The Archive

One-fifth the cost!

Unverified claim of Deepseek V4 AGI confirmation on social media, lacks evidence or official announcement.

Haven’t run a single request and my session shows 100% used. Weekly usage jumped from 5% → 20% out of nowhere. No inputs, no outputs, no cost. What exactly is being counted here?

https://preview.redd.it/nz2e10ujl6xg1.png?width=1594&format=png&auto=webp&s=455c898d9c0050748164d8f1fe8f935e66b39eff Last week I hadn't any problems with tokens and reset, those terrible issuing, looks as scam tentatives, start from yesterday: \- Weekly session limit reset was fixed on thursday at 1am, swithced at 10pm! \- This afternoon I end the session limit token in half hour, same usage of last week, and I haven't encountered any problem last week! \- Limits Reset 3 minutes ago, launched a prompt on Claude Code and I reached 52% in 3 MINS \- it resets EVERY 5 hours now! ...

Pi coding agent executes commands without sandboxing by default; community identifies permission-gate and sandbox extensions as mitigations.

Google announces it will invest up to $40 billion in Anthropic, its largest single AI investment ever. https://x.com/nolimitgains/status/2047709664423420358?s=46

Google plans up to $40B investment in Anthropic as AI rivals race to secure massive compute capacity, following the limited release of its powerful, cybersecurity-focused Mythos model.

Active experiment selection method reduces cost of fitting scaling laws for LLM training via uncertainty-aware sequential design.

14-author perspective paper argues a scientific theory of deep learning is emerging, synthesizing five recent research lines to explain why large learning systems work.

Claude Design product hits weekly rate limit with no export recovery, trapping users with stale project versions.

Systematic analysis of token consumption patterns in agentic coding tasks across eight frontier LLMs on SWE-bench Verified.

Study quantifies representational harms and stereotyping of Global Majority nationalities in LLM-generated text narratives.

Taxonomy of world modeling capabilities for AI agents across three levels (predictor, simulator, reasoner) organized by environmental laws.

Reddit discussion criticizing high registration fees and declining standards at ICLR and other ML conferences.

Training regularizers improve MILP embedding properties of neural network surrogates via relaxation-informed objectives.

A new era is on the way for Apple as Tim Cook plans to step down from his CEO role in September, handing the reins to hardware chief John Ternus. Ternus may be inheriting one of the most durable businesses in tech, but he’s also stepping into a very different ecosystem than the one Cook spent decades shaping. The App […]

Undecidability proof for plan existence in modal epistemic logic with depth-1 preconditions and no postconditions.

Anthropic, the AI lab valued at $380 billion, has acknowledged that a series of engineering missteps were behind a widely-experienced decline in the performance of its Claude Code tool that sparked a user revolt over the past month. The latest admission, which came after weeks in which Anthropic had initially implied in its communications that nothing was wrong and that users were largely to blame for any performance problems and later said some of the changes had been made for users’ benefit, has done little to calm Anthropic’s customers—some of whom say they have already cancelled their su...

BantuMorph transformer recovers 728 noun and 1,525 verb cognate candidates in Bantu languages via encoder embeddings.

I just played a game with my kids telling it to generate random images from gibbrish, and one of them was this, never mentioned Gemini or Nano Banana

Cross-lingual transfer + clustering discovers morphological patterns in low-resource Giriama language from minimal labeled data.

Utility-Aligned Embeddings framework distills LLM re-ranking signals into dense retrievers for high-performance RAG without inference cost.

Anecdotal observation of Qwen3.6-35B hallucinating context-full status at appropriate inference point, notes possible emergent behavior.

Reddit user requests 24-hour time format option in Claude UI instead of AM/PM.

Small business owner reports using Claude for custom app development after Gemini Pro struggled with security requirements.

In the first 24 hours of the assault on Iran, the US military struck more than 1,000 targets, nearly double the scale of the "shock and awe" attack on Iraq over two decades ago. This acceleration was made possible by AI systems that speed up the targeting process. Chief among them is the Maven Smart System. In her new book, Project Maven: A Marine Colonel, His Team, and the Dawn of AI Warfare, journalist Katrina Manson investigates the development of Maven from its inception in 2017 as an experiment in applying computer vision to drone footage. The project spurred employee protests at Google,...

Stratechery weekly digest covering Tim Cook's Apple departure, Cursor IDE, SpaceX developments, and geopolitical competition.

The AI-driven memory shortage is hitting Samsung's bottom line.

OpenAI and Anthropic job listings show unexpectedly high go-to-market hiring relative to engineering and research roles.

Abstract Chain-of-Thought uses reserved token vocabulary for latent reasoning, matching verbal CoT performance with shorter generation.