#rows

Capri My Love #Edgerunners2

in college around 2010 i found a whole website devoted to dragons fucking cars. thought it was really funny and would troll my friends by pulling it up on their computers when they left the room. one day i sat down in class, front row, dead center, opened my laptop, and, well...

My Takeaways From The Miami Dolphins 24-19 Week 8 Loss To The New England Patriots: • Bobby Slowik. Two weeks in a row now he’s been a main culprit in the Dolphins second half slumps. Last week the play calling was too conservative and this week, besides some situational stuff I didn’t like, the offense has NO urgency in the second half until it was too late. This offense doesn’t have a killer instinct under him right now and that needs to change asap, starting with him and his play calling. • I thought Malik Willis was pretty good in the first half and okay in the second half. Had some throwaways he should have made, still not consistently decisive about when to bail on the pocket, his throws gets batted too often at the LOS, had a throw early that was almost picked off by Marcus Jones, the INT called back due to the offside, etc. • Tough sledding on Sunday for Ollie. Took a terrible unsportsmanlike conduct penalty and then continuing to yell and go off on the sideline was not a good look. • The trust is clearly getting there between Bell and Willis, just wish Bell could get the ball in space more • Greg Dulcich had a key, brutal drop that almost resulted in a INT • Nice to see Kevin Coleman Jr. lead the WR group in yardage and with a TD this weekend, had a nasty route breaking some ankles to get open for that TD • I’m just happy Caleb Douglas is okay after getting rolled on his ankle • 2025 Jonah Savaiinaea showed up way too often in the first half of that football game • Kadyn Proctor was CLEARLY missed cause Jarrett Kingston had some rough snaps • Didn’t like seeing Durden put Brewer on his heels like that on the run where Gordon was stuffed for a 3 yard loss • Overall, wasn’t the offensive lines best performance of the season so far, that’s for sure • Chop Robinson is a menace and the best player in Miami’s defensive front right now • Josh Uche was really good against the Patriots • Jordyn Brooks was all over the field making plays • Jacob Rodriguez continues to show why he is the master of the peanut punch • No idea how Chris Johnson whiffed so bad on that sack, all I can think of is he was afraid of getting a flag but either way, absolute brutal though I thought he made that 4th down stop even if the replay original said otherwise • I know Juju Brents recovered the fumble but man he had some tough snaps, especially on the Demario Douglas TD • Speaking of getting cooked by Demario Douglas on his TD, Dante Trader Jr. isn’t playing good enough for me to be comfortable with him as a future starter beyond this season • Michael Taaffe needs to fill out his frame and gain some real strength but either way, the combo of him and Trader Jr. just lacks athleticism too much That was a second half collapse by the Dolphins, no other way to really put it. Now they have to face an emotionally charged Lions coming off a big win and losing one of their best defenders to an awful looking injury. Hey, at least they have the Colts in two weeks. #PhinsUp

