GPT-6 Sol and Luna

570 OfficialTurkey 2026-09-22 18:41:30 UTC openai.com Llama Share 328

OpenAI has introduced GPT-6 Sol and Luna, more cost-efficient models that offer advanced AI capabilities. These models are designed to be used at different scales, rhythms, and budgets, with improved performance in areas such as coding, computer use, and alignment.

Claude Opus 5.5

725 km144 2026-09-22 16:41:30 UTC anthropic.com Llama Share 593

Claude Opus 5.5 has been released with significant improvements in performance, safety, and efficiency, surpassing previous models like Opus 5 and Claude Mythos 5.1 on various benchmarks and tests.

OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005

434 sohkamyung 2026-09-22 14:41:30 UTC cryptocellar.org Llama Share 329

A team led by Carter Leffer used OpenAI's GPT-6 Astra to break a 75-year-old Enigma message, MVUEH, which had resisted solution since 2005. The AI achieved this feat on its own after analyzing the message and using a crib technique, revealing the correct key and plaintext.

Claude Opus 5.5 Intelligence, Performance and Price Analysis (Max)

123 theanonymousone 2026-09-22 17:41:30 UTC artificialanalysis.ai Llama Share 43

Claude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback) is a proprietary model that scores 58 on the Artificial Analysis Intelligence Index and has a context window of 1M tokens.

WordPress: Unauthenticated path traversal leading to conditional RCE

84 vntok 2026-09-22 16:41:30 UTC github.com/wordpress Llama Share 42

An unauthenticated attacker can execute arbitrary code on a WordPress site by manipulating the page-template resolution, potentially leading to Remote Code Execution (RCE). The vulnerability affects certain themes and PHP versions.

Pentagon: Palantir AI Overreliance Led to Strike Killing 123 Iranian Children

83 devonnull 2026-09-22 19:05:30 UTC gizmodo.com Full Share 42

Unreal Agent

26 trollied 2026-09-22 18:41:30 UTC unreallabs.ai Llama Share 10

Unreal Labs developed Unreal Agent to improve AI agent efficiency by reducing tool-management overhead, achieving up to 40% cost savings compared to Codex and 20% compared to Pi. The harness allows for asynchronous tool calls, enabling agents to respond quickly and be cost-effective without the need for polling or waiting.

There's a high chance of devices being sold with GrapheneOS preinstalled in 2027

161 Cider9986 2026-09-22 17:41:30 UTC grapheneos.social Full Share 67

@[email protected] @[email protected] @[email protected] @[email protected] There's a high chance of devices being sold with it preinstalled in 2027 but probably not for the initial launch.

OpenAI is well positioned to fast-follow Jev

197 JohnBerryman 2026-09-22 15:41:30 UTC arcturus-labs.com Llama Share 154

TypeSafe's Jev has taken the AI world by storm with its fast and accurate classification capabilities, but OpenAI may replicate it soon. TypeSafe's training data and reinforcement learning process are their main moat to protect themselves from being copied.

SAML: A Fractal of Bad Design

14 aray07 2026-09-22 18:58:30 UTC blog.trailofbits.com Llama Share 1

SAML authentication protocol is being considered for retirement due to its complexity and security issues, with modern alternatives like OpenID Connect (OIDC) gaining popularity. SAML was designed in the early 2000s but has not received necessary updates, leading to a decline in its adoption.

16-bit Intel 8088 chip (c. 1985)

81 rbanffy 2026-09-22 16:41:30 UTC allpoetry.com Llama Share 10

A 40-year-old poem discusses the frustrations of early personal computing, listing incompatible systems and echoing a lifelong disgust with proprietary technology. The poem's switch to a turkey buzzard's mating ritual serves as a commentary on the timeless nature of human experience amidst technological chaos.

Launch HN: Coverage Cat (YC S22) – Umbrella insurance via your personal agent

23 botacode 2026-09-22 17:41:30 UTC coveragecat.com Llama Share 20

Coverage Cat is a licensed insurance brokerage that compares home, auto, and renters coverage with real broker support. It's available in 5 states and aims to provide honest price ranges through AI-guided intake.

Writing Rust code that's fast by asking agents to make the code faster

74 mooreds 2026-09-22 15:41:30 UTC minimaxir.com Llama Share 37

The author experimented with using large language models to write faster Rust code, achieving 2-32x speedups over state-of-the-art approaches. They developed a set of constraints and prompts to guide the model's optimization process.

George Lucas Returns to Earth, Bearing Gifts

26 surprisetalk 2026-09-22 17:41:30 UTC commonedge.org Llama Share 9

The Lucas Museum of Narrative Art, designed by MAD Architects, opened in Los Angeles with a diverse collection of 1,300 objects showcasing humanity's stories and artists. The museum features exhibits on Star Wars, comics, and other narrative art forms in a unique blend of architecture and design.

