How OpenAI delivers low-latency voice AI at scale

126 Sean-Der 2026-05-04 20:11:32 UTC openai.com Llama Share 59

OpenAI rearchitected their WebRTC stack to improve real-time AI interactions by reducing latency and preserving standard WebRTC behavior for clients. They achieved this by introducing a relay service that routes packets to the correct transceiver, allowing for deterministic first-packet routing and a small public UDP footprint.

I am worried about Bun

312 remote-dev 2026-05-04 17:11:32 UTC wwj.dev Llama Share 214

The author is concerned about Bun's future after Anthropic acquired it, citing Claude Code's decline in quality and restrictions. They're moving away from Bun for package management and recommend pnpm instead.

Securing a DoD contractor: Finding a multi-tenant authorization vulnerability

136 bearsyankees 2026-05-04 18:11:32 UTC strix.ai Llama Share 61

A company, Schemata, handling sensitive military data had no API authorization, allowing a low-privilege user to access users, organizations, and confidential training materials. The issue was fixed after 150 days of disclosure attempts, highlighting the importance of open disclosure channels and robust authorization testing.

Talking to strangers at the gym

967 thitran 2026-05-04 12:11:32 UTC thienantran.com Llama Share 477

The user, who was struggling to make friends after college, approached 30 people at the gym over a month, resulting in some meaningful connections and friendships. Through this experience, the user became more resilient and confident in initiating conversations with strangers, eventually forming connections with people who share similar interests and hobbies.

Formatting a 25M-line codebase overnight

58 r00k 2026-05-04 21:11:32 UTC stripe.dev Full Share 31

This is the story of how our Developer Productivity team extended and rolled out rubyfmt, our Rust-based zero-config, ultra-fast autoformatter across the world's largest Ruby codebase.

GameStop makes $55.5B takeover offer for eBay

590 n1b0m 2026-05-04 10:11:32 UTC bbc.co.uk Llama Share 543

GameStop made a $55.5bn offer to buy eBay at $125 a share, with CEO Ryan Cohen planning $2bn in cost savings. The deal would link GameStop with eBay, lifting GameStop's valuation and giving eBay a national network for its business operations.

Microsoft Edge stores all passwords in memory in clear text, even when unused

294 cft 2026-05-04 19:11:32 UTC twitter.com/l1v1ng0ffth3l4n Full Share 113

Something went wrong, but don’t fret — let’s give it another shot. Some privacy related extensions may cause issues on x.com. Please disable them and try again.

Does Employment Slow Cognitive Decline? Evidence from Labor Market Shocks

160 littlexsparkee 2026-05-04 16:11:32 UTC nber.org Llama Share 149

Researchers Noah Arman Kouchekinia, David Neumark, and Tim A. Bruckner received grants for their study on employment and cognitive decline. The study was supported by the Alfred P. Sloan Foundation and the Lynde and Harry Bradley Foundation.

Redis array: short story of a long development process

196 antirez 2026-05-04 15:11:32 UTC antirez.com Llama Share 74

The user implemented a new Array data type for Redis, taking four months with AI assistance, which helped with specification, implementation, and testing. AI provided a safety net for complex tasks, allowing the user to venture into higher complexity and achieve high-quality system programming.

US healthcare marketplaces shared citizenship and race data with ad tech giants

356 ZeidJ 2026-05-04 18:11:32 UTC techcrunch.com Llama Share 124

Pixel trackers on websites collect personal info, including health data, often for advertising. Government websites sharing health data with tech companies affects millions of Americans.

How Monero’s proof of work works

207 alcazar 2026-05-04 15:11:32 UTC blog.alcazarsec.com Llama Share 163

Monero's RandomX proof of work makes mining expensive for custom chips by using CPU strengths like dynamic code and memory access. It generates random programs and uses memory-hard functions to make mining look like general-purpose computing.

UK Fuel Price Intelligence – Market analytics from reporting stations

142 theazureguy 2026-05-04 16:11:32 UTC fuelinsight.co.uk Llama Share 70

This site provides independent UK fuel price data sourced from the UK Fuel Finder scheme, with no influence on prices. It stores and analyses publicly mandated data, using Google Analytics to understand site usage.

Let's talk about LLMs

88 cdrnsf 2026-05-04 18:11:32 UTC b-list.org Llama Share 58

The author discusses the potential impact of large language models (LLMs) on software development, arguing that they may not offer the promised order-of-magnitude improvements in productivity.

Welcome to Gas City

19 teruakohatu 2026-05-04 21:21:32 UTC steve-yegge.medium.com Llama Share 30

Gas City is a new SDK for building dark factories, allowing users to deploy teams of collaborating agents in any topology. It provides a foundation for building custom orchestrators backed by Git, with features like declarative deploys, audit trails, and version history.

Pomiferous: The most extensive apples (pommes) database

87 Ariarule 2026-05-04 15:11:32 UTC pomiferous.com Llama Share 37

Several apple varieties are described including Sekai Ichi, French cider apple, and Pink Lady. These apples have unique characteristics such as flavor, texture, and origin.

