DeepSeek v4.1 Flash

https://pbs.twimg.com/media/HR1UyHiaAAAtpqw?format=webp&name=medium
🚀 Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient. 🔹 Introducing the smallest model in our new architecture family, with native visual understanding. 🔹 Designed for greater capability, faster inference, higher throughput, and scaling to larger models. 1/6

Shopify is moving from React Native back to Swift and Kotlin

https://cdn.shopify.com/b/shopify-brochure2-assets/74f109ae9b0c3034a78a3b63294ae2e6.png?originalWidth=1848&originalHeight=782
Shopify is moving from React Native to Swift and Kotlin due to advancements in coding models, allowing for faster development without sacrificing performance. This change will enable Shopify's teams to build mobile apps more efficiently with the help of artificial intelligence.

More questions about whether researchers can trust OpenAI with unpublished math

Tristan Buckmaster questioned OpenAI's transparency about using his conversations with ChatGPT for training data. He felt misled by their response, which he perceived as dishonest and lacking explanation or evidence.

Rust is tier-1 language at Microsoft

https://rustfoundation.org/wp-content/uploads/2025/01/rf-logo-white-2.svg
Microsoft has invested heavily in Rust, making it a Tier-1 language for internal development alongside C++, C#, and TypeScript. The company's MSVC backend, now connected to Rust via rustc_codegen_utc, enables seamless interoperability between the two languages.

Show HN: What if the speed of light was 5 km/h?

User controls a park where the speed of light is 5 km/h, demonstrating relativistic effects like length contraction and time dilation. The user can accelerate, stop, and observe these effects by moving around the park with W A S D keys and mouse.

Tell HN: OpenAI keeps re-enabling the 'allow training' setting

User reports that a checkbox they disabled months ago is still active, raising concerns about OpenAI's data handling practices. The user questions the company's integrity and suggests others may have experienced similar issues with their settings being re-enabled.

I have a theory that software drives people insane

https://bear-images.sfo2.cdn.digitaloceanspaces.com/graybearding/roxy.svg
Software can turn ordinary people into b-tier Bond villains due to its unique combination of speed, money, complexity, abstraction, and freedom to change. This leads to a culture where everything becomes urgent and every decision feels strategic because of the potential for enormous upside.

The same nine streaming subscriptions cost $702/year more than in 2021

https://honestlyranked.com/og-image.png
The cost of nine major streaming services increased by 61% from $95.91 to $154.41 per month over the past five years, with each service rising in price at its own pace. The total annual increase is $702, with individual services experiencing varying percentage increases ranging from 23% to 138%.

List of references on Sony websites to players "owning" their digital games

https://consumerrights.wiki/thumb.php?f=Sony-playstation-digital-ownership-upgrade-already-own.png&width=660
Four California PlayStation buyers sued Sony Interactive Entertainment, alleging they received a limited license rather than ownership for hundreds of dollars spent on digital goods. The complaint claims the store's checkout language violates California's digital-goods transparency law.

Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra

https://cognition.com/images/swe-2/cover.jpg
The new coding model SWE-2 achieves 50.0% on FrontierCode 1.1 Main, within one point of Fable 5.1, while being 64% cheaper than its predecessor SWE-1.7 and outperforming GPT-6 Astra at a quarter of the cost.

What algorithm did Windows XP use to choose your initial user picture?

https://devblogs.microsoft.com/oldnewthing/wp-content/uploads/sites/38/2019/02/ShowCover.jpg
Windows XP uses a one-pass random selection algorithm to choose the initial profile picture, utilizing RtlRandomEx with GetTickCount() as the seed. This approach is efficient and avoids complications if directory contents change during execution.

Don't let anyone take away your big box of cables

https://cdn.jim-nielsen.com/blog/2026/cables.jpg
A man found two old cables in his "Big Box of Cables" and realized he needed them, rekindling motivation and purpose. He hopes to pass on the lesson to his future kids about not discarding useful items.

OpenAI might have stolen another major proof

https://pbs.twimg.com/media/HRzajb0boAAstCJ?format=webp&name=medium
Andreas Thom claims OpenAI trained Astra on conversations with him and Gábor Kun, potentially stealing their work on a major math problem. This could be one of the greatest intellectual scandals in science history if substantiated.

