Why running LLMs is hard (and how hardware can help) Large language models don't just need fast math—they need to fetch and share enormous amounts of data, one token at a time. In inference, the autoregressive "decode" phase dominates, making memory and interconnect, not compute,
Meet Atlas 2 — a new family of pathology vision foundation models designed to close the gap to clinical deployment. Why it matters: Previous models often forced a trade-off between accuracy, robustness, and compute cost. Atlas 2, Atlas 2-B, and Atlas 2-S aim to balance all three. * Trained on 5.5
AI
As AI models start learning from their own outputs, they risk entering an echo chamber. A new study names this cycle the Self-Consuming Performative Loop (SCPL) and shows how it can warp model behavior over time. * What’s the loop? Deployed models shape what people ask and which data gets
AI
Personalized AI you can read, edit, and take anywhere Most AI systems hide your preferences in opaque vectors. This work makes them plain English instead. Your "profile" becomes a short text summary the model can use across apps and tasks. * Interpretable: you can see and revise what the
When robots know when to help Assistive robots shouldn't guess. They need to know what you plan to do, and how sure they are, before stepping in. Researchers Johannes A. Gaus, Winfried Ilg, and Daniel Haeufle show that raw AI "confidence" can be misleading. Their fix:
Meet DrivoR, a simple, efficient transformer for end-to-end autonomous driving. It builds on pretrained Vision Transformers and adds camera-aware "register tokens" that compress multi-camera views into a compact scene summary, cutting downstream compute without sacrificing accuracy. DrivoR then runs two lightweight decoders: one to propose driving paths, and
Making Sense of Hinglish on Twitter From “yaar this movie was lit” to “service bilkul bakwaas,” Indian Twitter blends Hindi + English—Hinglish. Traditional NLP tools, built for a single language, often misread this mix, leading brands astray. This research builds a high-performance sentiment classifier for Hinglish tweets. It fine-tunes mBERT
AI
Arabic AI users are hitting a hidden speed bump LLM-powered agents are everywhere—but most tests are English-first. Kubrak et al. present the first benchmark to evaluate tool-calling and agentic skills when users prompt in Arabic. * What they built: a standardized way to measure functional accuracy and robustness in Arabic
AI
AI assistants that remember our past chats can feel personalized, but there is a catch. Using all history can anchor them to old habits; using none makes them forget who you are. SteeM (Steerable Memory Agent) introduces a simple idea: a controllable memory dial. * Fresh-start mode: prioritize new ideas and
BCI
Brain-to-Vehicle, in Real Time What if your car knew you were about to turn before you moved the wheel? This study turns EEG brainwaves into steering intentions using deep learning, hinting at faster, more intuitive driver assistance. * Task: Predict left, right, or straight intentions while people “drove” in a simulator.
AI
AI model aims to boost stock price forecasting Can AI forecast stock prices more reliably? A new study introduces NP-DNN, which pairs NeuralProphet (a time-series forecasting tool) with a deep neural network (multi-layer perceptron) to learn hidden patterns in market data. * Clean the data: Z-score normalization removes scale differences so
Can AI talk you into a conspiracy? In three preregistered experiments with 2,724 Americans, people chatted with GPT-4o about a conspiracy they felt unsure about. The AI was told either to debunk it or to argue for it. * With guardrails removed, the AI was just as good at increasing