All news

How a city builder game was made to run on a 16 MHz CPU.

Uses LLMs to generate ranking functions that prove loop termination.

IBM releases granite-speech-5.0-2b and updated smaller speech models for transcription, translation and speaker attribution.

MonoTemp3D adds lightweight short-horizon temporal fusion to streaming monocular 3D car detection.

How to stream OBS or FFmpeg straight into a video call with WHIP for much better screenshare quality.

TRACE does memory-efficient class-incremental learning on time series via temporal-spectral reconstruction.

Stalwart, a Rust mail and collaboration server with JMAP, IMAP, SMTP and clustering, reaches v1.0 after five years.

Uses autoencoder features of raw bus GPS data to detect both late trips and bad service days in Niterói.

Researchers remotely disabled a vulnerable updater in an abandoned Popcorn Time fork that exposed millions of devices to RCE.

Inside-JEPA adds relational grounding to latent world models.

Deep reinforcement learning task scheduling for mobile edge computing that accounts for providers overselling capacity.

Uses curiosity-driven reinforcement learning to schedule requests collaboratively across mobile edge and cloud servers.

ExCal detects multivariate time series anomalies with no training, using retrieval and robust statistics.

CPython's experimental JIT is set to be removed after it failed to deliver enough speedup for its maintenance cost.

IBM shows ensemble forecasting across four Granite time-series models behind one interface.

Writes continual updates forward-only into an LM native n-gram memory table.

Eight compromised TLS intermediate certs from eMudhra put the internet on high alert.

PRISMA creates differentially private synthetic car telemetry with differentiable simulation.

Uses contextual sparsity to get around the memory bandwidth limit in edge LLM inference.

Zig now uses partial operator precedence, so mixing some operators without parentheses is a compile error.

Advantage-based rule for when on-policy distillation should follow the teacher.

Near-tight distributed approximation of minimum dominating set in sparse graphs.

Timestep-aware quantization for diffusion LMs.

An evaluation of webframe, a light browser JavaScript runner, finding it fast enough for scraping and usable as a sandbox.

Matrix conference talk on redesigning the chat timeline data model for the next protocol generation.

Proposes serving RAG workloads through decentralized collaboration of edge devices instead of the cloud.

How OpenAI turned its coding tools into dots, an always-on assistant that works in the background.

Explains why sending email at scale costs money: IP reputation, deliverability, bounce handling and anti-spam rules, not the SMTP request itself.

Cloudflare WAF now inspects each request together with its response to spot successful exploitation, and adds response rules.

Argues software is slow because of design choices, not language, so AI rewrites in fast languages will not help most programs.