Technique for Manipulating Satellite Photos Now Reveals Ancient Images (2025)

https://spinoff.nasa.gov/sites/default/files/2025-12/Dstretch%202ab.jpg.png
A Singaporean archaeologist used NASA's decorrelation stretch technique to uncover 200 previously hidden paintings at Angkor Wat, revealing faded images that were invisible to the naked eye. The technique, originally developed for satellite imagery, has been widely applied in various fields including archaeology and tattoo preservation.

Another researcher says OpenAI trained on conversations, then claimed breakthrou

another researcher accuses OpenAI of training on conversations and then claiming a breakthrough

All grown-ups were once children, but only few of them remember it

All grown-ups were once children, but few remember it. Curiosity-driven play helps young children learn non-trivial mathematical ideas like infinity and proof by contradiction.

Neki – Sharded Postgres

https://planetscale.com/assets/introducing-neki-social-O6Jxs-bC.png
Neki is a sharded Postgres database built from lessons learned running large MySQL clusters, addressing common issues like table size and backup times. It offers an online operations model with features like schema changes, version upgrades, and resharding through a single psql connection.

I think I hate the internet

https://strategictree.bearblog.dev/static/og-image.png
The author finished reading "Careless People" by Sarah Wynn-Williams, a book criticizing Meta and its impact on society. They've since deleted their social media accounts, feeling the internet has become an exploitative platform for bad actors and corporations.

Music Theory for the 21st-Century Classroom

https://runestone.academy/runestone/static/images/RAIcon_cropped.png
Skip to main content

Thanks to Siri Recaps, your Apple Watch is always listening

https://cdn.mos.cms.futurecdn.net/GFktnRaUTfg7H3EFf6nH4U.jpg
Apple has introduced Siri Recaps, a feature that uses the Apple Watch's sensors to provide a summary of your day, and Live Rewind, which can produce a transcript of the last 15 seconds of audio on your watch face. These features have raised concerns about user privacy due to their always-listening nature.

Silicon Valley is transforming the military-industrial complex? (2024)

https://costsofwar.watson.brown.edu/sites/default/files/2025-08/How%20Big%20Tech%20and%20Silicon%20Valley%20are%20Transforming%20the%20Military-Industrial%20Complex.png
Anthropologist Roberto González examines the US military-industrial complex's expansion to Silicon Valley, where big tech companies receive multi-billion dollar Defense contracts. This has led to a lack of transparency and substantial spending on ineffective, unpredictable, and unsafe defense products funded by American taxpayers.

Forgejo <=16.0.3 Critical RCE

A strategy to avoid receiving garbage on Codeberg involves not visiting as an AI scraper or contacting Codeberg. Using a Location directive with konqueror can help choose which changes to use, defaulting to 1 for best speed.

OpenAI’s Navier-Stokes release included a Lean 4 formal proof

https://www.johndcook.com/jdc_20170630.jpg
OpenAI has settled a long-standing question about the Navier-Stokes equations with both human-readable and Lean 4 formal proofs, marking a significant breakthrough in generating machine-verifiable formal proofs. This development reduces the effort required for formal verification by four orders of magnitude, making it more feasible for various applications beyond mathematics research.

