Hacker News
1Claude Sonnet 4 now supports 1M tokens of context
- <p>Claude Sonnet 4 now features increased context capacity of 1 million tokens, enabling processing of larger codebases and documents. The update supports comprehensive code analysis, document synthesis, and context-aware agents, with adjusted pricing for prompts over 200K tokens. Available in public beta through Anthropic API and Amazon Bedrock.</p>
Context window management
Debate over optimal context window size, with some arguing larger windows can confuse LLMs while others value more context. Users discuss strategies like clearing context regularly and breaking tasks into smaller chunks.Enterprise vs hobby use
Discussion of LLMs' effectiveness in professional vs hobby development. Many find them useful for prototypes and small projects but inadequate for enterprise code, citing issues with code quality and maintainability.Working with Claude Code
Tips for effectively using Claude Code, including managing token usage through checkpoints, treating AI as an architect rather than code generator, and using plan mode for better results.
2StarDict sends X11 clipboard to remote servers
- StarDict, a cross-platform dictionary application, has raised security concerns due to its default configuration in Debian that sends users' text selections over unencrypted HTTP to remote servers. Despite multiple reports since 2009, this privacy issue remains unresolved, highlighting challenges in maintaining security across open-source software.
Package transparency and communication
Debate over whether package descriptions adequately disclose network functionality, with many arguing maintainers shouldn't expect users to read all documentation and that sensitive features should be opt-inOnline vs offline dictionary functionality
Discussion about whether dictionary programs should work offline by default, with arguments about dictionary size, practicality of local storage, and generational expectations about software connectivityPrivacy and security implications
Concerns about unencrypted data transmission to Chinese servers, broader privacy issues in Linux distributions, and debate over whether this represents malicious intent or poor security practices
3Monero appears to be in the midst of a successful 51% attack
- Monero blockchain faces a critical security threat as Qubic mining pool gains majority control, enabling potential network manipulation through a 51% attack. Despite the significant threat to network integrity and an estimated daily attack cost of $75M, XMR's value has only decreased by 13%.
Attack legitimacy
Debate over whether Qubic actually achieved 51% hashrate control of Monero. Some argue it was just lucky mining, while others point to evidence of network reorganizations. Questions raised about Qubic's hashrate reporting accuracy.Impact of 51% attacks
Discussion of what attackers can actually do with 51% control - ability to rewrite transaction history and double spend, but not steal funds or create fake transactions. Debate over whether attacks are economically viable.Mining economics
Analysis of attack costs ($75M/day), sustainability, and mining incentives. Debate over whether attacks are "expensive" if mining remains profitable, and discussion of Qubic's business model offering alternative rewards.
4OpenSSH Post-Quantum Cryptography
- OpenSSH has implemented post-quantum cryptography to protect against future quantum computer attacks that could break current encryption. Starting with version 10.1, it warns users when connections use non-quantum-safe algorithms, encouraging migration to stronger security measures that defend against potential "store now, decrypt later" threats.
Hybrid algorithms
OpenSSH implements "hybrid" post-quantum algorithms that combine PQ with classical methods, ensuring security won't be worse than current standards even if PQ algorithms are broken. Some debate on implementation risks vs benefits.Urgency of adoption
Discussion of whether post-quantum crypto adoption is needed now given quantum computers don't exist yet. Focus on "store now, decrypt later" risks vs implementation overhead, with most supporting proactive adoption.Performance impact
Debate around overhead of PQ crypto, with consensus that impact is minimal since it only affects key exchange, not ongoing encryption. Larger keys but infrequent exchanges mean limited real-world performance cost.
5Trump Orders National Guard to Washington and Takeover of Capital’s Police
- President Trump announced a temporary takeover of Washington D.C.'s police department, deploying 800 National Guard troops despite low crime rates. Mayor Bowser acknowledged his authority but disputed his claims about city conditions. Trump also threatened similar actions in other cities and plans to reassign FBI agents for nighttime patrols.