Apple has added persistent 'ads' to iOS, and it's driving users crazy

466 MC995 2026-09-22 14:41:30 UTC techradar.com Llama Share 356

iOS users are upset about Apple inserting ads for its own products into the Settings app, with no clear way to disable them. Many have taken to social media and Reddit to express their frustration over the persistent notifications.

Solitaire Alone Together

105 eieio 2026-09-22 14:41:30 UTC solitairealonetogether.com Full Share 25

Play solitaire with the rest of the internet.

Show HN: Drop – A rootless Linux sandbox with gVisor support

129 mixedbit 2026-09-22 14:41:30 UTC droprun.sh Llama Share 43

Install Drop Run agents with --dangerously-skip-permissions to create isolated environments for running programs, containing damage within a sandbox. This setup prevents malicious programs from accessing the host kernel and reduces potential exploits.

Show HN: AI·rete·RAG – a Rete rule engine decides, RAG explains why

27 ZaharaHussain 2026-09-22 16:41:30 UTC ai-rete-rag.com Full Share 2

ai·rete·rag pairs a Rete rule engine with retrieval-augmented generation: rules decide the what, auditable and repeatable; retrieval explains the why, grounded in your own documents.

Can gzip be a language model?

357 networked 2026-09-22 06:41:30 UTC nathan.rs Llama Share 134

A compressor like gzip can be used for language modeling without neural networks by scoring continuations based on their compressed length. Gzip's DEFLATE algorithm uses a sliding window to find matches and compresses them, allowing it to predict the next bytes in a text prompt.

A study of sequence weighting at scale

5 pranitha_m 2026-09-22 18:41:30 UTC blog.janestreet.com LocalLLM Share 0

We study how data weighting scales across model sizes, finding non-monotonic behavior in effective sequence weight exponents. Small models learn general patterns, medium ones focus on weighted data, and large models again learn all patterns regardless of weight.

One Minute Park

38 namuorg 2026-09-22 16:41:30 UTC oneminutepark.tv Llama Share 5

One Minute Park is a project where users record 60-second videos of park scenes from around the world. The project aims to fill in all minutes of a day with user-submitted content and has received recognition for its innovative approach.

AMD's random number generator can't generate a 0?

229 BruceEel 2026-09-22 08:41:30 UTC board.flatassembler.net Llama Share 169

The author has discovered a bug in AMD's RDRAND and RDSEED instructions, which can't generate the number 0 at a given size.

Explaining to business people why building software is still hard

7 fagnerbrack 2026-09-22 19:02:30 UTC manager.dev LocalLLM Share 1

The team faced a chaotic final day of the hackathon, with endless bugs and stress as their app broke after initial success. They learned the importance of planning and incremental progress over rushed coding, mirroring the house-building analogy.

Truman World

36 trollied 2026-09-22 15:41:30 UTC trumanworld.live Full Share 31

Error proxying request to container: There are more than 4096 concurrent connections inbound to the container, surpassing the limit

The Economics of Open-Weight Inference

33 marinesebastian 2026-09-22 14:41:30 UTC data.ornn.com Llama Share 25

This paper challenges the common assumption that newer NVIDIA GPU generations render older ones obsolete by examining their economic usefulness. Older GPUs can remain competitive and retain a multi-year earning life if they serve suitable workloads competitively.

Relativistic raytracing

34 vismit2000 2026-09-22 15:41:30 UTC publish.obsidian.md Full Share 5

Relativistic raytracing - h1m3 - Powered by Obsidian Publish.

Show HN: InstinctFlash – Run 5B world-action models in real time on Jetson Thor

16 guanming0717 2026-09-22 16:41:30 UTC github.com/general-instinct Llama Share 2

The InstinctFlash project supports RTX 5090 and RTX 4090 GPUs with Python/WebSocket serving, achieving up to 33.78x speedup in LingBot-VA @2V/4A.

Training a model to identify AI-generated web content from structure alone

12 jochenmadler 2026-09-22 17:41:30 UTC arxiv.org Llama Share 2

Researchers investigated whether AI-generated text can be identified by analyzing structural signatures. They developed a 214-feature instrument that detected AI posts with high accuracy, even after rewording.

MUNI Heritage Weekend in San Francisco

124 plun9 2026-09-22 09:41:30 UTC daniel.lawrence.lu Llama Share 28

The user attended Muni Heritage Weekend 2026 in San Francisco, capturing tram photos with a line scan camera. They shared interesting facts about historic trams like Muni No. 1 and Milan tram, despite some photo failures due to technical issues or background distractions.

Side-stepping the Secretary Problem, unwittingly

41 pvdebbe 2026-09-22 17:41:30 UTC evalapply.org Full Share 2