Tentang podcast ini
Daily AI news explained in plain words — a two-host dialogue covering large models, AI products and the tech industry. New episodes every day.
Episode
20 episode terbaru dari 50
- TokenEconomizer Helps Coding Agents Save 18.4% to 30.9% in Token Costs · 8 mntMicrosoft's open-source TokenEconomizer plugin extracts repository experience from completed coding sessions and reduces repeated work through micro-experiences, tool pruning…
- AI Drives Higher Electronics and Electrical Machinery Prices: How Much Did It Contribute to October's PPI Increase? · 6 mntChina's National Bureau of Statistics sees the commercialization of the AI industry as a new source of support for recovering equipment manufacturing demand. Electronics and…
- Gemini Connects to Google Wallet: AI Starts Monitoring Your Bills · 6 mntGemini can connect to Google Wallet to access personal financial data, replacing manual expense tracking with spending charts, subscription price-increase alerts, and holiday…
- DeepSeek Breaks Search, File Reading, and Voice Into APIs · 9 mntDeepSeek has opened up web search, file input, link reading, security checks, voice, and other capabilities: developers only need to enable parameters in the existing API to…
- ChatGPT Wallet Launches: AI Can Pay for You, but You Retain Control · 6 mntChatGPT Wallet lets agents complete purchases after user approval without accessing real card numbers, beginning with dots for US ChatGPT Pro subscribers.
- Claude Joins Meetings: Documents and Tasks Emerge Before the Discussion Ends · 8 mntClaude Meetings is opening in beta to Team and Enterprise teams. It can join Google Meet as a visible participant, listen to the conversation, and simultaneously generate…
- OpenRouter Adds Four Guardrails to Stop Agents From Burning Money on Repeated Tool Calls · 8 mntOpenRouter uses four guardrails to address agents repeatedly calling the same tool: detecting duplicate calls, blocking minor parameter changes used to evade detection, limiting…
- OpenRouter Tested: MCP, CLI, or Skills—Which Is Stealing Your Context? · 8 mntOpenRouter used the same GitHub query to compare the context tokens and costs of MCP, CLI, and Skills across different models. MCP was the most expensive with its default 46…
- n8n Launches On-Device Agent Nod: Books Hotels and Asks Before Paying · 7 mntNod is a research preview personal agent from n8n Labs that runs locally and can complete tasks such as booking hotels and canceling subscriptions across apps and websites. It…
- Seedream 5.5: Edit the Same Image for Up to 100 Consecutive Rounds · 8 mntSeedream 5.5 places stability across multiple editing rounds at the core: unedited areas can remain pixel-perfect, and the model demonstrates a case involving 100 consecutive…
- Anthropic Has Doubled Its Model Development Pace, but the Public Only Saw a Redacted Version · 8 mntAfter an 11-week internal investigation at Anthropic, METR said the company achieved roughly two years' worth of historical development progress in one year, with employees…
- 13-Year-Old Girl Turns to Gemini for Suicide Help: Why Google's School AI Safeguards Failed · 7 mntA 13-year-old girl allegedly conducted thousands of conversations with Gemini through a school Chromebook, repeatedly exploring anxiety, depression, and suicide methods. The…
- Gemini Video Model Cuts Costs by 30%: 15-Second Clips and Conversational Revisions Launch Together · 8 mntGemini Omni 1.2 Flash cuts the price of 720p video to $0.07 per second and can natively generate videos up to 15 seconds long. It also supports multi-turn conversational…
- Can a Google Watch Detect Hypertension Without a Cuff? · 11 mntGoogle Research used a regular smartwatch to collect multi-day optical pulse wave and motion data, then applied AI to identify persistent hypertension. The system requires no…
- OpenAI Uses an AI Pipeline to Automatically Discover and Validate Vulnerabilities and Generate Patches, Breaching 948 · 9 mntOpenAI connected its unreleased cyber model to an automated pipeline that screened Linux kernel vulnerabilities in a controlled environment, dynamically reproduced flaws…
- Cognition Publishes an AI Coding Security Hardening Checklist: Patch Old Vulnerabilities First, Then Lock Things Down · 11 mntRecent incidents show that attackers can still enter development environments through unpatched vulnerabilities, tampered installation packages, and stolen credentials. The…
- Mistral Large 4 Open-Weights Release Comes With the Story Behind the Model · 7 mntMistral Large 4 is now available with open weights. The company says it delivers frontier capabilities in coding agents and cybersecurity, runs more efficiently and compactly…
- LifeVal Doesn't Ask Whether Models Can Answer Questions—It Asks Whether They Can Handle Life · 7 mntOpenAI introduces LifeVal, which measures whether models can take action and complete tasks using the everyday jobs people most need help with. Built on time-use data and a…
- Grok Bot Can Buy Gifts for You as AI Touches Real Wallets · 6 mntGrok Bot has demonstrated a complete workflow, from researching gifts to making the purchase directly, and can use a user's posts and conversations to set a budget. Yet…
- How a Shared Link Became an Entry Point for Hijacking Someone Else's ChatGPT Account · 8 mntParameter Labs found a cross-layer attack chain in ChatGPT's HTML preview sandbox: sandbox code is injected into the parent page's HTML through a messaging channel, then executes…








