Загрузка видео...

Не удалось загрузить видео

Возникла проблема при загрузке этого видео. Это может быть связано с временными проблемами сети или видео может быть недоступно.

На главную

Two 𝗧𝗢𝗨𝗖𝗛𝗗𝗢𝗪𝗡𝗦 on UCLA's first two offensive snaps! 🏈 UCLA 14, NCCU 0 13:46 – 1st quarter 📺: P12LA 📲: #GoBruins

UCLA Football

160,613 subscribers

19,005 просмотров • 3 лет назад •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

.Jalen Berger finds the endzone for the first time this season! 🙌 UCLA 10, Michigan State 7 1st quarter – 1:39 📺: Big Ten Network 📲: #GoBruins

.Jalen Berger finds the endzone for the first time this season! 🙌 UCLA 10, Michigan State 7 1st quarter – 1:39 📺: Big Ten Network 📲: #GoBruins

UCLA Football

53,088 просмотров • 10 месяцев назад

North Gwinnett has scored TDs on their first two drives. Then a little trickery for two. 15-0 over Grayson in the 1st Q. Corky Kell + Dave Hunter Classic

North Gwinnett has scored TDs on their first two drives. Then a little trickery for two. 15-0 over Grayson in the 1st Q. Corky Kell + Dave Hunter Classic

ChadSimmons

26,488 просмотров • 18 дней назад

Division II State Semifinal 🌟 2Q | 5:01 Avon - 14 Archbishop Hoban - 0 Avon extends its lead to two possessions on a 5-yard TD run by Cameron Wendell. #NEOFB 🏈

Division II State Semifinal 🌟 2Q | 5:01 Avon - 14 Archbishop Hoban - 0 Avon extends its lead to two possessions on a 5-yard TD run by Cameron Wendell. #NEOFB 🏈

NEO Zone HS Sports

22,510 просмотров • 1 год назад

Auburn LB commit Jaquez Wilkes with a big 39-yard run on just the second Wadley offensive play of the game… which came with two minutes left in the first quarter Wadley marches down and scores to cut the Fayetteville lead to 21-8

Auburn LB commit Jaquez Wilkes with a big 39-yard run on just the second Wadley offensive play of the game… which came with two minutes left in the first quarter Wadley marches down and scores to cut the Fayetteville lead to 21-8

Christian Clemente

19,645 просмотров • 11 месяцев назад

#AVCContinental: WADA IS ON FIRE 🇯🇵🔥 Yukiko Wada is up to 14 points after two sets, leading all scorers as Japan take a 2-0 lead over Chinese Taipei 🤩 One more set for a spot in the semifinals 👀 📺 LIVE NOW on VBTV: 🏐 #Tianjin2026 #Volleyball

#AVCContinental: WADA IS ON FIRE 🇯🇵🔥 Yukiko Wada is up to 14 points after two sets, leading all scorers as Japan take a 2-0 lead over Chinese Taipei 🤩 One more set for a spot in the semifinals 👀 📺 LIVE NOW on VBTV: 🏐 #Tianjin2026 #Volleyball

Volleyball World

43,960 просмотров • 4 дней назад

my college roommate got fired from Two Sigma last month he was making $480K/year building statistical arbitrage models instead of interviewing at other funds - he ported his entire system $800 → $127,400 in 5 weeks total cost: $0 he said: "the alpha I found for Two Sigma works better here, because at hedge funds you fight decay on $60B AUM. On Polymarket you compound edge on $800" he deployed 5-factor model overnight same math Two Sigma ran on client billions: Deep Wallet Analysis: mean reversion detector - identifies wallets that fade extremes, copies their entries Priority Mode: co-location wasn't advantage at Two Sigma, it is here - 0-2 second fills vs 14 second retail lag Correlation scanner: 200+ market pairs monitored - when linkage breaks >12%, one is mispriced Volatility regime: hidden Markov model switches strategy every hour - bull/bear/sideways auto-detected Position sizing: information ratio optimization, max 6% per trade, portfolio variance targeting all 5 factors score simultaneously ensemble combines signals Kelly optimal, max 6% allocation, 15 positions max Two Sigma's flagship fund returned 12% last year on $60B his bot returned 15,825% in 5 weeks on $800 former PM asked him to come back at $600K he said he makes that monthly now deploy tonight: deposit $50, enable 5-factor system: first trades execute within 24-48 hours doesn't need their carry anymore