Pulitzer Prize Winners 2026

5 brightbeige 2026-05-04 21:16:32 UTC pulitzer.org Prefix Share 1

For piercing the veil of secrecy around the Trump administration's chaotic overhaul of federal agencies and chronicling in rich detail the human impacts of the cuts and the consequences for the country. For its coverage of a shooting at a back-to-school Mass at a Catholic school that left two children dead and 17 wounded, powerful stories marked by thoroughness and compassion. For deeply ...

1966 Ford Mustang Converted into a Tesla with Working 'Full Self-Driving'

83 Brajeshwar 2026-05-04 16:11:32 UTC electrek.co Llama Share 64

A Sacramento auto parts shop owner spent $40,000 and two years converting a 1966 Ford Mustang into a Tesla with a Model 3 drivetrain and Full Self-Driving capabilities. The project demonstrates Tesla's technology is portable and adaptable, despite licensing struggles with other automakers.

Stop big tech from making users behave in ways they don't want to

183 andsoitis 2026-05-04 18:11:32 UTC economist.com Llama Share 120

A 2019 study found teens can't switch off from Instagram, leading to a landmark ruling against Meta and YouTube for designing addictive products. The ruling's implications are being digested as the world adjusts to its impact.

Heat pump sales rise across Europe

163 doener 2026-05-04 18:11:32 UTC pv-magazine.com Llama Share 87

Residential heat pump sales rose 17% in 11 European countries in Q1 2026 due to high gas and oil prices. France, Germany, and Poland saw 25% growth, while Austria's sales fell 30% due to lack of government subsidies.

Sierra Raises $950M at $15B Valuation

64 doppp 2026-05-04 16:11:32 UTC sierra.ai Llama Share 90

Sierra raises $950 million at over $15 billion valuation to invest in AI customer experience. The company now has over $1 billion to deploy AI agents for businesses, transforming customer interactions and experiences.

Show HN: nfsdiag – A NFS diagnostic application

25 lsferreira42 2026-05-04 19:11:32 UTC github.com/lsferreira42 Llama Share 1

nfs-doctor is a command line tool to help debug NFS servers from the client side. It checks network, rpcbind, NFS versions, and other common issues.

The Visible Zorker: Zork 3

30 zarlez 2026-05-04 17:11:32 UTC eblong.com Full Share 1

A software exhibit that allows you to explore Zork and also Zork's source code at the same time.

Transformers Are Inherently Succinct

11 bearseascape 2026-05-04 20:11:32 UTC arxiv.org Llama Share 4

Transformers are highly expressive in describing concepts, outperforming finite automata and LTL formulas in succinctness. Verifying transformer properties is intractable due to their expressivity.

A little comparison between R and Kap

11 tosh 2026-05-04 20:11:32 UTC blog.dhsdevelopments.com Llama Share 0

The user compared Pandas and R, finding Kap solutions shorter but R with nice defaults. They implemented examples from a blog post in Kap, highlighting differences between the languages.

Frizbee is a tool you may throw a tag at and it comes back with a checksum

5 mooreds 2026-05-04 21:11:32 UTC github.com/stacklok Prefix Share 0

Frizbee is a tool you may throw a tag at and it comes back with a checksum. It's a command-line tool designed to provide checksums for GitHub Actions and container images based on tags. # Using Go go install github.com/stacklok/frizbee@v0.1.2 # add the path to your .bashrc export PATH=$PATH:$(go env GOPATH)/bin # Using Homebrew brew install frizbee # Using winget winget install ...

“Kitten Space Agency”, a Spiritual Successor to “Kerbal Space Program” (2025)

108 Tomte 2026-05-04 18:11:32 UTC space.com Llama Share 37

Stefan Moluf, a former SpaceX engineer, is working on Kitten Space Agency, a spiritual successor to Kerbal Space Program, with a focus on a solid technical foundation and modding capabilities. The game aims to be both authentic and accessible, with a free "pay-what-you-want" distribution model.

Offenders sentenced up to 10 years for spying on TSMC

87 ironyman 2026-05-04 18:11:32 UTC taipeitimes.com Llama Share 5

A former TSMC engineer was sentenced to 10 years in prison for leaking trade secrets to Tokyo Electron Taiwan. Three other TSMC employees and Tokyo Electron Taiwan were also sentenced for their roles in the trade secret leak.

Newton's law of gravity passes its biggest test

117 pseudolus 2026-05-04 13:11:32 UTC science.org Llama Share 104

Researchers confirmed Newton's law of gravity on the largest scales using galaxy clusters. The study tightens the vise on alternative theory MOND, which tries to explain dark matter without gravity.

Why are neural networks and cryptographic ciphers so similar? (2025)

118 jxmorris12 2026-05-04 10:11:32 UTC reiner.org Llama Share 34

Language models and cryptography share similarities in their algorithms, including sequential absorbing, parallel processing, and repeated mixing layers. Both fields prioritize low-level hardware performance and emphasize parallelism, resulting in similar solutions despite different problem statements.

Using “underdrawings” for accurate text and numbers

353 samcollins 2026-05-04 00:11:32 UTC samcollins.blog Llama Share 126

User shares a technique to generate images with Gemini 3.0 Pro and ChatGPT-Images-2 by combining deterministic underdrawing with generative painting. This method still outperforms the models on certain tasks.