Skip to content
Tech News
Topics: Today This Week This Month This Year

Qwen3.8 27B benchmarked on Mac Studio M3 Ultra: half the speed, third the tokens

A user ran Alibaba's new Qwen3.8 27B model locally on a Mac Studio M3 Ultra for 10 days, using it for daily tasks like summarizing RSS feeds and renaming scanned PDFs, then formally benchmarked it against its predecessor. The Q4_K_M version generated at about 14 tokens per second versus 28.6 tokens/s for Qwen3.6 27B, but produced answers using roughly a third as many tokens, making total response time nearly equal. A 1-bit quantized version ran much faster at 27 tokens/s and stayed factually accurate but struggled to commit to definitive answers.

EasyEffects lets Linux users tune laptop speakers via PipeWire audio plug-ins

EasyEffects is a Linux application that taps into the PipeWire audio system to apply real-time sound processing—equalization, compression, bass enhancement, limiting and more—to a laptop's built-in speakers. Because it works directly with the audio streams entering and leaving the system, users can noticeably improve the thin, tinny sound typical of laptop speakers without buying external hardware. The tool bundles plugins from projects like Calf Studio Gear, MDA.LV2 and GStreamer with adjustable presets, all managed through one interface instead of separate mixers or command-line scripts.

Z.ai releases GLM-5.3 with open model weights

Z.ai has published GLM-5.3 as an open-weight model, making its parameters publicly available for developers to download and run. Unsloth is preparing GGUF quantized versions of the model to allow it to run efficiently on local hardware.

Nvidia launches employee PAC to boost its Washington lobbying power

Nvidia has registered a new federal political action committee that will collect voluntary donations of up to $5,000 a year from eligible employees to fund candidates and party committees. The company told staff that upcoming congressional decisions on AI regulation, export controls and infrastructure could significantly affect the industry. This move follows Nvidia's hiring of a new chief external affairs officer and five outside lobbying firms, alongside a jump in federal lobbying spending to over $2.5 million this year.

Debate over TUIs vs GUIs misses the real issue: keyboard navigation design

A widely discussed Hacker News post argued developers should favor GUIs over terminal user interfaces (TUIs), sparking pushback from commenters who claimed TUIs are superior because they're keyboard-driven. The author counters that keyboard-driven design isn't exclusive to TUIs, pointing to guidelines like GNOME's HIG that explicitly require GUIs to support full keyboard navigation.

Benchmarks show musl libc lags glibc by up to 26% even with mimalloc swapped in

A developer investigating containerized Rust performance found that musl's default memory allocator is significantly slower than glibc's, not just under high concurrency but even on modest 4-core EC2 instances. Swapping in mimalloc or jemalloc narrows the gap but doesn't close it—one workload remained 26% slower—because several of musl's core memory routines are themselves inefficient, not just its allocator.

Oak Deer and Hikrobots showcase AI cooking robots at Beijing's World Robotics Conference

At the World Robotics Conference in Beijing, Oak Deer Robotics unveiled a suite of automated kitchen tools including a multimodal AI cooking model called CookingMuse and a humanoid food-cart chef named RobotCook, demonstrating stir-fried tofu and pork with peppers using a 300-degree robotic wok. Separately, Hikrobots showed humanoid robots performing basic fast-food prep, while a robot-run noodle shop opened for trial operation elsewhere in Beijing.

Today's top topics: iphone duo openai apple ai agents meta tesla php2shell robotaxi microsoft elon musk
Browse all topics →  ·  Today's trending topics →