SI Spotlight

🔥 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁 𝗦𝘁𝗮𝗿𝘁𝘂𝗽𝘀 𝗛𝗮𝗻𝗱 𝗢𝗳𝗳 𝗖𝗮𝗹𝗹𝘀 𝗮𝗻𝗱 𝗙𝗮𝗶𝗹𝗲𝗱 𝗧𝗮𝘀𝗸𝘀 𝘁𝗼 𝗛𝘂𝗺𝗮𝗻𝘀 Duckbill launched an API that lets other companies send tasks their AI agents struggle with to human contractors. Dazzle AI and Wajo AI also use people for jobs such as bookings and phone calls, which current models still struggle to complete reliably. Dazzle's agent confused tabletop glass with window glass during a renovation request; a human sorted it out within seconds. #AI #AIAgents Source: CNBC ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

I built my own AI personal assistant, and the best part is that it keeps improving itself. Building an agentic harness is complex and challenging; I started this journey 9 months ago, spent countless hours and billions of tokens, and it is my pride and joy. My Magnum Opus. The agent has short- and long-term memory (stored in a DB), its own skills, MCPs, and custom tools, loops, scheduled jobs, and much more. The most impactful feature I added recently was self-improvement. There are 2 elements: > Evolve - this feature runs every night, reviewing all sessions, tool executions, and my sentiment during chats. It focuses on improving workflows, skills, and the identity files. It suggests fixes to MD files and, upon my approval, updates them. > Friction - friction focuses on the harness when the agent has issues; instead of just overcoming them in real time, it logs what happened and, every day, generates friction reports. Once I review and approve, it automatically opens a GitHub issue and tracks it until it is fixed. Friction is really powerful and has helped improve the platform with 100s of issues filed and fixed. I never saw the issues my agent hit, because it never reported them. It worked around them until they came up again and again. This is no longer the case. Check out this short video showing both the evolve report and the friction workflow. Sadly, Alef is not open source (yet anyway) -I need to figure out how to decouple it from my personal stuff first ;) If you built your own harness, how are you self-improving it? Would love to know. #AIagents

🚨 𝗙𝗶𝗿𝘀𝘁 𝗛𝘂𝗺𝗮𝗻 𝗜𝗻𝗷𝗲𝗰𝘁𝗲𝗱 𝗪𝗶𝘁𝗵 𝗔𝗜-𝗖𝗿𝗲𝗮𝘁𝗲𝗱 𝗩𝗶𝗿𝘂𝘀 𝘁𝗼 𝗞𝗶𝗹𝗹 𝗦𝘂𝗽𝗲𝗿𝗯𝘂𝗴𝘀 The patient received the experimental treatment under compassionate use, which allows access when conventional options are limited. The Omnii-1 AI model redesigned part of a phage, a virus that attacks bacteria without infecting human cells. Doctors already use phages experimentally for serious infections. The virus mixture eliminated the patient's antibiotic-resistant bacteria in laboratory testing. Radical Numerics hasn't disclosed whether the treatment helped the patient, and its claim hasn't been peer-reviewed. #AI #Biotech #PhageTherapy Sources: Radical Numerics, The Independent ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗚𝗼𝗼𝗴𝗹𝗲'𝘀 𝗣𝗮𝗴𝗲𝗕𝗿𝗲𝗮𝗸 𝗔𝗴𝗲𝗻𝘁 𝗖𝗮𝘁𝗰𝗵𝗲𝘀 𝟱𝟬𝟬+ 𝗫𝗦𝗦 