fleyta

Google DeepMind CEO, Demis Hassabis: "AGI is now on the horizon, and it will be the most profound and impactful technology ever invented. When we look back at this time, I think we will realize that we were standing in the foothills of the singularity." In this Google I/O 2026 keynote, Hassabis unveils Gemini 3.5, AlphaEarth Foundations, and Gemini for Science, aimed at one day solving all disease. Watch it today, then read how AI simulations are already predicting hurricanes before they form in the article below.

Andrej Karpathy just did a 29-minute fireside chat that works as the syllabus for agentic engineering in 2026: 00:00 – Why he said he's never felt more behind as a programmer 02:32 – Software 3.0, prompting as the new programming layer 09:41 – Verifiability, why models are jagged and what that means for founders 13:37 – Where to build when labs already own math and code 15:49 – Vibe coding vs agentic engineering, and why 10x stopped being the ceiling 19:29 – What AI-native coding actually looks like, and why hiring is still stuck in the old paradigm 21:34 – Animals vs ghosts, the framing for why yelling at a model changes nothing 25:19 – Agent-native infrastructure, sensors and actuators instead of docs written for humans 27:57 – "You can outsource your thinking, but you can't outsource your understanding" This one talk covers more than 10 agentic engineering threads combined. Watch it today, then check the car wash example at 09:41, it says more about frontier models than any benchmark will.

ONE WRONG PORT AND A $599 MAC MINI M4 THAT WON'T BOOT STAYS A DEAD SLAB OF ALUMINUM FOREVER, HE SHOWS THE ONE PORT THAT ACTUALLY REVIVES IT. 00:02 he pinches a braided usb-c cable and lines it up against the mac mini's rear panel, skipping the power connector and the hdmi side completely, then pushes it into the single thunderbolt port sitting right next to the power button before flipping the unit over to hold that button down. this isn't a settings menu or a software reset. it's the one hardware level port that talks to the chip before macos even tries to load, so a mac mini that won't show an apple logo, won't boot, won't respond to anything becomes recoverable through one cable and a second mac instead of a trip to a genius bar appointment. apple's $599 mac mini m4 ships with 4 rear ports and exactly 1 of them is dfu capable, apple configurator then pulls a restore image that runs past 15gb, and the whole sequence needs the power button held for roughly 10 seconds while that single cable stays connected the entire time. the interesting number here is not the 15gb restore file or the $599 price tag. it is 1, as in exactly 1 port out of 4 that brings the machine back, plug into any of the other 3 and the same $599 box just stays bricked.

A GUY 3D PRINTED A HINGE FOR $14, SCREWED AN 8.3 INCH iPAD MINI ONTO A $599 MAC MINI M4, AND BUILT A WORKING MACBOOK NOBODY SELLS. 00:03 he pulls up 3D print files from a youtuber, prints a white plastic hinge and base shell using maybe 150 to 200 grams of filament, then screws the 293 gram ipad mini to the top of the 670 gram mac mini like a laptop lid, one usb-c cable running through the hinge to keep both talking. the ipad isn't decoration, it becomes the actual 2266x1488 resolution screen running macos off the m4 chip inside the mini, so a $599 desktop and a $499 tablet that used to sit 2 feet apart on a desk fold into one 12.7 x 12.7 cm clamshell he can close and carry. the printed parts ran just over 100 yuan, roughly $14, on top of 2 devices he already owned worth around $1100 combined, add maybe 6 to 8 screws and 3 to 4 hours of print time and that's the entire build, no $1600 macbook pro, no new panel, no new logic board. the interesting number here is not the $14 print job or the $599 mac mini. it is 0, as in 0 apple parts bought new, he fused a $599 computer and a $499 screen apple has sold separately for over 10 years into a laptop apple itself has never made.

