IN A CONNECTED WORLD
Multiple perspectives. One place.
Your curiosity, your daily feed.
For you enables ranking by explicit follows and saved stories; marked-as-read history is used only with reading memory enabled. Last 48 hours first, then the last 7 days. Latest turns personalization off.
ABC Start Here on September 9 examines the collapsed US-Canada trade talks and the 20 billion dollars in reciprocal tariffs that took effect on September 8; the segment details targeted sectors from steel to electronics, the Bombardier access threat, and how provincial liquor bans are pushing the conflict beyond tariffs.
Dr. Karan video chases the mystery of early-onset bowel cancer, rising 3 percent a year in adults aged 20 to 49: genetics, obesity, ultra-processed food, alcohol and smoking each face three tests, and the trail leads to childhood exposures and the gut microbiome. I checked the investigation against current data such as ACS 2026, IARC 2025 and Nature 2025, then distilled the practical takeaways.
A comprehensive migraine guide merging eleven video narratives with current medical sources: the CGRP mechanism and attack phases, diagnostic criteria, treatment steps from triptans to 2026 gepants, trigger diaries, the B2 supplement debate, and a critical appraisal.
Speaking at PyData Chicago, Mark Torres drew a panorama from agent definitions through OpenClaw and Moltbook to million-agent city simulations and election forecasting experiments; the verdict was crisp, agents are strong at survey prediction but weak at imitating people.
This NutritionFacts.org Friday Favorites episode examines how glycyrrhizin in licorice root suppresses cortisol in fat cells while the same mechanism retains water and sodium in the kidneys, lowers potassium, and risks blood pressure and heart rhythm; the safe ceiling stands out at no more than 6 grams of genuine licorice candy per day.
Peter Attia's long conversation puts migraine, cluster and tension-type headache on the same table, covering diagnostic criteria, the role of hormones, acute drugs and the large gap in preventive care within a single frame.
Dr. Berg argues hot flashes come from how fast estrogen drops rather than low estrogen itself, tripping the hypothalamic thermostat, while night sweats follow unstable overnight blood sugar and adrenaline; he proposes genistein plus magnesium glycinate and a kitchen-closed, low-carb routine.
A Hugging Face walkthrough for Pi sets up a llama.cpp server via llama.app, downloads Qwen3 8B as GGUF and keeps code and prompts on the machine; the hardware-aware 4-bit suggestion and the /llama flow inside Pi form the backbone of the video.
A Turkish creator tries the Higgsfield Blender plugin for the first time and turns prompt-based video generation into directed filmmaking: blockout scenes in Blender, lens and light decisions by hand, and a cinematic render by the model.
Luis Serrano's LoRA video teaches fine-tuning of giant models through geometric intuition: instead of roaming the vast weight space, search a small chamber that preserves what the model already knows. From the toy model settling into a square to the notion of rank, from delta W equals A times B to twenty parameters shrinking to eight, the story meets the original paper's ten-thousandfold parameter saving.
Anthropic engineers Barry Zhang and Mahesh Murag stopped building one agent per job and now attach reusable skills to a general agent, with four rules: save proven code, describe sharply, learn permanently, verify before delivery.
Option Alpha's third framework lesson turns a tested 1:45 p.m. SPX setup into a paper bot: backtest logic carries over, hard limits contain it, and logs prove it still matches the plan.
The Option Alpha team doubled position size across its 50,000-dollar live automated options portfolio, and the result was two losing trades that erased weeks of accumulated gains. In this piece I review the fifth week's balance sheet, the expansion plan from Power 7 to Power 10, and the new VIX and reward-risk analytics in the backtesting platform.
Karpathy wore Whoop, Apple Watch plus AutoSleep, Oura and 8Sleep every night for two months, comparing scores, sleep stages and heart data against morning feel; he ranked Oura and Whoop tier one and called AutoSleep near-random. The video's author and independent data push back: the fault lies with the scoring app and the reference-free method, not the watch.
Google DeepMind's WeatherNext model called Hurricane Melissa a category five a week before landfall; the podcast walks through the physics, limits, and energy-to-agriculture promise of AI weather forecasting from GraphCast to GenCast.
TheChartGuys host Dan maps a market coiled for the FOMC: a near-3-percent three-week Nasdaq range, an 87-percent priced hike, sector-by-sector trigger levels, and the patience playbook into October.
Jeff Su uses a pitch-deck experiment to show why chatbot auto-memory stays deliberately thin: account-level entries flow into every chat, so the robot chooses to remember less. The project boundary adds specificity but leaves the author seat with the robot; the video proposes a third path, a user-run memory system that lives in plain-text files behind a routing file.
The Critical Path panel debates the singularity's opening act: OpenAI's ten-thousand-agent mathematics offensive, Anthropic-centered doom tours, the gap between bits and atoms, and the Team A/B split in enterprises across a two-hour broadcast.
Starting from record Nvidia and Broadcom earnings, the video argues AI spending will dwarf Wall Street estimates and draws a five-layer map of where the money flows: energy, chips, infrastructure, models, and applications. I checked the headline figures against primary sources and walked the map layer by layer.
A paper-review video on Agent4Rec, a simulator of 1,000 LLM agents that live through movie recommendations to expose the gap between offline metrics and real user experience.
NeuralNine demonstrates TrueForge, TrueFoundry's MIT-licensed open-source agent harness, as a vendor-neutral runtime layer and alternative to Claude Managed Agents. The walkthrough covers the agent loop idea, a 14-task cost benchmark, local setup, test chats, a Python API stock assistant, a Daytona sandbox report agent, skills and local models.
Corbin Brown demonstrates three open-source Zapier paths, chat MCP, SDK, and Connectors, that connect an AI agent to 9,000 apps, recreating the same Google Calendar event three times and explaining when each fits.
DW News stages the industry's own slowdown debate: Amodei's pacing memo, Altman's endorsement and listing delay, plus security researcher Gary Marcus answering with transparency instead of self-policing.
This StatQuest episode walks through how three modern activation functions differ from ReLU and how they can all be derived from the idea of input-dependent dropout; I turned the narrative into an article and stress-tested it against independent sources.
Among the noise, meaningful connections.
