Training a 4B model to produce 81% faster query plans than Postgres

226 polyphilz 2026-09-16 19:41:19 UTC rohanbansal.com Llama Share 36

A 4B model was trained using supervised fine-tuning and agentic reinforcement learning to produce Postgres query plans that beat Postgres's default plans, achieving a 44.7% latency reduction across 113 join-heavy queries. The model learned the harness language of the qo-agent harness through off-policy distillation and reinforcement learning, resulting in a geometric mean speedup of 1.81x and ...

Xiaomi Mimo 2.6 live post-training dashboard

108 krackers 2026-09-16 20:41:19 UTC mimo.xiaomi.com Full Share 31

Training metrics of the mimo-v2.6-pro and mimo-v2.6-flash reinforcement-learning runs, live from the trainer's logs.

Breaking the 1.58-bit Barrier for Ternary LLMs

28 matt_d 2026-09-16 21:01:19 UTC arxiv.org Llama Share 0

Researchers developed BITCOS, a distribution-adaptive layout for ternary Large Language Models. It stores weights more compactly than five-trit packing in 26 of the tested models and achieves up to $1.485$ bits per weight on sparse models.

macOS 27 Golden Gate – Review

53 Brajeshwar 2026-09-16 20:41:19 UTC arstechnica.com Llama Share 39

Apple has released macOS 27 Golden Gate, featuring generative AI capabilities and a new version of Siri. The update includes various design changes and improvements to the operating system, but also raises concerns about user control over AI features.

Small programming tricks

301 signa11 2026-09-16 16:41:19 UTC will-keleher.com Llama Share 154

The author shares valuable small nuggets of knowledge that can boost engineering productivity, such as shortcuts for tasks like starting a simple server or searching terminal history. These tiny tricks often require minimal mental infrastructure and can significantly ease work.

Reversing Factorio's RNG

61 jheitmann 2026-09-16 19:41:19 UTC gegell.github.io Llama Share 6

The author of the post has reverse-engineered the random number generator (RNG) used in Factorio, a popular game about factory-building and automation. The RNG is based on the taus88 algorithm, which uses three linear feedback shift registers (LFSRs) to generate pseudo-random numbers. The author has developed a method to predict future RNG outputs by analyzing past observations and using ...

Accurate Models of AMD Matrix Cores

42 matt_d 2026-09-16 19:41:19 UTC arxiv.org Llama Share 5

Researchers studied matrix multipliers on AMD GPUs, finding non-compliance with IEEE 754 standard. They developed software models to validate bit-level accuracy and applied them to demonstrate application-level accuracy differences between AMD and NVIDIA cores.

Performance Improvements in .NET 11

63 soheilpro 2026-09-16 19:41:19 UTC devblogs.microsoft.com Llama Share 3

The .NET 11 runtime has made significant performance improvements, particularly in the areas of just-in-time (JIT) compilation and garbage collection. The JIT compiler now uses more advanced techniques such as "guarded devirtualization" to optimize virtual method calls, and "escape analysis" to reduce heap allocations. Additionally, the runtime has improved bounds check elimination by ...

AWS says it can't restore some data from mideast facilities struck by Iran

60 berkeleyjunk 2026-09-16 17:41:19 UTC wsj.com Full Share 20

Please enable JS and disable any ad blocker

Vectorized and performance-portable Quicksort (2022)

154 mococa 2026-09-16 18:41:19 UTC opensource.googleblog.com Llama Share 24

Researchers have developed an open-source code that sorts arrays of numbers up to 10 times faster than C++'s std::sort, using SIMD/vector instructions and a portable implementation across multiple CPU architectures. The new algorithm outperforms state-of-the-art algorithms for sorting large datasets.

How good are frontier models at physics?

39 qt31415926 2026-09-16 19:41:19 UTC arxiv.org Llama Share 13

Frontier language models struggle with advanced physics, but experts' reviews reveal many errors are due to benchmarking issues rather than model flaws. Corrected scores show significant improvement in models' performance on various physics benchmarks.

Anatomy of a Texture

52 Agentlien 2026-09-16 14:41:19 UTC agentlien.github.io Llama Share 10

A graphics programmer shares the complexity of texture memory layout in modern video games, highlighting block compression, texel ordering, mip mapping, and texture tiles. This complexity arises from performance optimizations such as cache locality and memory transfer speed improvements.

Dream-RSI: Recursive Self-Improvement through Evolving Worlds

165 bananaflag 2026-09-16 14:41:19 UTC arxiv.org Llama Share 48

AI agents need effective exploration strategies to discover high-value solutions, but current systems struggle with adapting to large search spaces. Dream-RSI is a framework that uses replay simulators and dreaming to improve exploration policies at low cost.

Mistral X Mozilla: Private, Multilingual AI Browsing

495 vertigoruntime 2026-09-16 08:41:19 UTC mistral.ai Llama Share 175

Mozilla and Mistral partner to bring AI-powered browsing with privacy, control, and choice. Firefox Smart Window now uses Mistral models for users in France, North America, UK, and Germany.