my college roommate got fired from Two Sigma last month he was making $480K/year building statistical arbitrage models instead of interviewing at other funds - he ported his entire system $800 → $127,400 in 5 weeks total cost: $0 he said: "the alpha I found for Two Sigma works better here, because at hedge funds you fight decay on $60B AUM. On Polymarket you compound edge on $800" he deployed 5-factor model overnight same math Two Sigma ran on client billions: Deep Wallet Analysis: mean reversion detector - identifies wallets that fade extremes, copies their entries Priority Mode: co-location wasn't advantage at Two Sigma, it is here - 0-2 second fills vs 14 second retail lag Correlation scanner: 200+ market pairs monitored - when linkage breaks >12%, one is mispriced Volatility regime: hidden Markov model switches strategy every hour - bull/bear/sideways auto-detected Position sizing: information ratio optimization, max 6% per trade, portfolio variance targeting all 5 factors score simultaneously ensemble combines signals Kelly optimal, max 6% allocation, 15 positions max Two Sigma's flagship fund returned 12% last year on $60B his bot returned 15,825% in 5 weeks on $800 former PM asked him to come back at $600K he said he makes that monthly now deploy tonight: deposit $50, enable 5-factor system: first trades execute within 24-48 hours doesn't need their carry anymore

bl888m

18,118 просмотров • 4 месяцев назад

> be IO Interactive > spend years finally landing the James Bond license > first console Bond in 14 years, first under Amazon ownership > cast a young Bond. fans say he looks smug > respond with "if no one has opinions on him, he'd be the dullest Bond ever made" > let physical copies leak days before launch > issue a copyright strike on the leak > then upload the same 13 minutes yourself the next day > watch Kotaku call your opening "boring men in a boring helicopter" > two days before launch, drop the big cast reveal > "you will meet a familiar face in Vietnam" > it's Khaby Lame, the silent TikTok guy ICE self-deported last summer for overstaying his visa Bond has a license to kill. his new partner couldn't keep his license to stay.

> be IO Interactive > spend years finally landing the James Bond license > first console Bond in 14 years, first under Amazon ownership > cast a young Bond. fans say he looks smug > respond with "if no one has opinions on him, he'd be the dullest Bond ever made" > let physical copies leak days before launch > issue a copyright strike on the leak > then upload the same 13 minutes yourself the next day > watch Kotaku call your opening "boring men in a boring helicopter" > two days before launch, drop the big cast reveal > "you will meet a familiar face in Vietnam" > it's Khaby Lame, the silent TikTok guy ICE self-deported last summer for overstaying his visa Bond has a license to kill. his new partner couldn't keep his license to stay.

frostzy

506,679 просмотров • 3 месяцев назад

240423 Gentle Monster IG Update Introducing JENTLE SALON 🦄☁️ Imagined by Gentle Monster and Jennie. The two creative forces present a collaborative collection that transcends the realm of eyewear through its eye-catching charm designs, encouraging a diverse expression of style. Get notified at the link in bio now. Launching on 5.1 Experience Jennie’s captivating touch on the unique collection through special Jentle Salon Pop-ups in 13 global cities: New York 45 Grand Street, New York May 1st–May 12th Costa Mesa South Coast Plaza 1F, South Coast Plaza 3333 Bristol St, Costa Mesa June 1st–June 20th Seoul Haus Dosan 50 Apgujeong-ro 46-gil, Gangnam-gu, Seoul May 1st–June 30th Tokyo Gentle Monster Aoyama 5 Chome-3-2 Minamiaoyama, Minato City, Tokyo May 1st–June 30th Osaka Hankyu Umeda CIRCUS, 3F, 8-7 Kakudacho, Osakashi, Osakafu May 1st–May 29th Shanghai Haus Nowhere Shanghai 798-812, Mid-HuaiHai Rd., Huangpu District, Shanghai May 1st–End of June Beijing Gentle Monster Beijing Sanlitun Taikoo Li 1F S10-15, 2F S10-23 & 3F S10-32, South Zone, Taikoo Li Sanlitun, 19# Sanlitun Rd, Chaoyang District, Beijing May 1st–End of June Hong Kong Harbour City Atrium II, Harbour City, 3-27 Canton Rd, Tsim Sha Tsui, Hong Kong June 28th–July 11th Bangkok Siam Paragon Jewel zone, M floor, 991/1 Rama I Rd, Pathum Wan, Bangkok May 1st–May 31st Singapore Gentle Monster Singapore MBS Marina Bay Sands, 2 Bayfront Avenue, B2 - 103/104/105, Singapore 018972 May 1st–June 30th Manila Gentle Monster Manila Shangri-La Shangri-La at The Fort, Manila, 5th Ave, Taguig, 1634 Metro Manila May 1st–June 30th Taipei Gentle Monster Taipei Breeze Nanshan Shop F1-04-06, 1/F, Breeze Nanshan, No.17, Songzhi Rd., Xinyi Dist., Taipei City TO BE UPDATED Kuala Lumpur The Exchange TRX Ground Floor, Persiaran TRX, 55188 Kuala Lumpur July 5th–July 28th 🔗

