GoKawiil
Tech News
A technical blog post walks through how the Fourier transform can be used to reconstruct a hand-drawn path, such as a llama, by summing rotating vectors (epicycles) of different frequencies, radii, and starting angles connected tip to tail. The author shows progressively complex examples, from a straight line and square to a fish, culminating in a llama drawn using 1024 circles.
Apple has introduced LensVLM, built on Qwen3.5-9B-Base, which lets vision-language models scan compressed text-as-image inputs and selectively expand only relevant portions using learned tools rather than reading everything at full resolution. The model reportedly matches full-text accuracy at 4.3x compression and beats retrieval and compression baselines up to 10.1x compression across seven QA benchmarks, plus document and code tasks.
Apple's AI research team has published LensVLM-9B, a 9-billion-parameter vision language model that processes long documents by first compressing text into image form and then selectively decompressing only the pages it judges relevant, using learned tools. The model, code and demo scripts have been released on GitHub and Hugging Face under Apple's Machine Learning Research Model License and Sample Code License, with configurable compression ratios of 5x, 10x, or 15x.
A blogger demonstrated that gzip, primed with a corpus of Shakespeare text, can 'continue' a prompt by finding byte sequences that compress most efficiently against its sliding window, producing text that is semi-coherent but clearly reflects patterns from the source material. The experiment was inspired by a paper arguing that prediction and compression are mathematically equivalent, since both rely on assigning probabilities to what comes next.
Researchers from Stealth Software Technologies and the University of Vermont have developed ZK-JPEG, a zero-knowledge cryptographic system that proves a JPEG image was correctly compressed from a genuine, committed source photo. The tool also supports common edits like blurring or redacting regions while still generating a valid proof, using the PicoZK framework to translate Python-based image editing into a zero-knowledge circuit built on line-point zero knowledge proofs.
PrismML, a startup founded by Caltech researchers led by CEO Babak Hassibi, has released Bonsai 2 27B, a compressed version of Alibaba's open-source Qwen3.8 27B model that shrinks it from full size down to 5.9 GB. The compressed model retains about 98% of the original's benchmark performance, up from 95% in the prior Bonsai release, while being small enough to run on PCs and possibly high-end smartphones. The company has raised a $22.25 million seed round and counts Databricks co-founder Ion Stoica as an advisor.
DeepSeek released V4.1 Flash, a model initially mistaken for a minor update but revealed via its technical report to be a significant architectural overhaul, effectively a V5-class release. It achieves near 420 tokens/second throughput while compressing KV cache by 4x through techniques including cross-layer compression, sparse attention indexing optimizations, and FP4 precision, alongside a YOCO-inspired prefill design that only activates 8B parameters during prefill versus 16B during decode across its 40 layers.
A developer working at 505 Games on porting a custom PC game engine to consoles published a technical blog post explaining the hidden complexity of converting texture data between platforms. Using a 1024x1024 BC7-compressed wood texture as an example, the post walks through concepts like block compression, texel ordering, mip levels, and texture tiling that must all be correctly handled together to avoid corrupted images.
A compression engineer revisits JPEG XL's standing on the Web, noting that despite its technical strengths over JPEG and WebP, Chrome dropped support in 2023. He points out that a new Rust-based JPEG XL decoder has since appeared in Firefox and parts of Chrome, partly to avoid repeating a security flaw once found in WebP, but argues this alone doesn't settle whether JPEG XL deserves broad Web adoption.
BZip3, an open-source successor to BZip2, combines a Burrows-Wheeler transform with suffix arrays, an LZ77/PPM-style matching pass, and an order-0 context-mixing entropy coder to boost both compression ratio and speed. In a benchmark compressing every historical Perl 5 release bundled into one tar file, BZip3 with a large block size produced a much smaller output than bzip2, xz, and zstd, while also decompressing faster than the other tools tested.
A new open-source utility called balrogg recompresses existing Ogg Vorbis and Opus audio files without any loss of quality, typically shaving 8-12% off .ogg files and 3-8% off .opus files. The command-line tool, licensed under GPLv3 and hosted on GitHub, supports batch processing across multiple cores, adjustable compression effort levels from -1 to -9, and optional progress reporting during encoding and decoding.
A report explains that streaming platforms must compress 4K video heavily to send it over the internet without buffering, which strips out fine details like film grain, shadow gradients, textures and complex motion data. A 1080p Blu-ray, by contrast, uses a much higher bitrate and preserves that visual information more faithfully, so it can look sharper in demanding scenes despite having fewer pixels.