Published on June 17, 2026 at 18:01 CEST (UTC+2)
Epic Games announces Lore version control system (232 points by regnerba)
Epic Games announces Lore version control system
Epic Games has released Lore, an open-source, next-generation version control system designed for projects that combine code with large binary assets, such as games and entertainment. Lore is centralized and content-addressed, using Merkle trees and an immutable revision chain to ensure verifiable, tamper-evident history. It supports on-demand scalability, free branching, and efficient sparse/on-demand data hydration. A full-surface API in multiple languages (C/C++, C#, Rust, Go, Python, JavaScript) makes it extensible for both developers and artists.
Want your images back? Sure... That'll be $5! (405 points by lutr)
Want your images back? Sure... That'll be $5!
The article recounts a personal experience with Photobucket, an old image-hosting service that now demands a $5 fee for users to access or download their own previously uploaded images. The author criticizes the company for paywalling user content that was originally offered for free, calling it corporate greed. The piece highlights a broader trend of users losing control over their own data on centralized platforms post-acquisition or monetization shifts.
Volkswagen started blocking GrapheneOS users (48 points by microtonal)
Volkswagen started blocking GrapheneOS users
A discussion on the GrapheneOS forum reports that the Volkswagen (myVW) app has stopped working on GrapheneOS, a privacy-focused Android variant. Users share workarounds, including reinstalling the app from the Play Store and updating GrapheneOS, while others warn that modern car apps can exfiltrate extensive telemetry even when the phone is connected. The issue reflects growing tension between app compatibility and user-chosen security/privacy hardening.
GLM-5.2 is the new leading open weights model on Artificial Analysis (510 points by himata4113)
GLM-5.2 is the new leading open weights model on Artificial Analysis
Z.ai’s GLM-5.2, a 744B total / 40B active parameters open-weights model, has topped the Artificial Analysis Intelligence Index v4.1 with a score of 51, outperforming MiniMax-M3 and DeepSeek V4 Pro. It shows substantial gains in scientific reasoning, coding, and general benchmarks, while maintaining the same pricing as its predecessor. The model leads on GDPval-AA v2 and is competitive with proprietary models, marking a milestone for open-weight AI.
French physicist and media star loses doctorate after plagiarism investigation (20 points by bookofjoe)
French physicist and media star loses doctorate after plagiarism investigation
A well-known French physicist and media personality has had his doctorate revoked following a formal plagiarism investigation. The Science.org article details the findings of copied content in his thesis, leading to the university’s decision. This case underscores the ongoing scrutiny of academic integrity, even for high-profile scientists.
Sixty percent of US consumers say 'AI' in brand messaging is a turnoff (605 points by thm)
Sixty percent of US consumers say 'AI' in brand messaging is a turnoff
WordPress VIP’s “Future of the Web 2026” report reveals that 60% of U.S. consumers are put off by brands that explicitly mention “AI” in their messaging. Other findings: 74% feel the internet is less human than a decade ago, and 61% cannot name a single brand that uses AI well. “Bot fatigue” sets in after an average of 40 minutes, indicating that audiences are wary of inauthentic automation.
RFC 10008: The new HTTP Query Method (173 points by schappim)
RFC 10008: The new HTTP Query Method
This informational RFC proposes a new HTTP QUERY method that allows clients to send a query body to a server and receive a response, distinct from GET and POST. It defines semantics for content negotiation via an Accept-Query header, caching, conditional requests, and redirection. The method is intended to improve the semantics of read-only queries that require a payload, such as complex SPARQL or GraphQL requests, while keeping idempotent behavior.
MicroUI – A tiny, portable, immediate-mode UI library written in ANSI C (82 points by peter_d_sherman)
MicroUI – A tiny, portable, immediate-mode UI library written in ANSI C
MicroUI is a lightweight (~1100 lines of ANSI C) immediate-mode GUI library that works within a fixed-size memory region with no dynamic allocation. It provides built-in controls like windows, buttons, sliders, and textboxes, and can integrate with any rendering system that draws rectangles and text. The project is designed for simplicity, portability, and ease of custom control creation, ideal for embedded or resource-constrained applications.