240423 Gentle Monster IG Update Introducing JENTLE SALON 🦄☁️ Imagined by Gentle Monster and Jennie. The two creative forces present a collaborative collection that transcends the realm of eyewear through its eye-catching charm designs, encouraging a diverse expression of style. Get notified at the link in bio now. Launching on 5.1 Experience Jennie’s captivating touch on the unique collection through special Jentle Salon Pop-ups in 13 global cities: New York 45 Grand Street, New York May 1st–May 12th Costa Mesa South Coast Plaza 1F, South Coast Plaza 3333 Bristol St, Costa Mesa June 1st–June 20th Seoul Haus Dosan 50 Apgujeong-ro 46-gil, Gangnam-gu, Seoul May 1st–June 30th Tokyo Gentle Monster Aoyama 5 Chome-3-2 Minamiaoyama, Minato City, Tokyo May 1st–June 30th Osaka Hankyu Umeda CIRCUS, 3F, 8-7 Kakudacho, Osakashi, Osakafu May 1st–May 29th Shanghai Haus Nowhere Shanghai 798-812, Mid-HuaiHai Rd., Huangpu District, Shanghai May 1st–End of June Beijing Gentle Monster Beijing Sanlitun Taikoo Li 1F S10-15, 2F S10-23 & 3F S10-32, South Zone, Taikoo Li Sanlitun, 19# Sanlitun Rd, Chaoyang District, Beijing May 1st–End of June Hong Kong Harbour City Atrium II, Harbour City, 3-27 Canton Rd, Tsim Sha Tsui, Hong Kong June 28th–July 11th Bangkok Siam Paragon Jewel zone, M floor, 991/1 Rama I Rd, Pathum Wan, Bangkok May 1st–May 31st Singapore Gentle Monster Singapore MBS Marina Bay Sands, 2 Bayfront Avenue, B2 - 103/104/105, Singapore 018972 May 1st–June 30th Manila Gentle Monster Manila Shangri-La Shangri-La at The Fort, Manila, 5th Ave, Taguig, 1634 Metro Manila May 1st–June 30th Taipei Gentle Monster Taipei Breeze Nanshan Shop F1-04-06, 1/F, Breeze Nanshan, No.17, Songzhi Rd., Xinyi Dist., Taipei City TO BE UPDATED Kuala Lumpur The Exchange TRX Ground Floor, Persiaran TRX, 55188 Kuala Lumpur July 5th–July 28th 🔗

