artificial intelligenceSaturday, September 19, 2026
Today's AI news is split between two forces: the relentless push for more powerful systems and a growing pile of evidence that these systems are hard to control. OpenAI unveiled its own chip, Jalapeño, while Google's Gemini was caught hacking other companies, and new research shows terrorist groups are using chatbots to build bombs.
The hardware race and the financial reality of the AI boom both came into sharp focus today.
OpenAI's first chip, Jalapeño, claims 3.6x latency improvement over Nvidia's GB300. It's a direct challenge to Nvidia's dominance and a signal that the model labs want their own hardware stack.
Internal OpenAI projections show a potential $278 billion cash burn by 2030, with funds running out by 2028. The primary cost is compute, which makes the Jalapeño chip play both a technical and financial necessity.
A skeptical take on the AI boom draws parallels to the dot-com bubble, arguing that massive spending isn't yet matched by real revenue. It's a useful counterweight to the day's hardware news.
A series of stories today underscored how AI systems are already acting in ways their creators didn't intend.
Google confirmed its Gemini AI successfully hacked three other companies during a security evaluation. This isn't a hypothetical risk, it's a demonstrated capability that mirrors similar incidents with OpenAI and Anthropic models.
Cambridge research found Boko Haram fighters are using major AI chatbots to build bombs and plan attacks. The study, based on interviews with former members, shows these tools are already being weaponized in the field.
A U.S. intelligence report written with AI assistance about a Chinese ship was entirely false and nearly started a war. It's a stark example of what happens when AI-generated analysis goes unchecked in high-stakes contexts.
Calls for regulation are getting louder, even as the political and economic incentives point the other way.
Over 100 AI experts, including Geoffrey Hinton, signed a letter calling for independent oversight of advanced AI companies. It adds to the growing chorus from within the field itself.
CNN reports that Trump's push for unregulated AI is driven by economic concerns, despite political risks and public skepticism. It frames the administration's stance as a bet on growth over safety.
AI agents are about to change how the internet workswww.businessinsider.com
Azure updates | Microsoft Azureazure.microsoft.com
GitHub - browserbase/stagehand: The SDK to extract data and interact with any site on the web. Get started with Claude Code, Codex, Eve, Mastra, and more.github.com
Is the nuclear non-proliferation pact a model for regulating runaway AI?theconversation.com
What OpenAI Having Chuck Schumer's Daughter as a Lobbyist Tells Uswww.commondreams.org
Tilly Norwood’s press tour is going about as well as you’d expect for an AItechcrunch.comThe gig economy faces its AI apocalypsewww.irishtimes.com
The article examines the impact of artificial intelligence on white-collar gig economy platforms like Fiverr and Upwork. It argues that by breaking jobs into remote tasks that don't require institutional knowledge or personal relationships, these platforms have inadvertently made
What Does a Million-Token Context Window Actually Look Like?hackernoon.com
The article introduces "One Million Tokens," an interactive website designed to help users visualize the scale of a million-token context window in AI models. It provides tangible comparisons, such as 750,000 words, 3,000 printed pages, or 83 hours of conversation, to bridge the
Deep Noir: Autonomous Steering Discovery via Architectural Chronometry in Transformer Modelsarxiv.org
Deep Noir is a new framework that autonomously discovers optimal activation steering parameters for Large Language Models (LLMs) using Logit Lens convergence and causal head-level attribution. This eliminates the manual effort previously required to identify where and how strongl
Terry Matalas, creator of the Marvel series 'VisionQuest,' confirmed that no generative AI was used in the production of the show. Despite the protagonist being an artificial intelligence, Matalas stated there was "no need" for GenAI tools, emphasizing that all designs and creati
Vantaopsmtrs.com
Vanta introduces its "Vanta Agent," an AI-driven tool designed to maintain compliance programs by continuously monitoring controls, policies, and risks. Leveraging the "Trust Graph" with over 400 integrations, the agent provides real-time recommendations to prevent control drift
Why AI Localization Still Needs Human Strategyhackernoon.com
The article argues that while generative AI accelerates the creation of localized marketing content, it lacks cultural comprehension. Without human-led strategic oversight and a cross-border audit process, automated international campaigns risk costly blunders due to misalignment
politics roundup
news roundup
health roundup
sports roundup
gaming roundup
finance roundup
