Will Estes

This tweet is awesome bc the poster managed to trigger an entire state by claiming that our country’s only legitimate business destinations are delta hub cities- and they fell for it

@mccauley318 Heat

Paul Wall & Chamillionaire

@HoodFamousTV_ RED ROLLS WHITE CEILING

Multiple women have exposed their DMs with The Game where he uses the same pickup line😭😭😭

@aakashgupta This is so dumb bro

Blockbuster made $800 million a year charging late fees. Customers hated it so much the company went bankrupt. Two indie developers just made a game where you charge those same late fees, and it launched with a 99% positive review score. The difference is one word: consent. Handing a late fee to an NPC is play. Getting charged $4 for returning Titanic two days late was punishment for enjoying a Friday night ritual you loved. The browse. The wall of new releases. The kid begging for candy at the counter. That experience was Blockbuster’s actual product. The late fee was a tax on it. In 2000, Netflix offered to sell itself to Blockbuster for $50 million. Blockbuster’s CEO laughed them out of the room. By September 2010, Blockbuster’s 9,094 stores were worth $24 million combined. Netflix is worth $400 billion today. The reason is the same reason this game works. Blockbuster’s management looked at the P&L and saw late fees as a revenue line. They never saw them as the compound interest on customer resentment. $800 million a year in recurring hostility. When Netflix offered the same movies with no punishment, the switch was instant. 84,300 employees. 9,094 locations. Gone. Meanwhile, two developers at Blood Pact Studios built the part Blockbuster accidentally threw away. The Friday night ritual, the shelving, the customer interactions, the tape rewinding. Simulation games now account for 9.76% of all Steam revenue. Job simulators alone have generated $1.36 billion lifetime. The shop sim is the single most predictable path to indie success on the platform. Retro Rewind hit #5 top seller on Steam on launch day with zero marketing budget. Blockbuster had $6 billion in annual revenue and couldn’t survive the thing two people just turned into a $16 game. The movie was never the product. The store was.

@sweeterpool Instant bookmark

Im the only one wearing Rick Owens at this HIV clinic

This point would be stronger if the historical encoding feature were consistent. English is better described as a Frankenstein’s monster- Norse and Latinate appendages sewn into a Germanic base. Modern spelling & usage reflect not a uniform historical logic but rather a random walk. None of which makes it any less beautiful.

Noam Chomsky once called English spelling "a near optimal system." You might think he was being ironic. Far from it. The silent 'b' in "bomb" reappears in "bombard." The silent 'n' in "hymn" is pronounced once again in "hymnal." The silent 'g' in "sign" comes back in "signal." English spelling keeps these words looking like the family they are, even when pronunciation pulls them apart. The past tense ending "-ed" is pronounced three different ways (-t in "jumped," -d in "played," and -ed in "painted"), but spelled the same every time. One spelling, one meaning: something happened in the past. English spelling is full of inconsistencies and silent letter because it’s not simply encoding how words sound. If English spelling were aiming to represent sound alone, it would indeed be a total failure. But that's not the kind of system English has. It encodes words' meaning and history as well.

@CanesDavid You’re right that sounds different… frustrating it isn’t more useful

Is it possible you’re describing a different AI feature? I spoke to the residents about this and indeed on the inpatient side there’s a discharge summary AI helper tool which they agree is very useful. Gives summary narratives. This is different. This is a module on the outpatient side, which is supposed to summarize what has happened since I last saw the patient as an outpatient. Bullets only, no narrative, no rhyme or reason to what gets elevated to a bullet.

First experience with EPIC’s AI Summaries “Generate Summary from Notes” This was early access in my health system in the outpatient setting, where it can generate a summary of what's gone on with the patient since my last visit. I'm overall pleased that Epic is headed in this direction, but so far my first impression was that it falls short in several major ways: 1. The latency essentially makes it unusable for me. It takes 40 seconds plus to generate a summary. 2. Summaries can only “see” notes. They don't see labs, they don't see imaging, and for a subspecialist that's severely limiting. Context is everything and that’s just not enough context. 3. The summary ends up showing me random snippets from my own previous note, and it's unclear how those are prioritized. They’re typically not the most important sentences. In the 40 seconds it took to render, I'm best off just pulling up my last note and reading it myself. Not quite ready for prime time, but a step in the right direction. If you have a clinician-facing AI product, you need to think very carefully about the user intent. And it should be FAST. It cannot slow people down. Has anyone tried this and found it useful?

Guess I’m in the minority here but I love it, esp as it’s improved enough to reliably summarize even a 20-30 day LOS. The latency gripe confuses me- I just flip to the next chart & keep rolling. If I have multiple discharges I keep a few of them cooking simultaneously while I work in another chart. DC summaries have gone from most annoying to my easiest notes. I agree that it could include more data (though this would make latency even worse). “Garbage in, garbage out” definitely applies; it requires a careful reading before clicking sign. But for what it is- a note summarizer to speed up writing dc summaries- it’s been a solid QOL upgrade.