DC 제니 갤러리

83,206 просмотров • 2 лет назад

The 49ers just finished their first 2-block practice with pads... 🏈Pass game looked sharp — until today. Perhaps some of the WR tumult hurt the operation. Ricky Pearsall is obviously done and the 49ers held De'Zhaun Stribling out Saturday for precautionary purposes 🏈CMC appears to be aging in reverse, possibly a testament to the distance he's getting from the 2024 Achilles tendinitis. He's had a ton of juice and ripped another long cutback run today 🏈LG will be a concern until it's not. So far in 1-on-1s, Brett Toth has seen more pass-protecting success than Connor Colby and Robert Jones, both of whom will need to improve there to allow the 49ers to breathe easier. But the 49ers prefer for Toth to be a backup center who isn't exposed to injury at the same time as Jake Brendel. Maybe he'll be a "break glass in case of emergency" option 🏈The much-hyped new looks of Raheem Morris' defense are a real thing — the 49ers have indeed expanded their repertoire both in terms of fronts and coverages. Their DBs made a lot of plays Saturday — PBUs for Lenoir, Green, and Stout 🏈On that note, Ji'Ayir Brown does seem to be winning favor as a Morris "guy". The DC specifically mentioned Brown today as he was rattling off names of supposed defensive pillars. Would be surprising if Brown isn't a Week 1 starter, IMO 🏈The DE injury concern is real, if only for the 49ers to conduct a proper training camp that includes two joint practices with the Titans and Chargers. They need more bodies there now and they were interested in adding one more vet even before the latest hardships. Jadeveon Clowney and Kyle Van Noy are both still available. Joey Bosa also technically is, but it seems that (if he isn't retiring) he wants to take the Eric Kendricks route of chilling now and signing later. Keep an eye on 49ers' workouts and signings at DE to get through the current storm. Maybe they sign a placeholder or two now and a more recognizable name later? We will see 🏈Deebo Day was fun. Of note that the 49ers wanted him even before Pearsall went down. So they've added — at the very, very least — a return man like they did with Skyy Moore last year, but with way more offensive upside

The 49ers just finished their first 2-block practice with pads... 🏈Pass game looked sharp — until today. Perhaps some of the WR tumult hurt the operation. Ricky Pearsall is obviously done and the 49ers held De'Zhaun Stribling out Saturday for precautionary purposes 🏈CMC appears to be aging in reverse, possibly a testament to the distance he's getting from the 2024 Achilles tendinitis. He's had a ton of juice and ripped another long cutback run today 🏈LG will be a concern until it's not. So far in 1-on-1s, Brett Toth has seen more pass-protecting success than Connor Colby and Robert Jones, both of whom will need to improve there to allow the 49ers to breathe easier. But the 49ers prefer for Toth to be a backup center who isn't exposed to injury at the same time as Jake Brendel. Maybe he'll be a "break glass in case of emergency" option 🏈The much-hyped new looks of Raheem Morris' defense are a real thing — the 49ers have indeed expanded their repertoire both in terms of fronts and coverages. Their DBs made a lot of plays Saturday — PBUs for Lenoir, Green, and Stout 🏈On that note, Ji'Ayir Brown does seem to be winning favor as a Morris "guy". The DC specifically mentioned Brown today as he was rattling off names of supposed defensive pillars. Would be surprising if Brown isn't a Week 1 starter, IMO 🏈The DE injury concern is real, if only for the 49ers to conduct a proper training camp that includes two joint practices with the Titans and Chargers. They need more bodies there now and they were interested in adding one more vet even before the latest hardships. Jadeveon Clowney and Kyle Van Noy are both still available. Joey Bosa also technically is, but it seems that (if he isn't retiring) he wants to take the Eric Kendricks route of chilling now and signing later. Keep an eye on 49ers' workouts and signings at DE to get through the current storm. Maybe they sign a placeholder or two now and a more recognizable name later? We will see 🏈Deebo Day was fun. Of note that the 49ers wanted him even before Pearsall went down. So they've added — at the very, very least — a return man like they did with Skyy Moore last year, but with way more offensive upside