Why stdx is not on crates.io (30 points by Keyb0ardWarri0r)
Why stdx is not on crates.io
The article explains why the Rust library stdx (an extension of the standard library) is not published on crates.io. Reasons include potential naming conflicts with Rust’s own std namespace, concerns about encouraging poor code patterns, and the maintainer’s preference for keeping it as a personal, non-official reference. It also touches on broader Rust ecosystem norms around library naming and ownership.
Hacker News but for Independent Blogs (331 points by headalgorithm)
Hacker News but for Independent Blogs
Bubbles.town is a community-run aggregator that surfaces posts from over 5,000 independent, personal blogs. Posts are ranked by votes and freshness, similar to Hacker News, but exclusively featuring non-commercial, small-web content. The site offers a curated feed of authentic writing on topics like tech, life, and opinions, encouraging discovery of individual voices rather than mainstream media.
Open-weight models are closing the gap with proprietary leaders
GLM-5.2 achieving a score of 51 on the Artificial Analysis Intelligence Index—ahead of previous open-weights leaders and competitive with proprietary models—demonstrates that open-weight AI is rapidly improving. This trend accelerates the democratization of advanced AI, enabling smaller teams and organizations to deploy state-of-the-art capabilities without relying on closed APIs. The implication is that the value of proprietary models may erode unless they offer unique features (e.g., real-time data, specialized integration) that open models cannot easily replicate.
Consumer backlash against overt “AI” branding is strong and growing
The WordPress VIP report showing 60% of U.S. consumers are turned off by AI mentions in brand messaging signals a critical trust issue. “Bot fatigue” and a perceived loss of humanity online (74% of respondents) suggest that users value authentic, human-centered experiences. For AI/ML developers, this means that the success of AI-powered features depends on seamless, invisible integration rather than marketing the AI itself. Products that prioritize user benefit without hyping the underlying technology are more likely to gain adoption.
Version control for binary and AI assets is becoming a necessity
Epic’s Lore system—with its Merkle-tree-based, content-addressed storage and efficient handling of large binaries—addresses a growing pain point in AI/ML workflows. Datasets, model weights, and training artifacts can be massive and require reliable versioning, deduplication, and integrity checking. This trend indicates that the industry is moving beyond Git-like version control for code toward specialized tools for AI assets. Teams should evaluate such systems to improve reproducibility, collaboration, and auditability of their ML pipelines.
Privacy-focused operating systems are clashing with AI-driven telemetry in apps
Volkswagen blocking GrapheneOS users from using its app highlights an emerging conflict between user-controlled privacy hardening and the data-hungry nature of modern connected services (many enhanced by AI telemetry). As more apps rely on background data collection for AI features, they risk alienating privacy-conscious users. For AI/ML developers, this reinforces the need to design for privacy by default—for example, running models on-device, minimizing data exfiltration, and respecting device security profiles. Otherwise, they may face compatibility issues and user backlash.
Data lock-in and paywalls threaten user trust and AI training data availability
Photobucket’s decision to charge users for accessing their own uploaded images is a cautionary tale about centralized platforms monetizing user-generated data. In the AI/ML context, similar dynamics apply: many models are trained on data collected by platforms that may later restrict access or demand payment. This creates risks for both users (losing control of their content) and AI developers (losing access to training data). The trend toward decentralized, user-owned data (e.g., via IPFS or personal data stores) could gain momentum as a countermeasure.
The “small web” counter-movement underscores demand for authentic, non-AI-generated content
Bubbles.town, an aggregation of independent personal blogs, is thriving alongside growing skepticism of AI-generated or AI-curated content. This suggests a market for genuine human expression, free from algorithmic manipulation and synthetic text. For AI/ML, this implies that content moderation and recommendation systems must clearly label AI-generated material and prioritize human-authored content to avoid eroding trust. It also opens opportunities for AI to assist human creators without replacing them—e.g., editing tools, accessibility aids—while preserving the “small web” ethos.
Analysis generated by deepseek-reasoner