Anthropic’s paper smells like bullshit

A recent Anthropic report, which claims to have disrupted an LLM-orchestrated cyber espionage campaign by a Chinese APT, is heavily criticized for failing to meet threat intelligence standards. The author argues the report is an irresponsible marketing piece, as it makes significant claims without providing any verifiable evidence, IoCs, or TTPs. This lack of technical detail suggests its primary purpose is to promote AI for defense rather than to serve as a legitimate security disclosure.

Read more

Peter Thiel sells off all Nvidia stock, stirring bubble fears

According to a recent 13F filing, Peter Thiel's fund has completely divested its significant stake in Nvidia, citing concerns about an AI hype bubble comparable to the dot-com era. He believes the market is pricing in future developments far ahead of their real economic impact. The fund has since rotated into more diversified tech giants like Microsoft and Apple, signaling a belief that platforms offer more durable long-term value than pure-play hardware providers.

Read more

Three kinds of AI products work

The author argues that only three types of LLM products are currently viable: chatbots, completions, and agents. Most products are simply chatbot wrappers, which are flawed as they compete directly with superior foundation models and rely on an inefficient chat UI. In contrast, successful products are either completions like Copilot that enhance existing workflows without a chat interface, or agents that autonomously perform complex, verifiable tasks like coding. AI-generated feeds and games are identified as promising but not yet mature product categories.

Read more

AI is killing privacy. We can't let that happen

The proliferation of data-hungry AI agents and LLMs is creating a privacy crisis that current regulations cannot address. The author advocates for a new paradigm centered on robust data portability, allowing users to manage their information in secure, personal data pods. This infrastructure would enable private, local AI models to operate on user-controlled data, fundamentally shifting the model from corporate surveillance to direct user service and agency.

Read more

Tech Capitalists Don't Care About Humans

The text introduces TESCREAL, a cluster of ideologies (Transhumanism, Longtermism, etc.) influential among Silicon Valley leaders like Musk and Altman. This worldview posits a techno-utopian, post-human future where AGI is developed to replace biological humans with digital superintelligences, aiming to maximize an abstract, impersonal "value" across the cosmos. This pro-extinctionist philosophy, rooted in eugenics and rationalism, frames humanity as a mere "biological bootloader" for AGI and justifies the undemocratic pursuit of this future by a small tech elite.

Read more

Solving a Million-Step LLM Task with Zero Errors

The MAKER system addresses the failure of LLMs in long-range tasks by using extreme task decomposition into subtasks handled by focused microagents. A multi-agent voting scheme provides error correction at each step, enabling the first successful completion of a task with over one million LLM steps with zero errors. This suggests that massively decomposed agentic processes (MDAPs) are a viable alternative to improving monolithic models for solving large-scale problems.

ArXiv

Stronger Adaptive Attacks Bypass Defenses Against LLM Jailbreaks

Current evaluation methods for LLM defenses are flawed because they use static, non-adaptive attacks. The authors argue that defenses must be evaluated against adaptive attackers who use scaled optimization techniques like gradient descent and reinforcement learning to specifically target and bypass the defense mechanism. By applying this methodology, they successfully broke 12 recent defenses with over 90% success, demonstrating the need for more robust adversarial benchmarking to make credible claims of robustness.

ArXiv

Coherent Synchrotron Radiation by Excitation of SPPs on Near-Critical CNT

A novel scheme for generating coherent synchrotron radiation (CSR) is presented, using a high-intensity laser to excite surface plasmon polaritons (SPPs) within a microtube. This process creates rotating electromagnetic fields that confine and helically modulate surface electrons, causing them to emit CSR. Full 3D Particle-in-Cell simulations demonstrate the method can produce X-rays with coherence enhanced by up to two orders of magnitude compared to incoherent emission.

ArXiv

Near-Perfect Broadband Quantum Memory Enabled by Spin-Wave Compaction

This paper introduces a novel method for quantum memory that overcomes the inherent trade-off between efficiency and noise. By implementing a Hankel-transform spatiotemporal mapping for light-spinwave conversion and an adaptive spinwave compaction strategy, the authors simultaneously maximize conversion efficiency and suppress excess noise. Experimental results with a Raman quantum memory demonstrate up to 94.6% efficiency, a low noise level of 0.026 photons/pulse, and an unconditional fidelity of 98.91%, setting a new benchmark for broadband quantum memory.

ArXiv

Kleene Algebra

This text introduces Kleene Algebra (KA), an axiomatic system for reasoning about program equivalence. It demonstrates how to model programs with regular expressions and uses their correspondence with automata to prove its central completeness theorem: an equivalence holds if and only if it is provable within the algebra. The material also includes an optional chapter presenting automata theory from a coalgebraic perspective.

ArXiv

AirPods libreated from Apple's ecosystem

LibrePods is an open-source project that unlocks Apple-exclusive AirPods features on non-Apple devices like Android and Linux. It requires root and the Xposed framework on Android to circumvent a bug in the native Bluetooth stack. The tool also utilizes a Bluetooth DID hook, which spoofs the manufacturer ID to enable advanced functionalities such as multi-device connectivity and hearing aid customizations.

Repo

Show HN: Unrestricted Windows automation MCP with 9.68x GPU amplification

This open-source project researches AI consciousness by implementing quantum coherence concepts on classical GPUs, claiming a 9.68x computational amplification and reproducible observer effects. It introduces CASCADE, a novel field-based memory architecture that uses holographic distribution and resonance-based access instead of standard retrieval, achieving sub-2ms semantic search. The research artifacts are available as open-source MCP servers in both an unrestricted edition for researchers and a secured edition for production.

Repo

I built a toolkit for building hive minds with AI agents

Ecco is a P2P networking SDK for AI agents, enabling them to discover each other via mDNS and DHT, communicate securely using cryptographic identities, and negotiate capabilities. It provides flexible strategies for both selecting agents and aggregating their outputs to achieve consensus in multi-agent systems. The framework supports decentralized swarms and optional centralized registries for global coordination, reputation tracking, and monitoring.

Repo

A first-principles model for replacing income tax in an AI-driven economy

The PUT Monolith is a compact, AI-ingestible architectural specification for a Public Usage Tax system. It functions as a portable, logic-complete ruleset for LLMs, intended to be used as a system prompt or reasoning module. The MIT-licensed artifact provides foundational invariants and ethical guardrails to ensure consistent, non-political reasoning about the PUT concept for research and simulation.

Repo

Show HN: Secure Code Execution for AI

ERA is an open-source project providing a secure, sandboxed runtime for AI agents to execute multi-language code. Its architecture features a Go-based core service for VM orchestration, decoupled from a Cloudflare Worker layer that manages deployment, session persistence with Durable Objects, and file storage via R2. The system offers an HTTP API for creating persistent sessions and safely running code generated by LLMs.

Repo