David Lombardi

77,975 просмотров • 29 дней назад

whoever leaked this has bigger balls than sense two researchers replaced $7,500 of human grading with $77.81 of model calls, and the number that fell out should stop anyone shipping an agent this week: the judge disagrees with itself 13.6% of the time it also prefers whichever answer it saw first, 72% of the time. cross-judge agreement is a kappa of 0.51: the statistical polite word for guessing this is Eval Engineering, the layer that arrives after loops and graphs, and it installs into the agent you already pay for: - run every comparison both ways and average the two, because one production judge picked whichever answer sat in the first slot 72% of the time - stop shipping on a single verdict: recovering the reference answer took 11 repeated trials before a majority vote landed it, and 15 on the questions that actually mattered - report the chance-corrected agreement instead of raw match, since the gap between those two ran 33 to 41 points on the same benchmark - never let a model grade its own family, because self-preference tracks capability upward rather than falling away as models improve - test stability and bias together, since two judges already running in production held above 0.95 on repeat while carrying severe positional preference - freeze the rubric wording and version it, because semantically identical templates flipped the majority outcome in a quarter of tested cases - write the rubric as one observable pass condition rather than a bundle of scores, so a coin-flip preference cannot hide inside an average - let plain code take every objective call, did the test pass, does the file exist, did the state change, and leave the judge only what genuinely needs reading - make the verdict do something structural to the run in flight, reject the handoff, swap the node, quarantine the branch, or you have built a dashboard the catch is that your judge is a measurement instrument nobody calibrates: single-trial judging is too noisy for anything you would stake money on, and the same judges reordered by as much as 14 positions depending on which benchmark you asked so the loop has to converge on the spec rather than on the score, because optimise against a judge long enough and the agent learns to look right instead of being right bookmark this, the full build with the five evals to start on and the gate that lets the fleet merge its own work is written out in the article ↓

whoever leaked this has bigger balls than sense two researchers replaced $7,500 of human grading with $77.81 of model calls, and the number that fell out should stop anyone shipping an agent this week: the judge disagrees with itself 13.6% of the time it also prefers whichever answer it saw first, 72% of the time. cross-judge agreement is a kappa of 0.51: the statistical polite word for guessing this is Eval Engineering, the layer that arrives after loops and graphs, and it installs into the agent you already pay for: - run every comparison both ways and average the two, because one production judge picked whichever answer sat in the first slot 72% of the time - stop shipping on a single verdict: recovering the reference answer took 11 repeated trials before a majority vote landed it, and 15 on the questions that actually mattered - report the chance-corrected agreement instead of raw match, since the gap between those two ran 33 to 41 points on the same benchmark - never let a model grade its own family, because self-preference tracks capability upward rather than falling away as models improve - test stability and bias together, since two judges already running in production held above 0.95 on repeat while carrying severe positional preference - freeze the rubric wording and version it, because semantically identical templates flipped the majority outcome in a quarter of tested cases - write the rubric as one observable pass condition rather than a bundle of scores, so a coin-flip preference cannot hide inside an average - let plain code take every objective call, did the test pass, does the file exist, did the state change, and leave the judge only what genuinely needs reading - make the verdict do something structural to the run in flight, reject the handoff, swap the node, quarantine the branch, or you have built a dashboard the catch is that your judge is a measurement instrument nobody calibrates: single-trial judging is too noisy for anything you would stake money on, and the same judges reordered by as much as 14 positions depending on which benchmark you asked so the loop has to converge on the spec rather than on the score, because optimise against a judge long enough and the agent learns to look right instead of being right bookmark this, the full build with the five evals to start on and the gate that lets the fleet merge its own work is written out in the article ↓