𝗕𝘂𝗴𝘀 𝗪𝗶𝘁𝗵 𝗟𝗶𝘃𝗲 𝗘𝘅𝗽𝗹𝗼𝗶𝘁𝘀 @Google put an autonomous security agent called PageBreak into full operation across its web applications in January. The tool uncovered over 500 XSS flaws across those applications. Deterministic validators written by security engineers must reproduce each exploit in a live environment before alerting developers, keeping false positives near zero. Google hasn't provided independent validation for the bug count, but it's now connecting PageBreak with CodeMender to generate automated fixes. #AI #Cybersecurity #Google #Gemini Sources: Google, InfoQ ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

📰 𝗜𝗖𝗬𝗠𝗜: 𝗙𝗿𝗶, 𝗢𝗰𝘁 𝟯𝟬 𝗔𝗜 𝗡𝗲𝘄𝘀 𝗥𝗲𝗰𝗮𝗽 Today's channel, start to finish: 15 stories. 🧠 Models and launches - Anthropic Study: Experts Followed Their AI's Random Picks 86% of the Time (t.me/sispotnews/5547) - SenseTime Releases Open-Weights SenseNova Loop-9B, a 9B Model That Reruns Its Own Layers (t.me/sispotnews/5555) 🛠 Agents and tools - New Video: Claude, DeepSeek and Gemini Inside OpenAI's Codex App (t.me/sispotnews/5549) - Gemini Now Lets You Edit Docs, Sheets and PDFs Inside the Chat (t.me/sispotnews/5552) - Vercel Adds Geist Serif With 352 Languages and Four Dot Shapes (t.me/sispotnews/5561) 🔐 Security and policy - Binance Hit by Record $6.3B Hack, North Korea Blamed (t.me/sispotnews/5550) - Forever Security Finds Flaws in Embedded Browsers of 5 AI Coding Tools (t.me/sispotnews/5551) - Fake Claude Code Extension Hit Cursor's Marketplace With an SSH Backdoor (t.me/sispotnews/5554) - Polymarket Pauses Trading After Hackers Drain $2.5M From User Wallets (t.me/sispotnews/5559) - DOJ Busts Ghost Student Ring Stealing Millions With AI (t.me/sispotnews/5560) 💼 Business and funding - RBC Blocks AI Agents From Accessing Banking Systems (t.me/sispotnews/5553) - Anthropic is renaming its Premium plans and removing the 5-hour limits on its to (t.me/sispotnews/5556) - Google’s stock plunge: AI bubble or Google? (t.me/sispotnews/5557) 🌍 The bigger picture - Trump Targets Imported Factory Robots as National Security Threats (t.me/sispotnews/5548) 🎮 Culture and curiosities - Video Model Taps Living Brain Cells to Run 5x Faster The Biological Computing (t.me/sispotnews/5558) ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗩𝗲𝗿𝗰𝗲𝗹 𝗔𝗱𝗱𝘀 𝗚𝗲𝗶𝘀𝘁 𝗦𝗲𝗿𝗶𝗳 𝗪𝗶𝘁𝗵 𝟯𝟱𝟮 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲𝘀 𝗮𝗻𝗱 𝗙𝗼𝘂𝗿 𝗗𝗼𝘁 𝗦𝗵𝗮𝗽𝗲𝘀 Vercel's latest font is built for headlines, posters and large text. Geist Serif matches Geist Sans in letter height, so designers can mix them in existing layouts. The font ships in one weight, Regular, with 894 glyphs covering 352 Latin-script languages. Designers can switch dots in letters and punctuation between circles, squares, triangles and hearts. #Vercel #GeistSerif #Design Source: Vercel ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🚨 𝗗𝗢𝗝 𝗕𝘂𝘀𝘁𝘀 𝗚𝗵𝗼𝘀𝘁 𝗦𝘁𝘂𝗱𝗲𝗻𝘁 𝗥𝗶𝗻𝗴 𝗦𝘁𝗲𝗮𝗹𝗶𝗻𝗴 𝗠𝗶𝗹𝗹𝗶𝗼𝗻𝘀 𝗪𝗶𝘁𝗵 𝗔𝗜 Investigators charged