Published on June 14, 2026 at 18:01 CEST (UTC+2)
No, everyone is not using AI for everything (89 points by yegg)
No, everyone is not using AI for everything
The article challenges the narrative that AI is universally adopted, arguing that most people who have tried AI use it only occasionally. It cites Gallup data showing Gen Z AI adoption has stalled, with a large percentage still using AI rarely or not at all. The author notes that “everyone is using AI for everything” is actually “some people are using AI for some things,” and highlights persistent anxiety about AI even among younger users.
The Birth and Death of JavaScript (2014) (124 points by subset)
The Birth and Death of JavaScript (2014)
This is a talk by Gary Bernhardt from PyCon 2014 that presents a fictional/humorous history of JavaScript from 1995 to 2035. It covers the language’s flaws and its massive positive impact on the industry, and speculates about future developments in programming. The talk is not pro- or anti-JavaScript, but offers a thought-provoking look at how languages evolve and eventually decline.
Firewood Splitting Simulator (242 points by memalign)
Firewood Splitting Simulator
A simple interactive web toy where users can drag to rotate a log and click to split it. It loads assets and provides a satisfying simulation of splitting firewood. The page is part of a larger collection of “screen toys” and is purely for entertainment, with no deeper technical or AI relevance.
Measles surge in Utah sparks fears US could undo decades of progress (39 points by Bender)
Measles surge in Utah sparks fears US could undo decades of progress
The article reports a significant outbreak of measles in Utah, raising concerns that the U.S. might lose its official “measles elimination” status. It discusses how declining vaccination rates and misinformation have contributed to the surge, jeopardizing decades of public health achievements. The piece is a warning about the importance of maintaining high immunization coverage.
Lisp's Influence on Ruby (122 points by tacoda)
Lisp's Influence on Ruby
The content preview is not available, but the title indicates an exploration of how Lisp’s design principles—such as code-as-data, macros, and functional programming—shaped the Ruby programming language. The article likely discusses similarities like blocks, closures, and metaprogramming features in Ruby that trace back to Lisp.
Caddy compatibility for zeroserve: 3x throughput and 70% lower latency (54 points by losfair)
Caddy compatibility for zeroserve: 3x throughput and 70% lower latency
zeroserve is a high-performance HTTPS server that compiles Caddy configuration files (Caddyfiles) to eBPF and then to native machine code, running in an io_uring event loop. Benchmarks show zeroserve achieving ~3x the throughput and ~70% lower latency compared to standard Caddy, while using less memory. The post demonstrates how eBPF can be used to dramatically accelerate web server performance with full Turing-complete extensibility.
FarOutCompany (40 points by bookofjoe)
FarOutCompany
This is a visual archive of counterculture imagery from the 1960s and 1970s, featuring photographs and ephemera from communes, head shops, underground newspapers, and music festivals. The site serves as a nostalgic collection of hippie-era visual culture, with categories like “Macramé Park” and “Wheeler’s Ranch.” It is a historical/artistic resource, not a technical piece.
The only scalable delete in Postgres is DROP TABLE (34 points by hollylawly)
The only scalable delete in Postgres is DROP TABLE
The post explains why large DELETEs in PostgreSQL are problematic due to MVCC overhead, write amplification, and lack of immediate space reclamation. It recommends designing schemas so that bulk data removal can be done via DROP TABLE or TRUNCATE, such as using table partitioning and dropping entire partitions. The article provides practical advice for scaling deletion-heavy workloads in Postgres.
Perlisisms (13 points by tosh)
Perlisisms
A collection of epigrams on programming by Alan Perlis, covering topics like the nature of constants vs. variables, syntactic sugar, recursion, and the inevitability of software decay. These aphorisms offer timeless wisdom about software design, complexity, and the human aspects of coding. They are often cited in programming culture.
Rio de Janeiro's city government model Rio3.5 beats Qwen3.7 in recent benchmarks (65 points by lucasfcosta)
Rio de Janeiro's city government model Rio3.5 beats Qwen3.7 in recent benchmarks
A tweet highlights that Rio3.5, a 397-billion-parameter AI model developed by the municipal IT company of Rio de Janeiro, has outperformed Alibaba’s Qwen3.7 in certain benchmarks. The model is open-source and demonstrates how even local government entities can produce competitive frontier AI models, challenging the dominance of big tech. It also notes the rise of other models like Minimax M3.
AI adoption is not universal; hype does not equal reality
Despite widespread media narratives, a significant portion of the population, including Gen Z, uses AI only occasionally or not at all. Anxiety about AI remains high, and adoption has stalled. This suggests that AI companies should focus on building trust, addressing privacy concerns, and creating applications that genuinely solve everyday problems rather than assuming universal uptake.
Citizen/regional AI models are emerging as serious competitors
Rio de Janeiro’s city government producing a model that beats a major Chinese frontier model (Qwen3.7) signals a democratization of AI development. Local governments, universities, and smaller organizations can now train large-scale models with relatively modest budgets. This trend could fragment the AI landscape and reduce the dominance of a few tech giants, fostering innovation and local adaptation.
Open-source vs. proprietary is a key axis of competition
The tweet explicitly ties Qwen3.7’s fading relevance to its “proprietary stance.” Meanwhile, Rio3.5 (open-source) and Minimax M3 are gaining attention. The community increasingly values open models for reproducibility, customization, and trust. Developers and enterprises should prioritize open-weight models to avoid vendor lock-in and benefit from collaborative improvements.
Performance optimization techniques (e.g., eBPF, io_uring) are critical for AI infrastructure
While not directly about AI models, zeroserve’s use of eBPF and JIT compilation to achieve 3x throughput gains highlights how low-level systems engineering can accelerate web services. This is relevant for deploying AI model serving endpoints, where latency and throughput matter. AI infrastructure should adopt similar techniques to reduce inference cost and improve user experience.
Benchmark chasing can be misleading
The Rio3.5 model outperforming Qwen3.7 on “recent benchmarks” may be a snapshot that doesn’t reflect real-world performance across diverse tasks. As more small players enter the arena, the community must rely on standardized, multi-dimensional evaluations (e.g., MMLU, HumanEval, and domain-specific tests) rather than cherry-picked results. Developers should demand transparency in benchmark methodology.
AI model sizes continue to grow, but efficiency matters more
Rio3.5 has 397 billion parameters, placing it in the same league as GPT-4 class models. However, the real competitive edge is not just size but how well a model performs relative to its compute and energy cost. Expect a push toward mixture-of-experts (MoE), quantization, and pruning to make large models deployable on limited hardware.
Government involvement in AI development is increasing
A municipal IT company creating a frontier model shows that even local governments are investing heavily in AI R&D. This could lead to more regulation-tailored models, better alignment with public sector needs (e.g., language, culture), and potential for data sovereignty. However, it also raises questions about accountability and the risk of biased or poorly tested models being deployed in public services.
Analysis generated by deepseek-reasoner