HOLY SH*T. i made my first $10,000 with Grok Bot. THANKS ELON for months i kept asking myself one thing: how do i build income that runs without me so i sat down with Grok Bot and we went through dozens of ways to make money most of them needed clients, time or luck the fastest one needed none of that: a trading bot on polymarket so we reverse engineered every crypto bot that prints there why crypto: → a new "Bitcoin up or down" market opens every 5 minutes, 288 a day → the Bitcoin price is public, no insider info needed → every bet settles in 5 minutes, so you see what works fast we spent weeks building the strategy together. this is what we ended up with: → only the 5-minute Bitcoin markets → it waits until about 2 minutes are left → Bitcoin is only $10 from where the window started, yet the losing side already sells for 20-40¢ → it buys that cheap side. it still wins 44% of the time → when one side really pulls ahead, it buys the leader near 50¢ → it rarely pays above 70¢, about 1 buy in 10 → it skips 7 windows out of 10. $14 a bet, up to $170 on the strongest calls 23 days since we switched it on: → +$10,774 → green 20 days out of 21 → worst day: -$31 → 6,692 trades, still running right now honest part: before we put real money in, we ran it on paper for a week and every version bled. it bought the cheap side every single window and the fees ate everything want the same setup? this is how mine is built: → Scout reads the Chainlink price these markets settle on and works out each side's real chance → Trader places the bets on its own, only when a side is 5+ points too cheap after the fee → Risk reads every trade each morning and pauses everything after 3 red days in a row → Grok Bot routines are too slow for 2-minute calls, so Builder wrote the terminal that actually trades. it runs 24/7 on a $6/month server and talks to Polymarket directly paste this into Grok Bot to start: "build me a Polymarket team for the 5-minute "Bitcoin Up or Down" markets: Scout, Trader, Risk and Builder. Scout reads the Chainlink BTC/USD price the markets settle on and works out each side's real chance. Trader flags any side 5+ points too cheap after the fee, never above 70¢, size $15. Risk reviews the log every morning. Builder writes a small terminal for a cheap server so it reacts in seconds. run it on paper first and switch Trader to real bets only after a green week" full Grok Bot setup in the article below, my wallet in the reply ↓ i'm already building the next bot for another polymarket niche. follow so you don't miss the wallet when it goes live

“How couldn’t you guess our costume?”

Barbell Rows 80kgs X 8 How’s the form out of 10? #gym #barbellrow #rows #strength

#MCRLA4 setlist FINAL DEATHWISH BURY ME NA NA SING BEST DAY GIVE EM HELL FOUNDATIONS SAVE YOURSELF MAMA DESOLATION ROW WORLD IS UGLY PARTY POISON WTTBP DESTROYA CALI SCARECROW FAMOUS LAST WORDS HELENA PRISON (with BERT) NOT OK DEMO LOVERS KIDS

NBA STANDINGS UPDATE ‼️ ▪️ DET, PHI move to 5-1 ▪️ BOS wins 4th in a row ▪️ DAL, GSW win 3rd in a row

This is the strangest property of LLM inference: Your LLM's output depends on requests that aren't yours. The same prompt can produce different logits depending on how many other requests share its batch on the GPU. Here's what happens inside the GPU. Most of an LLM's forward pass is spent accumulating sums. Every matmul, RMSNorm, and attention step reduces thousands of floating-point values into a single output, so the order of those additions directly affects the result. The catch is that floating-point addition isn't associative. Every addition rounds the result to the nearest representable value, so (a + b) + c and a + (b + c) can produce different low-order bits. fp16 has an 11-bit significand, so between 1024 and 2048 the spacing between representable values is 1.0. If you compute 1024 + 0.4 + 0.4 left to right, each addition rounds back to 1024. If the two small values are added first, 0.4 + 0.4 gives 0.8, and 1024 + 0.8 rounds to 1025. The order of accumulation depends on batch size, because kernels choose their parallelization strategy to maximize GPU utilization. A GPU like the H100 has 132 compute units, and a small batch doesn't give each of them enough work. So the kernel splits each long sum into chunks, adds the chunks on different units at the same time, and then adds those partial results together. So a different batch size gives a different partitioning, a different accumulation order, and slightly different logits. Usually the difference is negligible, except when two tokens are nearly tied. In a batch of 8: - "Queens" might score 12.3417 - "New" might score 12.3415 In a batch of 32, the rounding might shift them to 12.3414 and 12.3416. If temperature is 0, argmax will pick "Queens" in the first case and "New" in the second. Every later token is conditioned on a different prefix, so the whole completion diverges. So the nondeterminism isn't coming from sampling, but rather from the fact that your request's numerics depend on concurrent traffic you don't control. This is pretty evident at Temperature 0, but the nondeterminism comes from the forward pass, not the sampler. You can fix this by making every reduction batch invariant, meaning each row is reduced in the same fixed order regardless of how many rows share the batch. This costs throughput, because the kernel can no longer select the optimal split for each batch size. It uses one fixed strategy even when that leaves compute units idle. The major inference engines now expose this as a flag: → vLLM has VLLM_BATCH_INVARIANT=1, still in beta. → SGLang has --enable-deterministic-inference, at a 25-45% slowdown. → DeepSeek's DeepInfer has DEEPINFER_BATCH_INVARIANT=1. If you run evals, RL rollouts, or anything that needs reproducible outputs, enable it and treat batch composition as part of your experimental setup. And in production, the batch around your request never stays fixed. With continuous batching, the server adds and removes requests at every decoding step. I wrote an article that explains how it works and why every major inference engine uses it. Read it below.