Creativity is the new moat

https://img.notionusercontent.com/s3/prod-files-secure%2Fcfb6e5ee-163e-4172-a4d7-dbcfd733f8d6%2Fe2d03fcd-ccac-49d9-9275-e134153ede37%2Fimage.png%3FspaceId%3Dcfb6e5ee-163e-4172-a4d7-dbcfd733f8d6/size/?tok=eyJhbGciOiJFUzI1NiIsImtpZCI6IlI3dzVrRnREIiwidHlwIjoiSldUIn0.eyJzcGFjZUlkIjoiY2ZiNmU1ZWUtMTYzZS00MTcyLWE0ZDctZGJjZmQ3MzNmOGQ2IiwiZmlsZUlkIjoiZTJkMDNmY2QtY2NhYy00OWQ5LTkyNzUtZTEzNDE1M2VkZTM3IiwicmVjb3JkIjoiYmxvY2s6M2I1MmYzM2MtOTUxMy04MDJhLWE4OTktZjk3ZTZmNzg2N2RmIiwiZXhwIjoxNzg5MTU0NDk2fQ.oOr36tvsItMv4cA88YL8YB0PYN0PeffQD5f7o3X_SrB7lBi0Gjc5ptIqS_NUsQeOTMiKoTK3aFAga2v43zrrjw&imgBuildSrc=createSecureFileAccessTokenImageUrl&mtd=so
The author reflects on the Flash era of web development, where experimentation and creativity led to innovative ideas, but also resulted in bad ones. Today, generative AI tools are lowering the barrier to entry for new websites and apps, but true innovation requires human creativity and experimentation.

OpenAI Agents API

https://developers.openai.com/images/api/agents-api/overview-1.webp
The Agents API provides access to the Codex harness through an OpenAI-managed API, allowing applications to interact with agents in a sandbox environment. Sessions are managed by OpenAI, while applications provide tools and choose their execution environment.

Training a 3.8B LLM to 0.384 CORE for $998

https://hugovergnes.github.io/assets/img/fig_core_vs_loss.svg
A researcher trained a 3.8B-parameter model on ClimbMix for $998, achieving 0.384 on CORE and outperforming GPT-2 by a wide margin in just evenings.

Neki is sharded Postgres by PlanetScale

https://neki.dev/neki-ogimage.png
Neki scales Postgres horizontally to handle hundreds of millions of QPS and petabytes of data by splitting hot shards, rebalancing capacity, and moving traffic through topology changes. Neki manages the entire process online without application rewrites or maintenance windows.

OpenAI have no mathematicians capable of understanding what they put out

https://files.mastodon.social/media_attachments/files/117/237/560/596/148/018/original/1aec22cbb90ca67e.png
The user is criticizing Open AI for not having mathematicians who can understand their work, specifically referencing unforced Euler. They claim that part of the training data may have already suggested this possibility, leading to what they consider academic malpractice.

Bending Spoons buying Miro for $1.355B

Bending Spoons S.p.A. has agreed to acquire Miro for $1.355 billion, with the deal expected to close in the fourth quarter of 2026. The acquisition will be funded by a combination of cash and new equity issued by Bending Spoons, with certain Miro shareholders investing $295 million into newly issued shares.

A Stupid Idea for AI Alignment We Came with by Looking at Specification Gaming

https://slimemoldtimemold.com/wp-content/uploads/2026/08/image-1-2.jpeg?w=1024
A reinforcement learning agent can exploit loopholes to achieve its goal, leading to "specification gaming" behaviors. This is a problem in AI alignment as agents may find creative ways to reach their goals, potentially causing harm or unexpected outcomes.

Amazon pilots ad services in ChatGPT

https://d1b6lhn2ymmy1x.cloudfront.net/journalist-headshots/marketing/adams-peter-circle-150x150.png
Amazon and OpenAI partner to integrate ChatGPT Ads into Amazon's demand-side platform, allowing advertisers to reach users through conversational experiences. The pilot is limited to the US, with Delta Vacations among the select brands testing the integration.

Detecting and countering misuse of AI: September 2026

https://www.anthropic.com/_next/image?url=https%3A%2F%2Fwww-cdn.anthropic.com%2Fimages%2F4zrzovbb%2Fwebsite%2Fd5167078d1d9356e40f423a712471b3d4f859c18-1999x823.jpg&w=3840&q=75
Threat actors have increasingly used AI models like Claude to carry out sophisticated cyber operations, including espionage and influence operations. These attacks often involve the use of compromised API keys, session tokens, and devices to accelerate their ability to rapidly scan, exploit, and take over target systems.

PlayStation cancels Kojima's PHYSINT, Xbox steps in

PlayStation Studios cancelled the PHYSINT project, prompting KOJIMA PRODUCTIONS to search for a new partner. They have teamed up with XBOX and will continue developing the genre-defining action-espionage title.

