Online AI discussion on 22 September centered on the cost of generated text, uncertain measurements of model behavior, local image generation and dramatic media framing of safety incidents. » Why it matters: Community forums often surface useful experiments and criticism before formal publication. They also mix evidence, demonstrations and opinion, so each item below is labeled by what it can support. Hacker News Attention becomes the scarce resource The most active thread linked an essay arguing that abundant media makes human attention more valuable. Commenters ranged from browser-history design to the way language models lower the cost of producing text. This is worth attention because it frames an economic problem for publishing: readers pay the review cost even when an author pays almost no production cost. ...
Aikido Releases Local Cybersecurity Model
Reuters reports that Aikido Security released an open-weight model for code security that organizations can run locally instead of sending repositories to an external provider. Aikido Security, a Belgian application-security company, is moving part of AI-assisted code analysis closer to the customer. The reported model is designed for cybersecurity tasks and local deployment. The central promise is data control: sensitive source code can remain inside the user’s environment while the model examines it. ...
China Slows Humanoid-Robot IPO Reviews
Reuters reports that Chinese regulators are applying closer scrutiny to humanoid-robot listing applications as concern grows that valuations have outrun commercial revenue. China’s humanoid-robot sector has attracted capital on expectations that machines shaped for human environments will eventually work in factories and services. The reported regulatory response does not ban listings. It indicates a slower review of whether companies have enough revenue, technology and credible demand to justify public-market claims. ...
FT Tests Chatbots on Personal Finance
The Financial Times reports that leading assistants failed many personal-finance questions in its newsroom test, raising concerns about answers that readers may act upon. The Financial Times tested chatbots on financial questions and concluded that they were wrong most of the time. The finding matters because a plausible error about tax, borrowing or investments can lead directly to a costly decision. It should also be interpreted cautiously because the complete method was not accessible during this verification. ...
Iambic Files for a US IPO
Reuters reports that AI-assisted drug developer Iambic Therapeutics filed for a Nasdaq listing under IAM without yet setting the offering size or price. Iambic Therapeutics, a San Diego biotechnology company founded in 2019, is seeking access to public markets. The company uses machine learning in small-molecule drug discovery and is developing IAM1363, an experimental treatment in early clinical testing for solid tumors. Its filing arrives while investors are selectively returning to biotechnology listings. ...
Kev Brings Decision Models to Local Hardware
Jared Palmer released three small Qwen3.5-based models that answer structured decisions locally and expose probabilities instead of only a final label. Kev is an open project for questions such as whether to escalate a support case, which department should handle it or how strongly a text fits a rating scale. Its 0.8B, 4B and 9B models can run on common local hardware, including Apple Silicon, and use an API compatible with TypeSafe’s hosted System One service. ...
Linear Rebuilds CI for Agent-Written Code
Linear says AI agents now write most of its tests, pushing the company to redesign continuous integration as its test suite nearly quadrupled during 2026. Linear, the software company behind a project-management platform, found that generating code faster moved delay into verification. Every pull request still had to wait for checks. The company changed runners, reduced repeated setup, shortened gate jobs and rebalanced tests to keep feedback from slowing developers and agents. ...
Meta Muse Gains Shopify, Loses Amazon
Shopify will reportedly support purchases through Meta’s Muse assistant, while Amazon has blocked the same agent from shopping on its marketplace. Meta’s personal AI agent has encountered two different rules for online commerce. The Wall Street Journal reports that Shopify will connect Muse to Shop Pay, its checkout service. Forbes and other outlets report that Amazon has denied the agent access, citing unauthorized behavior. The split shows that an agent cannot complete a purchase merely because its model understands the task. ...
Mini-AGI Tests Continual Learning on 8GB
An open experiment called mini-AGI trains a byte-level language model on one 8GB GPU and keeps updating it as it reads, while warning that it remains a toy. Mini-AGI is a research repository, not a frontier assistant. Its author is testing whether a small model can keep learning from one stream of data without erasing earlier knowledge. The system stores expert weights as files, pages selected parts through memory and grows or removes capacity during training. ...
SpaceXAI Releases Grok 4.7
SpaceXAI released Grok 4.7 on 21 September for coding and professional knowledge work, with API access starting at $2 per million input tokens. SpaceXAI, the company that develops Grok, has introduced a larger model with longer reinforcement learning and a new safeguard system. The release is available through the Grok API, Grok Build, Cursor and other model platforms. Developers now have another frontier-priced option, but the evidence published with the launch comes mainly from the vendor. ...