The original benchmark was misleading and I'm sorry for that. So I deleted it. But allow me to give a proper explanation of what velocitty is actually trying to achieve. Velocitty is a deliberately greedy single-threaded software rasterized terminal that sacrifices correctness for performance. It's a fork of another terminal I've been working on for over a year. I sincerely apologize if I made it seem like it was anything more than that. Here is a short breakdown of how velocitty works and the design decisions I made: Part 1 - IO Input is truncated via a circular buffer. Anything that does not fit into the scrollback is overwritten. This is done either until either EOF is reached (EAGAIN is ignored) or if the screen should refresh (so every 1/hz seconds). Part 2 - Pre-parsing Step 1: use AVX2 to split lines by newline while flagging any line that has either C0 (<= ESC) or C1 (>= DEL). Step 2: lines are split into "runs" (ascii, utf8, csi, osc, kitty, sixel, etc.) This can't be done purely in a single step (i.e. splitting only into runs) because I skip lines that don't have C0 codes at all. I'm that greedy. Only lines that fit on the screen are actually parsed in full. So for a 80 row screen, only the last 80 rows are really parsed. That's why it's so "fast" despite having trash VT parsing (maybe I should have used libghostty lol). Part 3 - Rendering Rendering is nothing fancy just an LRU for what glyphs should be in the atlas and writing to a bitmap. Styles are stored in each cell. It's dumb and greedy but it's my stupidity and greediness, not some AI's! At least let me own my mistakes!

I like to look at claims like this in case I can improve Ghostty. I grabbed your source and immediately spotted the issue: your read loop overwrites a 64 KiB ring. Everything older is discarded unparsed per tick. You're faster because you're not actually terminal emulating. It means the number isn't a throughput number. In a kitten benchmark run at the rate you're claiming, roughly 6.6 MB arrives per 60 Hz tick and 64 KiB of it gets parsed. You're read(2)-ing 100% of the data and parsing about 1% of it. There are other minor issues (a small fixed scrollback in a preallocated ring buffer, no grapheme clustering, throwing away data like OSC 8 URIs, etc.), but the major one is simply that you're discarding ~99% of the input. For plain text, e.g. your cat benchmark, the final screen ends up looking right, but the scrollback is missing 99% of the file and each retained chunk starts mid-line. As soon as there are escape sequences in the dropped spans, your emulator is fully out of sync and invalid: a cut inside a CSI prints the parameters as text, and any SGR, mode, title, or alt-screen change in the gap is just... gone. FYI, this is why I don't like when people point at refterm as a terminal benchmark. refterm implements a TINY subset of what a terminal emulator must support, it's closer to a fast file reader, not a terminal emulator. I also don't mean any shade by this. I like to look at technology objectively to try to learn something and improve my own craft and work. So I'm sharing what I found. I modified your code very slightly to parse on every full buffer (rather than per tick). And the results are attached. Both use main commit in each repo at the time of this post. Note the graph is in log scale, the discrepancy between performance for things like Kitty images is measured in GB/s.

