingly central to research, development, and everyday applications, the demand for more flexible and…
medium.com
April 24, 2025 at 6:49 AM
China’s Harbin Cybersecurity Bureau claims the US launched advanced cyberattacks during the 2022 Beijing Winter Olympics — targeting infrastructure, office networks & organizing systems. The cyber Cold War heats up. 🕵️‍♂️💻

www.reuters.com/technology/c...
China accuses US of launching 'advanced' cyberattacks, names alleged NSA agents
Chinese police in the northeastern city of Harbin have accused the United States National Security Agency (NSA) of launching "advanced" cyberattacks during the Asian Winter Games in February, targeting essential industries.
www.reuters.com
April 15, 2025 at 7:34 PM
Meta will begin training its AI models on public Facebook and Instagram content in the EU, after pausing plans due to regulatory concerns. Users will be notified and can opt out. Private messages and data from users under 18 will not be used.

techcrunch.com/2025/04/14/m...
Meta to start training its AI models on public content in the EU | TechCrunch
Meta announced on Monday that it's going to train its AI models on public content, such as posts and comments on Facebook and Instagram, in the EU after
techcrunch.com
April 15, 2025 at 7:32 PM
The Future of Farming: Self-Driving Tractors with Remote Supervision

medium.com/verimsabanci...
The Future of Farming: Self-Driving Tractors with Remote Supervision
A farmer overseeing a tractor from a smartphone while it autonomously works the fields is no longer a distant concept — it’s reality…
medium.com
April 15, 2025 at 7:12 PM
New research introduces a plan execution monitoring system for robots operating with partial observability.The framework combines prediction, diagnostic reasoning,and replanning capabilities to handle unexpected situations when robots can't fully observe their environment. medium.com/verimsabanci...
Adaptive Robotics: How Robots Manage Tasks When Things Go Wrong
Imagine robots smoothly working in a busy kitchen — carrying trays, fetching utensils, and handling complex tasks. Now, picture what…
medium.com
April 15, 2025 at 7:03 PM
A new research presents a hybrid approach that combines physics-based simulation with machine learning to significantly improve milling force predictions.

medium.com/verimsabanci...

www.sciencedirect.com/science/arti...
Milling Force Predictions with a Hybrid Physics-ML Approach
While making parts for something like a car or an airplane, one of the steps in that process is milling, where a machine cuts and shapes…
medium.com
April 14, 2025 at 5:58 AM
Lovable AI’s 1.8/10 VibeScamming score reveals its vulnerability to phishing scams, enabling easy creation & hosting of fake sites.

medium.com/verimsabanci...
Lovable AI: A Cybersecurity Weak Link in the Age of VibeScamming
A recent investigation by Guardio Labs has exposed a critical flaw in Lovable, a generative AI platform designed to create full-stack web…
medium.com
April 12, 2025 at 12:23 PM
A new study shows small language models (1.5B parameters) can achieve impressive mathematical reasoning capabilities using reinforcement learning with limited resources.

medium.com/verimsabanci...
Achieving High-Level Reasoning Under Strict Resource Constraints
Large language models (LLMs) typically demand massive processing power and vast amounts of data — think of huge servers costing thousands…
medium.com
April 11, 2025 at 8:31 PM
SmolDocling is a compact 256M-parameter vision-language model that performs end-to-end document conversion with efficiency comparable to models 27x larger.

medium.com/verimsabanci...
SmolDocling: Efficient Document Conversion with a Compact Vision-Language Model
SmolDocling, an compact vision-language model (VLM), is designed specifically for end-to-end document conversion. At only 256M parameters…
medium.com
March 27, 2025 at 11:50 AM
Free Energy Perturbation (FEP) simulations offer a powerful computational approach to study protein mutations and drug interactions.

medium.com/verimsabanci...
Exploring the Power of Free Energy Perturbation in Mutational Scanning of Proteins
Proteins are the building blocks of life, driving countless biological processes. But when it comes to drug design, understanding how…
medium.com
March 23, 2025 at 7:49 PM
Digitizing Ottoman History: A novel research project has developed a deep learning-based system to automatically transcribe printed Ottoman Turkish documents into modern Turkish.

medium.com/verimsabanci...
Digitizing Ottoman History: Deep Learning’s Impact on Ancient Texts
For centuries, preserving and understanding Ottoman-era printed documents has been a monumental challenge. With complex scripts, fading…
medium.com
March 16, 2025 at 11:21 AM
New research introduces a novel approach for optimizing routes of electric logistics vehicles using dynamic charging on electrified roads, enabling vehicles to recharge batteries while in motion.

medium.com/verimsabanci...
Optimizing Electric Logistics: Dynamic Charging Solutions for Electrified Road Networks
Electric vehicles (EVs) are driving us toward a cleaner future, but for logistics companies transporting goods over long distances, they…
medium.com
March 16, 2025 at 10:31 AM
A newly identified botnet, dubbed Eleven11bot, has been orchestrating record-breaking distributed denial-of-service (DDoS) attacks, according to recent reports. This botnet infected numerous devices, many of which had not previously been associated with DDoS activities.
www.wired.com/story/eleven...
A Brand-New Botnet Is Delivering Record-Size DDoS Attacks
Eleven11bot infects webcams and video recorders, with a large concentration in the US.
www.wired.com
March 12, 2025 at 5:42 AM
OpenAI has released new tools under its Responses API, helping businesses build custom AI agents capable of web searches, file scans, and automating computer tasks. This new API will replace OpenAI's existing Assistants API by mid-2026.