Argona

237,525 просмотров • 1 месяц назад

You spend years training for the World Cup, then lose to a cat. This Cat vs Human Football match rivalry was created using GPT Image 2 + Seedance 2.0 on RoboNeo #Roboneo Prompt Create a 15-second cinematic FIFA World Cup 2026 sports-comedy commercial. Every scene must happen strictly in the exact sequence listed below. Do NOT skip, merge, shorten, combine, or reorder any scene. Maintain perfect character consistency across all shots. MAIN CHARACTER: Athletic western blonde female soccer player, age 24, blue eyes, long blonde ponytail, red soccer jersey #8, red shorts, red socks, professional football boots. ANIMAL HERO: Sleek athletic grey cat with green eyes wearing a matching red soccer jersey. OPPONENTS: Two muscular female football players wearing white uniforms. ENVIRONMENT: FIFA World Cup 2026 atmosphere. Massive packed football stadium during golden hour. Warm sunlight, dramatic lens flares, vibrant green grass, roaring crowd, FIFA World Cup branding, international supporters waving USA, Canada, and Mexico flags. Premium global football-event energy. STYLE: Photorealistic live-action sports commercial, realistic human anatomy, Pixar-quality cat animation, Nike-style energy, dynamic camera movement, cinematic lighting, ultra-detailed textures, realistic football physics except for the cat's comedic athletic abilities. DURATION: 15 seconds total. SCENE 1 (0:00–0:02) — WORLD CUP OPENING High aerial establishing shot of a massive FIFA World Cup 2026 stadium at golden hour. Visible FIFA World Cup 2026 banners around the stadium. Crowd packed with supporters waving USA, Canada, and Mexico flags. A recognizable host-city landmark is visible in the distance. The blonde player in red jersey #8 dribbles confidently across the pitch while the grey cat runs beside her. Camera slowly pushes forward. Warm sunlight washes over the field. Massive crowd roar. On-screen title card appears briefly: "FIFA WORLD CUP 2026". SCENE 2 (0:02–0:03) — GAME ON Medium action shot. Two muscular opponents in white uniforms notice the cat and immediately shift their focus toward stopping it. The cat glances toward the ball with determination. Quick cinematic push-in. Crowd anticipation rises. SCENE 3 (0:03–0:05) — CAT SPEED BURST Ground-level tracking shot. The cat explodes forward between players' legs at impossible speed. Grass, dirt, and turf particles spray upward. Intense acceleration effect and realistic motion blur. Stadium crowd reacts with surprise. SCENE 4 (0:05–0:06) — PLAYER REACTION Extreme close-up of the blonde player. She smiles confidently, then her eyes suddenly widen in genuine surprise as she realizes how unbelievably fast the cat is moving. Lens flare sweeps across frame. SCENE 5 (0:06–0:08) — HERO CAT Low-angle hero close-up. The cat locks eyes on the ball. Golden sunlight creates a glowing halo around its silhouette. Slow-motion moment. Crowd noise softens briefly while heroic music rises. Fur detail and jersey fabric rendered with ultra-realistic quality. SCENE 6 (0:08–0:10) — FIRST DODGE Wide action shot. One defender launches a tackle. The cat instantly sidesteps and slips perfectly between both defenders. The tackle misses completely. The crowd erupts with laughter and excitement. SCENE 7 (0:10–0:12) — SLIDE CHAOS Both opponents commit to aggressive sliding tackles simultaneously. Massive dirt explosion. The cat effortlessly avoids both slides with perfect timing. The defenders collide with each other and tumble dramatically. Realistic physics, comedic timing. Crowd reaction intensifies. SCENE 8 (0:12–0:13) — DRIBBLE MASTER Dynamic tracking shot. The cat skillfully dribbles the ball through the chaos with incredible close control. Opponents tumble helplessly in the background. The blonde player watches in amused disbelief. Stadium energy reaches its peak. SCENE 9 (0:13–0:14) — IMPOSSIBLE SHOT Dramatic low-angle shot. The cat crouches beside the ball and looks upward. Tension builds. Suddenly the cat launches into an impossible move and strikes the ball toward goal. Cinematic speed ramp. Ball rockets through the air toward the net. SCENE 10 (0:14–0:15) — WORLD CUP GOAL & BRAND PAYOFF Ultra-cinematic goal shot. The football smashes into the back of the net at high speed. The net deforms naturally and ripples realistically with detailed cloth physics. Goal frame shakes subtly from impact. Crowd instantly explodes in celebration. Fans wave USA, Canada, and Mexico flags. The blonde player throws her hands up in disbelief and joy. The cat lands heroically and looks toward the roaring stadium. Freeze frame on the celebrating cat. On-screen tagline appears: "IMPOSSIBLE IS JUST KICKOFF." FIFA WORLD CUP 2026 logo appears prominently with premium commercial-grade branding. End on triumphant crowd roar and stadium atmosphere. CINEMATIC NOTES: Strict chronological sequence. No scene skipping, merging, or reordering. Fast-paced sports-commercial editing. Warm golden-hour color grading. Dynamic camera movement throughout. Premium FIFA World Cup 2026 advertising quality. Realistic football and net physics. Massive crowd reactions escalating throughout the film. Consistent character appearance across every shot. Photorealistic visuals with cinematic lens flares and broadcast-quality production values. Final goal must feel epic, emotional, and unmistakably FIFA World Cup 2026.