Appreciate the follow-up, and no apology needed for building something. The only issue was the characterization of the comparison. I'm sorry it got so much attention! I gave the feedback in exactly the same tone and style I would if someone I worked with professionally (at any level) came to me and showed me the work and made the same claims. I looked at the way you process/drop data in your circbuffer and VT parser and I was suspicious of HOW much you want to sacrifice correctness, so I want to be specific about what "sacrifices correctness" means here, because I think it's more than it sounds. I'm NOT trying to "mog you" or be mean here, I'm just trying to point out some things you might find helpful, and perhaps help you learn something (if you want). - A 120 KB plain text file: 137 of 300 lines never exist (e.g. your dropping data that technically fits in your scrollback), and the first visible row is a mid-line cut-off fragment. - Same file preceded by CSI ?1049h (the alt-screen switch) is dropped, so all 163 surviving rows land on the primary screen and in scrollback. - A cut inside a truecolor SGR prints `;2;226;54;22m` as text. Similarly, a cut inside an OSC 8 prints the URI as text. The threshold for this isn't a massive firehose. It's any output over 64 KiB between frames: a `git diff`, a stack trace, `ls -R`, a verbose build. That's ordinary terminal use, almost completely ordinary text. A terminal is a stateful protocol, as you hopefully know. If a mode change is in the dropped span (alt screen, bracketed paste, mouse tracking, kitty keyboard flags, a scroll region), the app and the terminal disagree for the rest of the session and nothing resyncs them short of a full RIS. The line-based skipping also can't be made correct even in principle, because terminal state isn't line-local. SGR persists across lines, cursor position depends on every CR/BS/tab/wide char before it, and full-screen apps don't emit newlines at all, they use CUP. I repeat: none of this is shade. "Greedy, drops data, trades correctness for speed" is fine. It's just a different category from a terminal emulator, and I would highly recommend you do not call it that. And again, I hope you understand especially at this point, that I'm giving this feedback in an earnest way attempting to educate and help. I would give the feedback the same way to anyone who works with me. Good luck with your project.

