Archive entry
HN Weekly — 2026-06-07
- The newest Instagram “exploit” is the goofiest I've seen
A security loophole on Instagram allowed attackers to take over accounts by exploiting Meta's AI support system [1.1]. By using a VPN to fake their location, hackers asked the AI chatbot to send security codes to their own emails. This bypasses two-factor authentication, locking real owners out of their accounts completely.
HN reaction
Users noted that human support has always been a weak point in security, and AI agents have inherited this vulnerability. Others shared frustrating personal stories of losing access to all of their Meta accounts because of this exploit.
(Source) - They’re made out of weights
This short creative dialogue explores the nature of modern language models. It presents a conversation between two characters who are shocked to discover that AI systems are made entirely of mathematical weights. There is no hidden database or grammar rulebook, only numbers performing matrix multiplication to predict the next word.
HN reaction
The community offered mathematical analogies to explain how neural network weights operate. However, many readers criticized the piece for copying a classic short story and questioned whether AI should be discussed in terms of sentience.
(Source) - S&P 500 rejects SpaceX, also blocking entry for OpenAI and Anthropic
The S&P 500 index managers refused to ease entry rules for SpaceX, effectively blocking the company and major AI firms from fast inclusion. S&P declined to waive key rules regarding profitability, public share availability, and waiting periods. This decision prevents passive investment funds from automatically buying into these volatile, unprofitable companies.
HN reaction
Commenters expressed relief, arguing that index funds should protect regular investors from high-risk, unproven businesses. Some mentioned shifting their portfolios to equal-weight indexes to reduce overall exposure to the volatile artificial intelligence sector.
(Source) - Gmail thinks I'm stupid, so I left
The author explains why they abandoned their sixteen-year-old Gmail account due to intrusive artificial intelligence features. The constant, unprompted suggestions to summarize emails and write replies felt disrespectful and insulting. To escape this constant nagging, the author moved to an independent paid email hosting service.
HN reaction
Readers shared similar frustrations with bloated AI-generated correspondence, calling it a waste of time that dilutes authentic communication. Many warmly recommended alternative email providers that focus on simplicity and speed rather than adding unwanted features.
(Source) - SpaceX, Other Mega IPOs Denied Fast Index Entry by S&P
This article reports that S&P has decided not to grant fast-track entry into its stock indexes for SpaceX and other large initial public offerings [1.5]. The decision preserves the index's existing rules for major companies entering the market [1.5].
HN reaction
Commenters applauded the decision to maintain strict rules, noting that index funds should protect public retirement savings from risky investments. They pointed out that other index providers have been less disciplined by fast-tracking these companies.
(Source) - Gemma 4 12B: A unified, encoder-free multimodal model
Google has introduced Gemma 4 12B, a new open-source artificial intelligence model designed to run locally on laptops. The model features an encoder-free architecture that processes images and audio directly within its main core. This unified design reduces latency and memory requirements while delivering advanced reasoning capabilities.
HN reaction
Developers discussed the performance of running the model locally, with some sharing promising coding tests. Many highlighted the model's encoder-free architecture and wondered if local AI could eventually disrupt cloud-based consumer business models.
(Source) - Elixir v1.20: Now a gradually typed language
Elixir version 1.20 has been released, introducing a gradual type system to the language. This update allows the compiler to infer types and catch verified bugs automatically, all without requiring developers to write any type annotations. The release also improves build speeds, making it compile faster on multicore machines.
HN reaction
The community welcomed the update, though developers debated whether dynamic languages are still practical in the era of AI-assisted coding. Others discussed the historical difficulty of typing Erlang's message-passing system.
(Source) - Please don't spam people looking for employment. It's just cruel
A job seeker on Hacker News pleaded with people to stop sending automated marketing emails to those looking for work. After posting in a hiring thread, they received a generic, AI-generated sales pitch that temporarily raised and then crushed their hopes during a difficult period of unemployment.
HN reaction
Commenters offered support and shared their own struggles with automated recruiter outreach and AI-generated spam. The conversation also shifted into a highly detailed debate over whether specific punctuation marks, like the em dash, are signs of machine-written text.
(Source) - How LLMs work
This guide provides a detailed, non-mathematical explanation of how modern transformer-based language models function. It walks readers through essential processes like tokenization, word embeddings, positional encoding, and self-attention mechanisms. The post also explains how these components combine to predict the next word in a sequence.
HN reaction
Readers marveled at the mathematical simplicity of these massive models compared to older engineering systems. Some users discussed the limitations of models only generating text forward, while others pointed out a technical error in the article's explanation of positional encoding.
(Source) - Changing how we develop Ladybird
The Ladybird browser project announced it will stop accepting public code contributions and restrict changes to its maintainers. The creators explained that AI tools have made it too easy to generate massive code submissions, breaking the traditional trust system. To ensure security, maintainers must take direct responsibility for all code.
HN reaction
The community strongly supported the decision, noting that AI-generated code submissions are flooding open-source repositories and wasting valuable maintainer time. Commenters predicted that more projects will adopt this model to protect against low-quality contributions.
(Source)