Sharon Riley

60,716 просмотров • 2 месяцев назад

Seedance 2.0 on Haimeta is next level. It doesn’t even look AI-generated anymore, it looks filmed. Prompt: FORMAT: 15s / 180 BPM / ONE CONTINUOUS SHOT / 360 POV downhill stair run, viral energy, max chaos SUBJECTS: First-person cyclist, handlebars and front wheel flashing low in frame during drops and hard turns. Vendors, laundry, scooters, dogs, chickens, cars, and pedestrians erupt around the rider as sudden obstacles. ENVIRONMENT: Dense Brazilian hillside streets, painted concrete stairs, tight landings, tiled corners, hanging wires, murals, awnings, puddles, hot late afternoon light, deep alley shadows, city sprawling below. MOOD: Adrenaline, max chaos, and nonstop street speed with violent spatial intensity in every direction. COLOR LOGIC: Hyperreal Pop Look CAMERA DETAILS: 360 action-cam POV, horizon-stable but brutal, nonstop forward drive, minimal roll, heavy stair vibration, sharp side and rear parallax, full-sphere chaos, no release. TIMELINE: 0:00-0:03: POV freefalls down a steep stairwell. The front wheel punches over the first steps, bars jackhammer below frame, the whole sphere shuddering with every hit while a child lunges in from the left to yank a rolling soccer ball out of the rider's line and walls, rails, balconies, and faces whip past on both sides. Violent forward descent with brutal stair vibration. SFX: (city hum, distant funk beat, tire chatter, breath, rapid stair hits, frame rattle). Hard sun above, deep shadow pockets below. 0:03-0:05: POV slashes left across a tiny landing, skips a broken crate, and drops again as laundry cracks across the front hemisphere, shoulder missing painted concrete by inches while side-wrap onlookers recoil. Hard lateral shake, immediate snap back into the stair run. SFX: (cloth slap, skid, crowd shout, chain buzz, stair chatter). 0:05-0:08: POV hammers the next staircase as a stray dog cuts center frame and startled chickens burst upward from the side steps, wings flaring across the sphere. The rider flicks the bars, rear wheel skates loose, then needles between a fruit cart and handrail with almost no clearance while the sphere jitters from every stair impact. Fast vibrating continuous-shot chaos. SFX: (tire chirp, crate clack, wings flapping, squawks, paw skitter, stair thuds, bass from window, metal rattle). 0:08-0:10: POV blasts out of one stair run, skips across a short asphalt gap between two stair sections without slowing, slams the opposite curb, and drops straight into the next descending steps. Only once the rider is fully back on the stairs do horns burst behind while passing cars rip through the side wrap. Full-speed crossing with strong lateral parallax and zero release. SFX: (engine idle, tire buzz, chain rattle, curb thump, car horns behind). 0:10-0:13: POV keeps attacking the next stair section immediately after the asphalt gap, machine-gunning through two tight landings and another steep stepped lane, every hit punching a fresh jolt through the sphere as stacked homes, wires, and alley mouths curl around the viewer and the overlook rushes closer. Forward lunge with impact drive and nonstop shake. SFX: (air rush, heavy thump, stair chatter, horn echo fading, chain lash, wind buffeting). 0:13-0:15: POV rips through the final stepped approach to the overlook and throws the rear wheel into a savage sideways skid, dust and gravel spraying across the lower frame while stair-lined drops fall away to both sides and the skyline blooms around the entire sphere in one fast violent sweep. The bike stays hot through the slide as the city fills every direction. SFX: (rear tire screech, gravel spray, freewheel spin, city roar, music drop)."