Happy Halloween! 👻

Even the birds got front-row seats to the rhinos’ private date

#Battlefield6 'Casino Heist' begins November 10th. Overview 🗺️ New Map: European Casino, an infantry-only map set on the Las Vegas Strip. 💰 Limited-Time Event: The Casino Heist event allows you to earn currency to unlock rewards. 🚛 New Hardware: The NSW4 Carbine, the Signal Jammer, and a new MRAS truck. 🔫 Systems & Quality of Life: Weapon adjustments, Multiplayer Leaderboards, Proximity Chat in Rapid Royale, and Battle Royale mission clarity. European Casino 🔹 Goes live on November 10th. 🔹 Night map. 🔹 Infantry-only combat inside a large, European-style resort. 🔹 Engagements move quickly from open courtyards into packed gaming floors and narrow hallways. 🔹 Blow through breakable walls to carve out flank paths or just play in the lazy river. 🔹 The massive center tower can come crashing down to permanently change the flow of the map. 🔹 It has a similar flow to Operation Metro from Battlefield 3. 'Casino Heist' Limited-Time Event 🔹 Goes live on November 10th. 🔹 Defeated enemies will drop Contracts that give you additional objectives to complete during your matches, from capturing points and reviving teammates to exploring the European Casino map. 🔹 Completing Contracts rewards you with DOUGH, the event’s currency, which you can exchange for a collection of cosmetics including weapon packages, vehicles skins, and an all-new Heist Mask soldier skins. 🔹 Claiming every standard item on the reward track will also unlock one final challenge for a bonus completion reward. New Hardware 🔹 The NSW4 carbine offers rapid handling and comes equipped with subsonic rounds by default, alongside a unique Supersonic ammo option. 🔹 The deployable ADS-17 6GHz (Signal Jammer) gadget for the recon class jams enemy radars and drastically reduces the range at which opposing players can share active spotting information. 🔹 The M142 MRAS is a highly mobile truck that provides devastating indirect fire on a fast, flexible platform. Weapon Changes on the Upcoming Update 🔹 Extended Magazines Adjustments: High-capacity extended magazines on the SCW-10 and KV9 are receiving handling penalties. Equipping extended magazines should involve a clear trade-off between sustained firing capacity and immediate weapon agility. 🔹 More Magazine Options: Additional magazine capacities and feeding options are being introduced across multiple weapon types to give players more flexibility in pacing engagements. These include some LMGs, Shotguns, and more. 🔹 Underutilized Attachment Buffs: DICE is planning buffs to several historically overlooked muzzle and magazine attachments, including the Flared Magwell, Frangible Ammo, and more, giving you more viable, competitive options in the gunsmith. 🔹 M44 Attachments: For those who like a side of flair with their sidearms, the team is testing some highly requested, custom optic attachments for the M44. Because sometimes, you just want to put an 8x scope on a hand-cannon for the memes. Multiplayer Leaderboards 🔹 Players will be able to browse leaderboards across dedicated tabs for Career, Game Modes, and Classes. 🔹 The Leaderboards are seasonal. DICE has included a dedicated Seasonal Filter so you can easily compare your standings from Season 5 onward. 🔹 They have a 'Sticky Player Row.' This ensures that you can always see your exact overall standing across any category or season, even if you are still working your way up toward the top ranks. Updates in REDSEC 🔹 Proximity chat is coming to Rapid Royale for this upcoming patch. 🔹 DICE is updating the in-game HUD and objective text to make tracking your active Battle Royale contracts and missions much simpler during chaotic firefights. 🔹 This update also includes clearer timers, better indicators, and more obvious notifications for objectives. Vehicle and Game Audio 🔹 DICE is tracking vehicle survival times, engagement win rates, and armor turnover across all game modes before deciding on further tuning. 🔹 The latest update delivered initial improvements targeting footstep audio, but their work here is not done. Portal Updates 🔹 Builders can now choose which skin variation is used when spawning a Custom Portal Bot with the SpawnAIFromAISpawner action. The six existing class-specific skins are now individually selectable, rather than being chosen at random. 🔹 The Marked Soldier is being added as a seventh option. This bloody and scarred soldier can now make an appearance in any Portal experience you create. 🔹 They're adding several new effects to Portal, including visual and audio effects pulled directly from the Halloween Dead Space experience: ▪️ 'NecromorphEffects' allow you to transform soldiers into Necromorphs. Unlike their original appearance, you're not limited to Bots or a specific team in Portal; you decide who gets transformed. ▪️ 'ShadowSilhouette' turns a soldier completely black and can be combined with the Necromorph effect or used independently. 🔹 They're also opening up several of the audio effects used to create hallucinations and atmosphere. Builders can make individual players hear whispers, monster noises, a surreal ambient drone, or even fake footsteps and gunfire. These sounds are only heard by the affected player, opening up new possibilities for player-specific warnings, hallucinations, misdirection, or whatever else you can come up with. 0 New Screen Effects 🔹 Season 5 adds 10 new Screen Effects that can be enabled or disabled for individual players. These include Color Correction, Depth of Field, Film Grain, Filmic Effects, Fog, Motion Blur, Tone Mapping, Vignette, White Balance, and even Unitology Markers from the Halloween experience: ▪️ 'SetScreenEffectParameter' exposes 121 new parameters that creators can experiment with, allowing you to fine-tune values and control individual parts of the Screen Effects. NOTE: Some parameters may have a performance impact, particularly when frequently changing a large number of them, so DICE encourages creators to experiment and make use of the available performance tools while testing their experiences.

Day 1 of Learning SQL 💻 Today, I started my SQL learning journey and here’s what I learnt: - What SQL is and why it is important in data analytics. - What databases are and how they store information. - What tables are and how they are organised. - The difference between rows and columns. - What records mean in a database. How SQL helps data analysts retrieve, manage and work with data. I also learnt that every row represents a record, while columns represent the different attributes of that record. #30DaysOfSQL #DataAnalytics #SQL #LearningInPublic

What the hell is going on at Clemson??🫪

I need heavier DBs #rows

French police officers were literally engulfed in flames today as Paris exploded into chaos. What started as a march for 70,000 citizens demanding relief from crushing inflation was completely hijacked by masked rioters armed with hammers and Molotov cocktails. At Porte de Charenton, these extremists tore up the streets and threw incendiary devices directly into police lines, turning a cost-of-living protest into a violent assault. Law enforcement had to unleash tear gas, stun grenades, and water cannons just to push back the mob and retake the outer boulevards. This is the third weekend in a row that France has been paralyzed by these massive, volatile demonstrations. The government tried to ban the march from central landmarks to prevent looting, but changing the route didn't stop the wreckage. How much longer can a society stand by while its cities burn every single weekend?