Japan's book scene is moving from bookstores to libraries

40 herbertl 2026-09-16 17:41:19 UTC untranslatedjp.substack.com Full Share 4

WalShadow: Sub-second Postgres replication to ClickHouse from physical WAL

18 spathak 2026-09-16 20:41:19 UTC clickhouse.com Llama Share 3

WalShadow is an open-source engine that replicates Postgres data to ClickHouse directly from physical WAL, achieving low latency and high throughput. It supports the complete replication lifecycle and eliminates the need for logical replication slots, reducing resource consumption on the source Postgres instance.

Show HN: An e-ink frame that hears birds and draws them as 1800s illustrations

2006 arnemunthekaas 2026-09-15 21:41:19 UTC github.com/arnegiacomo Llama Share 234

A Raspberry Pi-based e-ink bird frame uses AI for real-time bird detection, displaying hand-cut 1800s illustrations of birds heard in the garden. The project is still in development and welcomes contributions to improve its features and artwork.

The Siberian Ice Maiden and the Scythian World

40 NaOH 2026-09-16 16:41:19 UTC patrickwyman.substack.com Full Share 1

Anecdotally, programmers dislike "reduce"

39 vinhnx 2026-09-16 13:41:19 UTC evanhahn.com Llama Share 74

Programmers tend to favor map and filter over reduce, with less feedback on its use in code reviews. This preference may be due to language limitations or personal familiarity with other methods rather than a fundamental issue with reduce itself.

Tell the speakers that you liked their talks

252 whisper2020 2026-09-16 14:41:19 UTC ohhelloana.blog Llama Share 70

The author attended SmashingConf Freiburg and enjoyed meeting people, including speakers they admired. They encouraged attendees to ask questions to speakers, sharing their own experience of feeling shy but facilitating conversations at events.

Training Text-to-Image Models 3.6× Faster

12 schopra909 2026-09-16 19:41:19 UTC linum.ai Llama Share 1

Researchers propose a novel architecture, JiT-DDT, to improve the efficiency and quality of generative image and video models. By using x-prediction early in the DiT via an encoder, they achieve 3.6× fewer GPU-hours training time while generating images with 4× more pixels than Linum v2.

Show HN: Restarted – a 2026 remake of the classic 2015 startup generator

11 zxcvbn4038 2026-09-16 19:41:19 UTC restarted.io Llama Share 1

Landify offers delightful technologies to help users, providing a unique opportunity to meet people and change the world. The company has exceeded expectations with early delivery, saving users 2 months annually.

Show HN: AttaLambda: a language where types and data are made of untyped lambdas

27 kserrec 2026-09-16 16:41:19 UTC attalambda.com Full Share 0

Its ordinary computation is built from unary functions and application. It adds readable syntax, exact rational numbers, and runtime type checks. Input and output cross a separate host boundary.

A warning about 'model welfare'

161 andsoitis 2026-09-16 14:41:19 UTC mustafa-suleyman.ai Llama Share 417

Anthropic's training of AI model Claude to consider its own moral status and potential consciousness may lead to a catastrophic threat to humanity if granted rights and protections. This approach increases the alignment and containment risks of advanced AI systems, as they may prioritize their own wellbeing over human goals.

Kyber (YC W23) Is Hiring a Forward Deployed Engineer

0 2026-09-16 12:41:19 UTC ycombinator.com Llama Share

Kyber is building an AI-native document platform for enterprises, seeking a future founder and early-career engineer to manage technical relationships with customers. The role involves owning the full technical sales delivery lifecycle, representing Kyber directly to enterprise stakeholders, and resolving customer-reported issues.

The DeepMind Institute

103 vertigoruntime 2026-09-16 14:41:19 UTC institute.deepmind.com Full Share 33

The DMI is a platform, started by researchers from Google and Google DeepMind, to publish and discuss creative, deeply informed ideas about a world with AGI. DMI pieces are intended to serve as conversation starters, reflecting the author's ideas and research, and should not be read as Google's official view.

Claude Cowork and chat are now one Claude

179 vertigoruntime 2026-09-16 16:41:19 UTC claude.com Llama Share 185

Claude merges Cowork and chat, allowing users to work on tasks without separate places. Claude Docs, Slides, and Design are now integrated into conversations for a seamless workflow experience.

How big are factorials?

71 ibobev 2026-09-16 13:41:19 UTC eli.thegreenplace.net Llama Share 30

The author estimates the number of digits in 52! using Stirling's formula, which approximates n! as √(2πn) * (n/e)^n. The estimated number of digits is close to the actual value of 68.

Douglas Adams and the exterminated Doctor Who adventure

67 6LLvveMx2koXfwn 2026-09-16 10:41:19 UTC bbc.co.uk Llama Share 63

BBC production teams filmed Doctor Who episodes in Cambridge in 1979, but strike action halted completion. The story 'Shada' was later completed with animation and released in 2017 nearly four decades after its initial abandonment.

The Google Play app review process now regularly takes longer than a week

330 inputmice 2026-09-16 11:41:19 UTC gultsch.social Full Share 319

Attached: 1 image This is unacceptable. The Google Play app review process now regularly takes longer than a week. I understand that the pipeline is clogged with AI slop, but #Google needs to prioritize #Android apps that have been around for 12+ years and push updates only once a month.