@ch3nweiii treating nanoseconds like dollars completely changes how you even approach building trading systems

citadel's engineering lead explained why your trading code loses before it even runs not your logic. not your signal. a number you've probably never thought about 30 microseconds that's the entire budget for a trade. not milliseconds. microseconds a human blink takes 400,000 of them here's how it gets spent: 2 micros to decide. 5 to check risk. 3 to send the order that's 10 accounted for the other 20 vanish before your code touches anything your signal has to physically travel to chicago. light speed is the ceiling and physics doesn't negotiate that chunk is locked. nothing you write makes it smaller so the only thing you actually control is the 10 micros where your code lives and if you show up 10 nanoseconds late, someone else already has the fill ten nanoseconds. that's light crossing a small room this happens billions of times a day in every liquid market on earth retail hears "HFT is fast" and stops there. that's the whole reasoning chain for most people but the real question is what these firms do with the sliver they can control how do you guarantee a 2-microsecond decision when the language you wrote it in might pause to collect garbage at the worst possible moment that's the answer citadel keeps landing on. not python. not java one language, decades later, still running the core of every serious shop not because it's elegant. because it's the only one that lets you promise the timing you weren't beaten by a better prediction you were beaten by someone who treated 10 nanoseconds like it was money

A CHINESE DEVELOPER PROPPED HIS iPHONE AGAINST A $599 MAC MINI M4 AND SHRANK A FULL DESKTOP INTO SOMETHING THAT FITS IN A JACKET POCKET. 00:45 he leans the phone against the mac mini's top edge as a tiny detachable monitor, screen mirroring a notes app, cursor jumping between folders while a slim keyboard and mouse sit right beside the unit on the desk. the whole point isn't the phone acting as a screen. it's that a 5.3x5.3 inch, 670 gram box stops being a fixed desktop object and becomes something you drop in a bag between a coffee shop table and a train seat, no monitor, no laptop, just the box and whatever screen is nearby. $599 for the mac mini, a phone most people already carry, and maybe $20 for a mount and cable, that's the entire upgrade, the clip pulled 10000 likes, 4478 saves and 4279 shares before most viewers even clocked they were watching a full desktop replace a laptop. the interesting number here is not the m4 chip or the $599 price tag. it is 0, as in 0 extra screens bought, 0 new monitors carried, 0 desks required, he swapped a full workstation for a phone he already owns and a computer smaller than his hand.

This 17-minute Claude + Obsidian section shows the whole AI second brain workflow from scratch: 22:48 Create your first project with Claude 25:35 Turn one idea into folders, inputs, process, outputs and skills 28:49 Open that project as its own Obsidian vault 29:41 Control the second brain from your phone with Dispatch 34:36 Use the same system to build a real app with paying users 38:58 Watch videos, take notes, export to Obsidian and keep the loop moving Most people use Obsidian as a place to store notes. He uses it as a project operating system where Claude knows the goal, the folder, the skill, the output and the next step. Watch this section, then read the article below on how to turn your own vault into an AI memory loop.

@ch3nweiii do you use the cftc report for other futures besides es, or is it mostly just for index trading signals?

this guy opens the video by saying he has a crystal ball for hedge fund positioning then he pulls up a free government report CFTC has published exactly what hedge funds, asset managers, and leveraged traders hold in every futures market since 1962 Friday. 3:30pm. no login. no paywall Congress mandated it in 1936 so retail could see what big money was doing retail never opened it quant desks read it religiously and built entire books on top of it when leveraged funds flip net short after 4+ weeks net long, something mechanical happens 40,000 contracts start covering someone has to absorb the flow price responds 67% win rate on ES futures tested to 2006 average hold 4 days reward/risk 2.1 not a signal you had to build a government report you were legally supposed to see watch the clip

HE TURNED A MESSY CLOUD DRIVE INTO AN OBSIDIAN MAP, 4 STORAGE LEVELS, 12 FOLDERS, 1 GRAPH VIEW, AND A SYSTEM AI CAN READ WITHOUT DIGGING THROUGH RAW FILES cloud storage looks organized until you actually need something. work, clients, invoices, finance, photos and archive keep sinking deeper until one file sits 3 folders away from where your brain expects it. obsidian changes the shape of the same information. instead of hiding files behind level 1, level 2 and level 3 folders, it shows how clients, docs, people, projects, bugs and decisions connect. for ai, that difference matters. claude does not need 200 random files dumped into context, it needs 5 linked notes, 1 index file and a clear path back to the source of truth. normal cloud storage stores files. obsidian shows relationships, and that is why it becomes far more useful once ai starts reading your work.