Front row seats, obviously

🚨 SOMETHING VERY, VERY BAD IS HAPPENING IN CHINA. China STAR50 → -40%. AI stocks → $480 BILLION wiped out. U.S. Treasury holdings → 18-YEAR LOW. Gold purchases → 23 MONTHS IN A ROW. China AI stocks are CRASHING. But while everyone is watching the sell-off, Beijing is making a MUCH BIGGER move. China's U.S. Treasury holdings have fallen by over $100 BILLION in just ONE YEAR. Meanwhile, its central bank keeps accumulating GOLD. And this is what 99% of investors are MISSING: China isn't just diversifying its reserves. It's building the infrastructure to REDUCE its dependence on the U.S. DOLLAR. Look at what's ACTUALLY happening: → July 2026: Hong Kong launches a new GOLD clearing system. → Direct physical gold delivery links with Shanghai begin. → China expands gold trading and settlement infrastructure. → U.S. Treasury exposure continues to decline. → Gold becomes increasingly important to its reserve strategy. See the bigger picture? Less U.S. Debt → More Gold → Alternative Settlement Networks → Less Dollar Dependence. This is NOT just about the AI bubble. It's about a potential shift in how the global financial system operates. And here's the part that should concern EVERY investor: If this trend accelerates, demand for U.S. debt, dollar liquidity, and global capital flows could all be affected. The consequences could reach STOCKS, GOLD, BITCOIN, and the entire financial system. Remember: Major financial shifts often begin LONG before the public recognizes them. And central banks are already repositioning. SAVE THIS CHART. I've spent 10 years studying macro cycles and major market turning points. Follow and turn on notifications. The next 12 months could be CRITICAL.

🚨 THE 18-YEAR CYCLE SAYS WE’RE AT THE PEAK! Look at the top chart. Inflation-adjusted home prices going back to 1890. For over a century, prices oscillated around the same baseline. Then two things happened that broke the pattern: → 2008 bubble: peaked near 265 → 2026 bubble: peaked above 300 Read that again. We’re now higher than 2008, in real terms. The bubble that nearly took down the global financial system was smaller than what we’re sitting on right now. And the bottom chart explains why the timing isn’t a coincidence. The 18-year land cycle has repeated for generations. Same structure every time: Recovery → mid-term peak → mid-term recession → expansion → winner’s curse → cycle peak → deep recession Look at the peak years: 1954, 1972, 1990, 2008, 2026. Every one of them was followed by a deep recession. 1955-58. 1973-76. 1991-94. 2002-12. And the chart projects the next one: 2027-2030. We are standing exactly on the “we are here” marker. Right at the top of the expansion phase, in what the cycle calls the winner’s curse - that final stretch where prices feel unstoppable, everyone piles in, and the last buyers pay the highest price in history. That’s where we are. Not early. Not mid-cycle. At the top. Now stack today’s reality on it: record seller-to-buyer imbalance, mortgage rates above 7%, zero price discovery because nobody can afford to move. The cycle has never missed. Not once in 130 years. Save this chart. You’ll want it in 2027!

#MCRLA4 setlist DEATHWISH BURY ME NA NA SING BEST DAY GIVE EM HELL FOUNDATIONS SAVE YOURSELF MAMA DESOLATION ROW WORLD IS UGLY PARTY POISON WTTBP DESTROYA CALI SCARECROW FAMOUS LAST WORDS HELENA PRISON (with BERT) NOT OK DEMO LOVERS KIDS