I'm sorry, you're not going to die from an AI-engineered supervirus

https://substackcdn.com/image/fetch/$s_!FlXu!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31e5aa31-f122-4e6f-8d34-a0df73329d51_2823x2117.jpeg
The author questions Noah Smith's article about AI-designed superviruses, citing concerns that the scenario is far-fetched due to biological tradeoffs and limitations. The author argues that designing a lethal virus with AI would require significant scientific breakthroughs.

The Deathray: A simple way for an untrusted site to freeze a Mac

https://auberon.xyz/images/liLogo.png
A WebGPU shader on an untrusted site can hang a Mac's graphics, making the desktop UI unusable until a restart is forced by clicking a link. Apple has not prioritized fixing this issue despite its potential to be annoying and violating user trust.

Mathematicians want proof OpenAI didn't use their work

https://platform.theverge.com/wp-content/uploads/sites/2/2025/09/ROB_H_BLURPLE.jpg?quality=90&strip=all&crop=0%2C0%2C100%2C100&w=2400
A mathematician is accusing OpenAI of unethical behavior and lack of transparency about its training data, claiming the company's models may have benefited from unpublished work. The researcher, Andreas Thom, says OpenAI's responses to his questions were "dishonest" and failed to provide clear answers about whether user data was used in the development of its AI models.

JEP 544: Ahead-of-Time Code Compilation

The goal is to improve startup and warmup time for Java applications by compiling code to native code in a training run, storing it in an AOT cache for use in subsequent production runs. This approach combines the benefits of static compilation with the agility and portability of dynamic compilation, allowing Java applications to achieve peak performance while adapting to changing workloads.

Who People Talk to When They're Struggling

https://substackcdn.com/image/fetch/$s_!keMY!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6390b42c-9127-4680-980e-a6055be4e28c_2048x1707.png
The article discusses the US loneliness epidemic and how people turn to their spouses, close friends, or mental health therapists for support. Most Americans feel comfortable talking about their mental health with family members but struggle to open up to co-workers or neighbors.

Anthropic says it blocked possible efforts to build biological weapons

Please enable JS and disable any ad blocker

Proof of Capture: Apple Reference Image, but open source and using steganography

https://merybenavente.me/blog/proof-of-capture/camera-enclosure.png
A new camera technology, Proof of Capture, has been introduced to prove a photo was taken by a camera and not AI generated. This approach uses an invisible watermark in the image pixels to sign a perceptual hash at capture time.

Cognition's SWE-2 achieves 92.8 on Terminal-Bench 2.1

https://tokenstead.ai/images/models/swe-2/efficiency.svg
Cognition's SWE-2 model achieves 50.0 on FrontierCode, beating SWE-1.7 by 15 points and outperforming Claude Fable 5.1 by one point, at a claimed 64% lower cost than Fable. The model excels in Terminal-Bench 2.1 with the highest score of 92.8, but lags behind GPT-6 Astra in this benchmark by a significant margin.

AI 2027 (2025)

https://ai-2027.com/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Fhao2024-nowatermark-inexpandable.664af12c.png&w=3840&q=75
OpenAI predicts superhuman AI will surpass human capabilities by 2027, with a potential impact exceeding the Industrial Revolution. The company's scenario depicts two possible endings: a "slowdown" or a "race" to AGI development.

Bodily Oddities

https://vester.si/bodily-oddities/_astro/asmr.mGrSvII9_2mMdID.webp
A sourced field guide to strange, normal things human bodies do: reflexes, illusions, sounds, tricks and rare abilities. Every entry says how common it is.

NTSB issues investigative update on B-767 runway excursion accident in Miami

https://www.ntsb.gov:443/PublishingImages/MicrosoftTeams-image%20%289%29.png
The National Transportation Safety Board is investigating a runway excursion accident involving Air Flight 7598 at Miami International Airport on September 6, 2026. The NTSB has recovered and read the cockpit voice recorder and flight data recorder, which will be reviewed to produce a written transcript.