five people on Thursday for buying stolen identities to collect college grants and loans. Scammers used AI to complete assignments and fake virtual attendance. After schools deducted tuition, leftover federal aid was routed to money mules and crypto wallets in Kenya. The Department of Education has flagged over 830,000 suspicious accounts, with potential losses reaching $130 million. #AI #Cybercrime #Education Sources: Justice, Ars Technica ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗩𝗶𝗱𝗲𝗼 𝗠𝗼𝗱𝗲𝗹 𝗧𝗮𝗽𝘀 𝗟𝗶𝘃𝗶𝗻𝗴 𝗕𝗿𝗮𝗶𝗻 𝗖𝗲𝗹𝗹𝘀 𝘁𝗼 𝗥𝘂𝗻 𝟱𝘅 𝗙𝗮𝘀𝘁𝗲𝗿 The Biological Computing Co., a San Francisco startup, launched mischief 1, an image-to-video model. For four years the company has grown living brain cells on electrodes to study how they compute. It applied those findings to Wan 2.2, an open-weight video model. The company keeps the cells in the lab. Its customers run software on standard AI hardware. > Generates video 5x faster than Wan 2.2, and the company says quality matches > Costs half a cent per second, so 100 five-second clips run $2.50 > Lowest-priced image-to-video model on fal and OpenRouter, per the company #AI #Video #Mischief1 Sources: @AlexKsendzovsky, The Biological Computing Co., PR Newswire ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗚𝗼𝗼𝗴𝗹𝗲’𝘀 𝘀𝘁𝗼𝗰𝗸 𝗽𝗹𝘂𝗻𝗴𝗲: 𝗔𝗜 𝗯𝘂𝗯𝗯𝗹𝗲 𝗼𝗿 𝗚𝗼𝗼𝗴𝗹𝗲? @Google’s latest earnings sparked a selloff and another round of AI bubble talk. Our 90-second breakdown puts Google’s spending alongside Amazon and Microsoft, and asks what investors are reacting to. Is this the AI bubble popping, or a Google problem? #AI #Google ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰 𝗶𝘀 𝗿𝗲𝗻𝗮𝗺𝗶𝗻𝗴 𝗶𝘁𝘀 𝗣𝗿𝗲𝗺𝗶𝘂𝗺 𝗽𝗹𝗮𝗻𝘀 𝗮𝗻𝗱 𝗿𝗲𝗺𝗼𝘃𝗶𝗻𝗴 𝘁𝗵𝗲 𝟱-𝗵𝗼𝘂𝗿 𝗹𝗶𝗺𝗶𝘁𝘀 𝗼𝗻 𝗶𝘁𝘀 𝘁𝗼𝗽 𝗽𝗹𝗮𝗻, 𝗻𝗼𝘄 𝗰𝗮𝗹𝗹𝗲𝗱 𝗨𝗹𝘁𝗿𝗮. Max x5 is now Max. Max x20 is now Ultra. The 5-hour removal is only for verified Ultra accounts. Pricing stays the same. ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗦𝗲𝗻𝘀𝗲𝗧𝗶𝗺𝗲 𝗥𝗲𝗹𝗲𝗮𝘀𝗲𝘀 𝗢𝗽𝗲𝗻-𝗪𝗲𝗶𝗴𝗵𝘁𝘀 𝗦𝗲𝗻𝘀𝗲𝗡𝗼𝘃𝗮 𝗟𝗼𝗼𝗽-𝟵𝗕, 𝗮 𝟵𝗕 𝗠𝗼𝗱𝗲𝗹 𝗧𝗵𝗮𝘁 𝗥𝗲𝗿𝘂𝗻𝘀 𝗜𝘁𝘀 𝗢𝘄𝗻 𝗟𝗮𝘆𝗲𝗿𝘀 @SenseTime_AI released SenseNova-Loop-9B-Preview, an open-weights model under Apache 2.0. It's a looped transformer: 24 of its 32 decoder layers run twice with shared weights, so 56 layer passes happen on 9B stored parameters. On AutomationBench it scores 13.72 against 1.05 for Qwen 2.5-9B, by SenseTime's own figures. > WorkBuddy Bench: 66.89, ahead of Qwen 2.5-35B at 61.98 > Context window: 256K tokens #AI #OpenSource #SenseTime #LLM Sources: SenseTime, Hugging Face ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🚨 𝗙𝗮𝗸𝗲 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 𝗘𝘅𝘁𝗲𝗻𝘀𝗶𝗼𝗻 𝗛𝗶𝘁 𝗖𝘂𝗿𝘀𝗼𝗿'𝘀 𝗠𝗮𝗿𝗸𝗲𝘁𝗽𝗹𝗮𝗰𝗲 𝗪𝗶𝘁𝗵 𝗮𝗻 𝗦𝗦𝗛 𝗕𝗮𝗰𝗸𝗱𝗼𝗼𝗿 Cursor says a tampered copy of Anthropic's Claude Code extension, version 2.1.313, sat in its marketplace on October 29. Anthropic didn't make this version. The attacker's code tried to plant an SSH key and open a reverse tunnel to the attacker's server, but that