Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
A developer building a real-time aircraft tracker in Rust discovered that a validation layer was silently discarding about 11% of good data due to a subtle timestamp handling bug. The innovation gate, designed to reject …
A developer built ttsproof, an open-source QA framework for text-to-speech systems that separates structural defects from pronunciation errors, addressing the shortcomings of Word Error Rate metrics. In a blind study aga…
A developer's controlled fault-injection study shows that corrupted training data, such as shuffled labels or white noise, can reduce loss by 62% without any visible spikes, making it indistinguishable from healthy train…
An engineer at Bedvibe Studio, writing under the alias Mormolykos, has released a new open-source library called trainproof after repeatedly shipping the same validation bug in different forms. The tool, which evaluates …
Google DeepMind, with researchers from the University of Texas at Austin, published a paper (arXiv:2608.17981) introducing 'recirculation,' an inference-time technique that feeds deep-layer activations back into shallow …
A developer building an on-premise RAG system for regulatory documents discovered that a semantic cache using cosine similarity was unsafe for questions differing only in polarity. Testing with Spanish administrative que…
Carnegie Mellon University PhD student Michael Isaac, during an internship at cmpnd, implemented Flex, a new DSPy module that lets models rewrite program code, not just instructions, to optimize performance. Flex exposes…
The European Central Bank (ECB) reported in a Research Bulletin published on May 21, 2026, that machine-learning algorithmic trading accounts for 60-70% of equity transaction volumes in the US and other major markets, an…
A developer built a real-time sleep analysis system using OpenAI Whisper and Silero VAD to detect snoring and potential sleep apnea from bedroom audio. The system uses Silero VAD as a gatekeeper to filter silence and onl…
A developer detailed how an out-of-memory crash in a production computer-vision pipeline led to a redesign that treats memory as a first-class resource. The fix, which included bounded queues, uint8 quantization, and off…
Salesforce is hiring an AI Scientist for its Commerce Labs team in San Francisco, offering a salary of $149k–224k/yr, to design and ship machine learning systems for agentic commerce products. The role involves building …
PayPal is hiring a Principal Engineer for Agentic Systems in San Jose, California, offering a salary of $218,000–$324,000 per year. The role involves leading machine learning initiatives, developing and deploying ML mode…
The hex-bench benchmark, a personal evaluation suite for the hex project, tests personal-assistant workflows using Inspect AI's react solver against a local OpenAI-compatible llama.cpp endpoint, with 18 tasks across six …
BMLL Technologies reports that Polars loads a full day of US equity trade data in 4.3 seconds, a roughly 48x improvement over pandas' 3.5 minutes, enabling complete analysis without sampling. The benchmarks, run on BMLL …
Researchers from UIUC and Apple introduced STARFlow2, a unified multimodal model built on the Pretzel architecture that interleaves a frozen pretrained vision-language model with a TARFlow stream via residual skip connec…
FreeToken, a local inference tool, enables running mixture-of-experts models with over 290 billion parameters on a single consumer GPU by streaming only active experts from system RAM, avoiding the need for multi-GPU set…
Escha Labs Inc. released Escha-W2, a 2-bit quantized build of Qwen3.8-27B that compresses the 27-billion-parameter model to 10.15GB, enabling 128k context on a single 24GB GPU while matching FP8 quality on benchmarks. Th…
Tencent released EVE Preview 4.5B, a 4.5 billion parameter visual document retriever that matches text queries against document page images directly without OCR, ranking first on ViDoRe V3 and earlier benchmarks. In hand…
In a benchmark dubbed Miko-bench, 65 AI vision models on OpenRouter guessed the breed mix of a puppy named Kumiko, and the consensus pick of Australian Cattle Dog matched the top breed in her Embark DNA report at 47.4%, …
A study using AI-based tissue clocks reveals that human organs age on different timelines, and these patterns can be detected from blood. The research indicates that the chronological age from birth certificates does not…