Epic will roll out this unusable tool and all hospital systems will say "we don't need to buy a separate AI summary tool, Epic has one!"

@GriffinHilly @le_polisson @giulio_mattioli Elite explanation thanks for this

@le_polisson @giulio_mattioli In the linear-scaled plot you observe heteroskedasticity, which is commonly corrected with log-log comparison share.google/aimode/waerbLB…

You know that graph plotting (logged) GDP against (logged) energy consumption per capita, highlighting the bottom right corner and saying that "there is no such thing as a low-energy rich country"? (done my own version below) There's a few problems with it... (THREAD)

@HuggingModels How can physicians implement this usefully though? Is this HIPAA compliant? Not sure how I could feed clinical info into it to get useful output without revealing PHI. Also doubt it could run/install on hospital enterprise systems which have tight restrictions..

Meet ClinicalBERT: a specialized AI that understands medical language like a pro. This isn't your average language model. It's been trained on clinical notes and medical literature, making it uniquely equipped to handle healthcare terminology. Think of it as a medical translator for AI systems.

@AgenticReadyHQ @HedgieMarkets Isn’t it still a concern that this new infrastructure may become obsolete within a more limited time horizon than previous infrastructure projects like rail or fiber? I want to believe the bulls but tough to ignore the risk

The spending numbers are staggering, but this looks less like pure speculation and more like a front loaded infrastructure cycle. We saw the same pattern with railroads, electricity, and the internet. Capital overshoots demand, prices spike, labor reallocates, then excess capacity gets absorbed over a longer horizon. The real risk is not that AI is useless, but that returns accrue unevenly and too slowly for the balance sheets funding it. The question is not whether the infrastructure gets used. It is who survives long enough to benefit when it does.

🦔 The AI boom is causing shortages across the economy. Five companies (Amazon, Google, Microsoft, Meta, Oracle) are on track to spend $700 billion this year on AI infrastructure, nearly double what they spent in 2025 and equal to three-quarters of the US military budget. Electricians are getting harder to find as data centers absorb skilled tradespeople. Construction projects unrelated to AI are being delayed. Memory chip prices are up 60% since September, pushing smartphone and PC prices higher for potentially years. Apple told investors it's having trouble buying enough chips for iPhones. Promising startups outside AI are being starved of funding, with the "middle class" of startups at a decade low for investment. JPMorgan calculated the tech industry needs an extra $650 billion in annual revenue just to earn a reasonable return on AI investment. OpenAI expects to lose more than $100 billion through the end of the decade. My Take I've been writing about the AI bubble for months but this puts the scale into perspective. An early Facebook investor said the amount invested in AI since 2022 may exceed all prior investments in the entire tech industry combined. Data center construction spending rose 32% last year while other commercial real estate is flat or declining. OpenAI told the White House its planned data centers would require 20% of the existing skilled tradespeople workforce. Your next phone costs more because data centers are buying all the memory chips. Housing projects are delayed because electricians make more money wiring server farms. This is what it looks like when an entire economy bets on one technology. If AI delivers on the promise, maybe the disruption is worth it. But if the returns don't materialize, and there's growing evidence they won't at this scale, the losses won't stay contained to tech companies. Everyone paying more for electronics, every construction project that got shelved, every startup that couldn't get funded because they weren't AI, that's the cost we're all absorbing right now for a bet that hasn't paid off yet. I keep thinking about the dot-com bust. The infrastructure eventually got used, but that didn't help the people and companies that got wiped out waiting for demand to catch up to capacity. Hedgie🤗

@alifmunim I’m an IM hospitalist, would it be possible at some point to pair this with POCUS technology like a phone-connected US probe? I’d be fascinated to give this a try at the bedside, assuming no privacy concerns.

We trained a foundation model on 18 million heart ultrasound videos to predict structure instead of pixels. Introducing EchoJEPA, the first foundation-scale JEPA for medical video. Paper: arxiv.org/abs/2602.02603 Code: github.com/bowang-lab/Ech… 🧵 1/n

@elaifresh I unironically really wanna sign up for Spotdaddy

Now That’s! What I Call Vibecoding

@mattyglesias Bruh that’s literally zero protein

@kendramiddleton In the last 3 seasons this team has 31 regular season wins and 2 playoff wins. Now they’re back in the playoffs playing better than ever & lead the league in scoring defense.

I’d just like to point out that the Texans have never even been to a conference championship game but everybody is doing tricks on it for them

@wannabejake9 Bruh AFCS has 2 teams founded after 1995, that’s most in the league

This gotta be the most poverty division ever holy shit 😭😭

@NoahMF Texas has at least 5 major professional sports franchises (excluding houston) vs how many for AR? Love htown but this ain’t it bro

Houston without Texas would be Singapore. Texas without Houston would be Arkansas.

@JillFilipovic Enjoy ur engagement

My hottest history take is that no one, including historians, can really explain what caused WWI.

@Bachscore ChatGPT’s finest work

Let's all enjoy again the college football comment of the year. 10/10, no notes 🏆