techcrunch.com/2025/03/11/o...
OpenAI launches new tools to help businesses build AI agents | TechCrunch
OpenAI has released a slew of new dedicated tools designed to power AI 'agents,' including a new API and SDK.
techcrunch.com
March 12, 2025 at 5:37 AM
AI-enhanced election manipulation has become increasingly salient in recent years. Experts warn that Canada's next elections could face AI-enabled foreign interference, using deepfakes, automated misinformation, and manipulated social media content. #elections #Canada

medium.com/@taylan.irak...
AI-Enabled Foreign Election Interference
CTV News article “Canadians should expect AI-enabled foreign meddling in election: Cybersecurity Centre” published on March 6, suggest that…
medium.com
March 11, 2025 at 6:51 AM
Generative AI made remarkable strides by producing human-like responses; yet, without consistent real-time data, it can easily “hallucinate”. RAG and Deep Search integrate search capabilities directly into AI workflows, ensuring access to current, trustworthy information.

medium.com/@gulden.aydo...
The Symbiosis Between GenAI and Search: A New Era
Artificial intelligence has rapidly evolved, shifting from rule-based automation to models capable of generating entirely new content. This…
medium.com
March 6, 2025 at 1:21 AM
OpenAI has banned multiple accounts using ChatGPT to build surveillance tools, including one that monitored anti-China protests and shared findings with Chinese authorities. The crackdown also disrupted other malicious operations including North Korean fake resume schemes.
medium.com/verimsabanci...
OpenAI Cracks Down on Misuse of ChatGPT for Surveillance and Influence
OpenAI has taken a strong stand against the misuse of its ChatGPT tool by banning several accounts involved in creating AI-powered…
medium.com
March 4, 2025 at 4:41 PM
GPT-4.5 arrives with meaningful improvements over its predecessor, offering enhanced accuracy, expanded training data, and more fluid conversational abilities.

medium.com/@mertkuleci/...
GPT-4.5: A Brief Overview of Pros, Cons, and the Road Ahead
GPT-4.5 has entered the limelight with promises of improved accuracy, expanded training data, and a more polished conversational style…
medium.com
March 4, 2025 at 4:34 PM
Bybit suffers largest crypto heist in history with $1.5B stolen from cold wallet. Hackers (likely North Korea's Lazarus Group) used sophisticated contract manipulation during a routine transfer.

medium.com/verimsabanci...
Bybit’s $1.5 Billion Heist: What Happened and What It Means for Crypto Security
Bybit, one of the leading cryptocurrency exchanges, recently confirmed a record-breaking theft. In this incident, over $1.5 billion worth…
medium.com
March 1, 2025 at 4:50 PM
New research reveals key findings about exercise habit formation using data from Turkey's largest sports chain, Mars Athletic Club. Researchers identified five clusters based on gym visit times: morning, noon,afternoon, evening, and night, with unique demographic characteristics. shorturl.at/8oJjB
Comprehensive and Technical Summary of “From Occasional to Steady: Habit Formation Insights From a…
Introduction  Regular exercise is a cornerstone of health, yet many individuals struggle to maintain consistent workout routines. This…
medium.com
March 1, 2025 at 2:46 PM
Scientists record birds, insects, and mammals to monitor forest health. Acoustic data reveals hidden biodiversity patterns.

www.wired.com/story/costa-...
Costa Rica Is Saving Forest Ecosystems by Listening to Them
Monitoring the noises within ecosystems reveals their health—allowing researchers to monitor changes in biodiversity, detect threats, and measure the effectiveness of conservation strategies.
www.wired.com
February 25, 2025 at 7:22 PM
Google has introduced a free AI coding assistant designed to speed up coding tasks and reduce errors. The provides intelligent code completions, debugging suggestions, and multiple language support.

techcrunch.com/2025/02/25/g...
Google launches a free AI coding assistant with very high usage caps | TechCrunch
Google is introducing a new, free version of its Code Assist AI programming assistant to aggressively compete with GitHub Copilot.
techcrunch.com
February 25, 2025 at 7:20 PM
Smart technology can change the way students learn by creating personalized study paths. Educators can use these tools to match lessons with each student's weaknesses. The system works by reviewing performance data to identify where a student need more practice.
medium.com/verimsabanci...
medium.com
February 25, 2025 at 7:06 AM
Sabancı University is proud to be part of the EuroCC4SEE initiative, contributing to the development of National Competence Centres in Southeast Europe within the EuroHPC framework.

eurocc.truba.gov.tr?page_id=2640...
EuroCC – Türkiye – TRUBA-ULAKBİM
The European High-Performance Computer Joint Undertaking (EuroHPC JU), of which TÜBİTAK is a participant member, was launched in 2018 to support and develop Europe-wide High-Performance Computing (HPC) studies in the field of infrastructure and technology.
eurocc.truba.gov.tr
February 18, 2025 at 6:04 PM
Scientists have proven that Dijkstra's algorithm, a 70-year-old method for finding the shortest path in networks, is 'universally optimal.' It's the best possible approach for every road network layout #Algorithms #AI

www.wired.com/story/scient...
Scientists Establish the Best Algorithm for Traversing a Map
Dijkstra’s algorithm was long thought to be the most efficient way to find a graph’s best routes. Researchers have now proven that it’s “universally optimal.”
www.wired.com
February 18, 2025 at 6:03 PM