@ch3nweiii that bit about citadel doing 10 million a day really puts scale into perspective

this is the Heston model closed-form solution ψ_T(v) = e^(-rT) · φ_T(v-(iα+1)) · S_0^(iv+α+1) / ((α+iv)(α+iv+1)) if you can't read it, that's the point this equation prices a single option under stochastic volatility one option. one strike. one expiration Citadel prices 10 million of these per day Susquehanna does more every options desk on Wall Street runs closed-form solutions like this in production, on GPUs, with sub-millisecond latency retail buys a call because "the chart looks good" the counterparty selling that call already solved a stochastic differential equation to figure out what price to quote you that's not a metaphor. this exact math sits between your buy button and the fill you get Steven Heston published it in 1993 Nobel-adjacent work. free on SSRN. every quant knows it by heart you were never trading against other retail you were trading against a PDE with a name watch the clip

SHE SPOKE ONE PROMPT INTO A POCKET RECORDER AND CLAUDE BUILT A VAULT WITH 4 AUTOMATED PASSES RUNNING EVERY 6 HOURS, FOR LESS THAN THE PRICE OF A COFFEE this is not a voice memo she transcribes later. one spoken prompt goes straight into claude, and claude writes the frontmatter, the topic, the source, and the status field on its own, no manual typing. no dragging files into folders by hand. speak once, and a structure appears that normally takes 30 minutes of clicking, with every note tagged and linked before she even opens obsidian. add the 4 pass loop on top and the vault starts working without her. one pass flags anything untouched for 60 days, one checks old ideas against what performed, one gives a blunt 2-week read, one flags duplicates worded completely differently, all 4 running every 6 hours for less than a coffee a day in api cost. oncees this becomes the habit, a second brain stops needing hands at all. one voice, one prompt, and 4 passes do the filing and the rereading before she even picks up her phone.

@ch3nweiii most retail traders focus on direction, but the big desks just care about implied vs realized vol spread

implied vol on Feb 19, 2020: 11.42% realized vol 30 days later: 95.33% the market was pricing calm what actually landed was one of the biggest moves in history the gap: 83.91 percentage points that number has a name. volatility risk premium it's how every options desk on Wall Street makes money when implied is way below realized, options are cheap. long vol prints when implied is way above realized, options are expensive. short vol prints the scatter in the video is every day of history plotted at once regression: y = -0.73x + 18.5% low implied vol → realized tends to explode high implied vol → realized tends to collapse mean reversion in vol itself. not price. vol Citadel, Susquehanna, Optiver don't ask "is the stock going up" they ask "is this option mispriced against what statistics say it should be worth" one is a bet the other is arbitrage Feb 19 wasn't luck. the math was screaming for weeks retail wasn't reading it watch the clip

HE BUILT A MAP OF HIS OWN BRAIN INSIDE OBSIDIAN, 300 NOTES, 80 LINKS, 15 PROJECTS, 6 MONTHS OF WORK AND ONE GRAPH THAT SHOWS WHAT HIS MEMORY KEPT DROPPING this is not just a pretty animation. every dot can be a book note, saved link, meeting idea, content draft, bug fix, old prompt, or project thought that would normally disappear after 24 hours. after 50 notes, you only have storage. after 150 notes, clusters start forming. after 300 notes, the graph starts showing what you keep returning to, what you keep avoiding, and which ideas are already connected. add claude and it gets stronger. instead of dumping 20 random tabs into one prompt, you can give it 5 linked notes, 1 project folder, 3 old decisions, and let it rebuild the context from your actual thinking. once the vault grows for 30 days, obsidian stops being a notes app. it becomes a memory layer where 100 small thoughts can finally connect into one useful direction.