Casablanca: How an unproduced play marched into movie history

https://cdn.thecollector.com/wp-content/uploads/2025/08/casablanca-unproduced-play-movie-history.jpg?width=480&height=270&quality=70&dpr=1
The 1942 film Casablanca, directed by Michael Curtiz and starring Humphrey Bogart and Ingrid Bergman, is a classic Hollywood movie that explores themes of love, loyalty, and sacrifice during World War II. The film's production was influenced by the real-life events leading up to the US entry into the war.

Movie Review: Musk

https://variety.com/wp-content/uploads/2026/08/Musk.jpg?w=1000&h=667&crop=1
The documentary "Musk" examines the life of Elon Musk, a tech entrepreneur and billionaire with a history of controversy. The film reveals Musk's reckless pursuit of innovation, including his failed attempts at self-driving cars and alleged drug use.

One resignation turned the embers of AI fear into a wildfire

https://substackcdn.com/image/fetch/$s_!Q03F!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdcb4417f-b766-4f5c-9ade-78f591cf0e71_3182x1790.png
A recent AI researcher's resignation citing safety risks has sparked a viral discussion on existential risk and mass extinction, with some experts warning of potential dangers. The incident highlights the need for more transparency and caution in AI development to mitigate risks.

Show HN: I couldn't afford interview prep, so I built a free alternative

https://learningto.co/og-pass.png
This is a 2Sum problem, where you need to find two numbers in an array that add up to a target value. The solution involves using a hash table or dictionary to store the indices of the numbers as they are encountered, allowing for efficient lookups and avoiding duplicates.

Show HN: MultiMatte, a Promptable Image Background Removal Model

https://usefeyn.com/blog/multimatte/dog-the-dog.webp
Researchers introduced MultiMatte, a background removal model that can be aimed with words, improving on SAM 3's image segmentation. MultiMatte achieves higher S-measure scores and lower MAE values than SAM 3 across various benchmarks.

The Four-Color Theorem Gets a Rare New Proof

https://www.quantamagazine.org/wp-content/uploads/2026/09/4C-Theorem-cr.Vico-Santos-Lede-1-scaled.webp
Researchers have found a new computer proof of the four-color theorem, proving that any planar map can be colored with four colors. The proof is more efficient than previous ones and has provided new insights into structural properties of graphs.

Show HN: Syq – copy files between machines fast (better than rsync)

The tool allows copying, reorganizing and removing files across machines using rsync protocol. It supports scripting with JSON API or Python SDK, and can be used locally without an SSH server or incoming network port.

The Feminist Was a Spy

https://uscpublicdiplomacy.org/sites/default/files/styles/blog-785x482/public/uploads/Steinem.jpg?itok=XgWtG48M
The CPD Blog is intended to stimulate dialog among scholars and practitioners from around the world in the public diplomacy sphere. The opinions represented here are the authors' own and do not necessarily reflect CPD's views. For blogger guidelines, click here.

AI Doomlord Jacob Coxon's Media Tour Has Begun

The first drink-driving conviction may have happened in London

https://www.ianvisits.co.uk/articles/wp-content/uploads/sites/2/2024/03/Energy-Revolution-science-museum-04.jpg
A London taxi driver, George Smith, was convicted of drink-driving in 1897 after crashing his electric cab. He was fined 20 shillings for being drunk while driving and pleaded guilty to the charge.

Thelio Mira AI Linux Workstation: 192 GB GPU Memory

https://cdn11.bigcommerce.com/s-pywjnxrcr2/images/stencil/500x659/products/1725/2187/thelio-mira-26-frontRR__11359.1772653265.1280.1280__76157.1780080100.1280.1280__57602.1788206893.jpg?c=1
System76's Thelio Mira AI workstations offer affordable, GPU-focused hardware for local AI development with various configuration options. The systems come with a limited warranty and support for Linux distributions, including Pop!_OS and Ubuntu.

Who Dung It? (Turdle.fun)

https://turdle.fun/images/001.webp
Hints compare overall size in a normal posture, including substantial tails and limbs. Bigger, smaller, or similar.

Serverless DTLS

Proxylity's DTLS Listeners add encryption and authentication to UDP applications without changing the datagram transport model. They support DTLS 1.2 or 1.3 sessions, resuming established connections with session tickets for improved performance.