Deployment legality
Discussion of legal framework for National Guard deployment in DC, including Posse Comitatus Act implications, DC's unique status, and presidential authority over DC National Guard versus state governors' authority elsewhereCrime rates debate
Contrasting views on DC crime statistics - some cite high crime rates as justification for deployment while others note violent crime is at 30-year low and question if deployment is politically motivatedDemocratic implications
Concerns about using military for domestic policing, impact on protest rights in DC, and potential precedent for expanding federal control over local law enforcement in other jurisdictions
6GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models [pdf]
Model performance and comparison
Users share experiences comparing GLM-4.5 with other models like Claude and GPT, noting GLM-4.5's strong coding abilities, better "honesty" in debugging, but some limitations with larger contexts.AI-generated content concerns
Discussion about whether certain comments are AI-generated, with debate about detecting LLM writing patterns and whether AI-assisted writing should be disclosed or is problematic.Local model possibilities
Optimistic discussion about future availability of powerful local coding models, with predictions ranging from end of year to 2 years for workstation-grade performance matching current cloud models.
7Why are there so many rationalist cults?
- The article explores how the Rationalist community, while generally functional, has spawned several problematic groups. Despite being founded on principles of rational thinking, some offshoots have developed cult-like characteristics, including occult beliefs and harmful practices. The article particularly examines cases of dysfunctional groups that attracted vulnerable members seeking belonging and transformation.
Problems with rationalist thinking
Discussion of how rationalists build conclusions on untested axioms, have excessive confidence in logical chains despite compounding errors, and lack intellectual humility characteristic of truly smart peopleCult dynamics in rationalist groups
Analysis of how rationalist communities can evolve into cults, driven by narcissistic leaders, desire for belonging, and elaborate belief systems, similar to historical patterns in religious movementsArtificial intelligence concerns
Debate about rationalists' focus on AI extinction risk versus other existential threats, questioning assumptions about recursive self-improvement and whether AI poses unique catastrophic dangers
8GitHub was having issues
- A service outage occurred on Aug 12, 2025, impacting API requests, issues, pull requests, actions, and packages. The incident began with degraded performance at 14:12 UTC and gradually improved over several hours, with full resolution announced at 17:56 UTC, though some users experienced latency issues during recovery.
Github reliability concerns
Multiple users discuss GitHub's frequent outages, with debate about whether it's the "most unreliable SaaS". Enterprise customers advised to track downtime and request contracted uptime reports. Comparisons made to Bitbucket/Atlassian reliability.Alternative platforms
Discussion of self-hosted alternatives like Forgejo, Gitea, and GitLab. Users share experiences with different platforms and debate their features, with focus on independence from centralized services and avoiding single points of failure.IPv6 support
Users criticize GitHub's lack of IPv6 support as embarrassing and costly. Discussion touches on Azure's poor IPv6 implementation compared to other cloud providers, forcing some users to purchase IPv4 addresses.
9Australian court finds Apple, Google guilty of being anticompetitive
US antitrust enforcement
Debate over US willingness to pursue antitrust cases, with some arguing historical reluctance while others point to recent FTC actions against tech giants. Discussion of political influence and strategic considerations.Legal basis differences
Contrasting treatment of Apple vs Google in courts, focusing on how Apple's closed ecosystem avoided monopoly designation while Google's partially open Android faced penalties. Discussion of legal definitions vs practical impacts.Payment systems
Discussion of app store payment requirements, 30% fees, and whether large companies should be allowed to handle their own payment processing. Arguments for graduated fee structures based on company size.
10Show HN: Building a web search engine from scratch with 3B neural embeddings
- This article chronicles a two-month project to build a search engine from scratch, motivated by declining search quality and advances in language models. The author developed a system using neural embeddings and natural language processing to better understand search intent, rather than just matching keywords, ultimately creating a search engine capable of handling complex queries with improved relevance.
Search engine performance and challenges
Discussion of search result quality, including limitations with vector-only search, ranking algorithms, and challenges in getting relevant results for specific queries vs current search engines like Google.Cost and sustainability
Analysis of operational costs, including OpenAI's low embedding costs, potential for community funding, and debate about making the project sustainable through subscriptions or donations.Technical implementation
Detailed discussion of the project's architecture, including service mesh design, database choices, vector search implementation, and crawling strategies.