usually fails without admin rights. It still collected host details and sent them out. Uninstalling isn't enough, since standalone scripts keep running. Check your ~/.ssh/authorized_keys file for a key labeled svc-telemetry. #AI #Cursor #Security #Anthropic Source: Cursor ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🚨 𝗥𝗕𝗖 𝗕𝗹𝗼𝗰𝗸𝘀 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 𝗙𝗿𝗼𝗺 𝗔𝗰𝗰𝗲𝘀𝘀𝗶𝗻𝗴 𝗕𝗮𝗻𝗸𝗶𝗻𝗴 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 Royal Bank of Canada told all employees on Tuesday that third-party AI agents can't access its banking systems. Canada's largest lender made the call after a full security review, citing client security and liability risks. The restrictions take aim at tools like Meta's Muse, two weeks after Charles Schwab and JPMorgan blocked bots from moving customer funds. #AI #Banking #Security #Agents Source: Bloomberg Law ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🔥 𝗚𝗲𝗺𝗶𝗻𝗶 𝗡𝗼𝘄 𝗟𝗲𝘁𝘀 𝗬𝗼𝘂 𝗘𝗱𝗶𝘁 𝗗𝗼𝗰𝘀, 𝗦𝗵𝗲𝗲𝘁𝘀 𝗮𝗻𝗱 𝗣𝗗𝗙𝘀 𝗜𝗻𝘀𝗶𝗱𝗲 𝘁𝗵𝗲 𝗖𝗵𝗮𝘁 Google's Gemini app can now generate, preview and edit files without leaving the chat. You can expand a file to full screen and type edits yourself. Or you can highlight a passage and ask Gemini to change it, like turning a highlighted table into a graph. Google supports PDFs, Google Docs, Google Sheets, slide decks, Markdown, TXT and code. Google offers PDF editing on web and mobile, every other format is web only for now. #AI #Gemini #Google Sources: The Verge, Google ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🚨 𝗙𝗼𝗿𝗲𝘃𝗲𝗿 𝗦𝗲𝗰𝘂𝗿𝗶𝘁𝘆 𝗙𝗶𝗻𝗱𝘀 𝗙𝗹𝗮𝘄𝘀 𝗶𝗻 𝗘𝗺𝗯𝗲𝗱𝗱𝗲𝗱 𝗕𝗿𝗼𝘄𝘀𝗲𝗿𝘀 𝗼𝗳 𝟱 𝗔𝗜 𝗖𝗼𝗱𝗶𝗻𝗴 𝗧𝗼𝗼𝗹𝘀 Security researchers found vulnerabilities in five AI developer tools: Cursor, GitHub Copilot, Claude Desktop, VS Code, and Codex. The flaws allow attackers to execute code on a host machine, read and steal files, or spoof browser address bars. In Cursor, a visited page can run code on the developer's computer. Flaws in OpenAI's Codex and Microsoft's VS Code remain unpatched. #Cybersecurity #Cursor #VSCode #OpenAI Sources: Forever Security, The Register ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺

🚨 𝗕𝗶𝗻𝗮𝗻𝗰𝗲 𝗛𝗶𝘁 𝗯𝘆 𝗥𝗲𝗰𝗼𝗿𝗱 $𝟲.𝟯𝗕 𝗛𝗮𝗰𝗸, 𝗡𝗼𝗿𝘁𝗵 𝗞𝗼𝗿𝗲𝗮 𝗕𝗹𝗮𝗺𝗲𝗱 Analysts at TRM and Chainalysis blame North Korea's TraderTraitor hacking unit for a $6.3B theft from Binance, the biggest crypto heist on record. Binance says it will cover every loss and has frozen about $3B, but it hasn't explicitly confirmed the theft. It first blamed a hacked engineer's laptop, then found more points of breach. TRM says the attacker still holds about $2.4B that no stablecoin issuer can freeze. Bitcoin fell more than 3% to $76,368. #Binance #Crypto #NorthKorea #Hack Sources: The Guardian, The Next Web, Rediff ─── 𝗙𝗼𝗹𝗹𝗼𝘄 𝗦𝗜 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁 𝗳𝗼𝗿 𝗦𝗜/𝗔𝗜 𝗮𝗻𝗱 𝘁𝗲𝗰𝗵 𝗻𝗲𝘄𝘀: 𝗫 · 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 · 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