Seedance 2.0 on Haimeta is next level. It doesn’t even look AI-generated anymore, it looks filmed. Prompt: FORMAT: 15s / 180 BPM / ONE CONTINUOUS SHOT / 360 POV downhill stair run, viral energy, max chaos SUBJECTS: First-person cyclist, handlebars and front wheel flashing low in frame during drops and hard turns. Vendors, laundry, scooters, dogs, chickens, cars, and pedestrians erupt around the rider as sudden obstacles. ENVIRONMENT: Dense Brazilian hillside streets, painted concrete stairs, tight landings, tiled corners, hanging wires, murals, awnings, puddles, hot late afternoon light, deep alley shadows, city sprawling below. MOOD: Adrenaline, max chaos, and nonstop street speed with violent spatial intensity in every direction. COLOR LOGIC: Hyperreal Pop Look CAMERA DETAILS: 360 action-cam POV, horizon-stable but brutal, nonstop forward drive, minimal roll, heavy stair vibration, sharp side and rear parallax, full-sphere chaos, no release. TIMELINE: 0:00-0:03: POV freefalls down a steep stairwell. The front wheel punches over the first steps, bars jackhammer below frame, the whole sphere shuddering with every hit while a child lunges in from the left to yank a rolling soccer ball out of the rider's line and walls, rails, balconies, and faces whip past on both sides. Violent forward descent with brutal stair vibration. SFX: (city hum, distant funk beat, tire chatter, breath, rapid stair hits, frame rattle). Hard sun above, deep shadow pockets below. 0:03-0:05: POV slashes left across a tiny landing, skips a broken crate, and drops again as laundry cracks across the front hemisphere, shoulder missing painted concrete by inches while side-wrap onlookers recoil. Hard lateral shake, immediate snap back into the stair run. SFX: (cloth slap, skid, crowd shout, chain buzz, stair chatter). 0:05-0:08: POV hammers the next staircase as a stray dog cuts center frame and startled chickens burst upward from the side steps, wings flaring across the sphere. The rider flicks the bars, rear wheel skates loose, then needles between a fruit cart and handrail with almost no clearance while the sphere jitters from every stair impact. Fast vibrating continuous-shot chaos. SFX: (tire chirp, crate clack, wings flapping, squawks, paw skitter, stair thuds, bass from window, metal rattle). 0:08-0:10: POV blasts out of one stair run, skips across a short asphalt gap between two stair sections without slowing, slams the opposite curb, and drops straight into the next descending steps. Only once the rider is fully back on the stairs do horns burst behind while passing cars rip through the side wrap. Full-speed crossing with strong lateral parallax and zero release. SFX: (engine idle, tire buzz, chain rattle, curb thump, car horns behind). 0:10-0:13: POV keeps attacking the next stair section immediately after the asphalt gap, machine-gunning through two tight landings and another steep stepped lane, every hit punching a fresh jolt through the sphere as stacked homes, wires, and alley mouths curl around the viewer and the overlook rushes closer. Forward lunge with impact drive and nonstop shake. SFX: (air rush, heavy thump, stair chatter, horn echo fading, chain lash, wind buffeting). 0:13-0:15: POV rips through the final stepped approach to the overlook and throws the rear wheel into a savage sideways skid, dust and gravel spraying across the lower frame while stair-lined drops fall away to both sides and the skyline blooms around the entire sphere in one fast violent sweep. The bike stays hot through the slide as the city fills every direction. SFX: (rear tire screech, gravel spray, freewheel spin, city roar, music drop)."

Iqra Saifi

25,925 просмотров • 4 месяцев назад

$I love that Higgsfield is making its best AI commercial workflows completely open-source. Seedance 2.0 on Higgsfield AI Full open-sourced prompts & assets below. Prompts : Style: 8K IMAX. Photorealistic — no 3D render, no game engine, no game-cutscene aesthetic. Cinematography: Emmanuel Lubezki × Roger Deakins. Lighting: Natural light only — contre-jour backlight, camera on shadow side, atmospheric haze throughout. Key light from sky and windows only. Color: 60:30:10 — dominant / secondary / accent. Camera: Physical cine lens. 180° shutter motion blur. Skin: Pore-level realism — vellus hair, asymmetric moles, capillary flush, pore-shadow matching on-set light. Acting: Hollywood — micro-pauses before reactions, precise eye-line, wet living eyes with catch-lights, visible breath and chest rise. Physics: Gravity and inertia respected — mass has real weight, correct contact shadows. No floating props. Composition: Rule of thirds + golden ratio. Every person moving from frame one. Continuity: Characters, props, environment identical across every cut. No identity drift. Technical: 24fps smooth motion. 8K detail. No jitter. Audio: Environmental SFX only. No music. No subtitles. SUBJECT — >>, character matches input 100%, >> can in right hand. By end fully transformed into >> — armor assembled mid-air in real-time, >> matches input 100%. WB 6500K. MULTISHOT — fast cinematic cutting, escalating energy across the full 15s. LOCATION — >> is a STYLE REFERENCE ONLY, not a fixed keyframe. Use it for the look, mood, architecture and lighting of the environment — a deserted NYC-type avenue in hard midday sun. Seedance should freely generate and extend the world: new buildings, new stretches of avenue, fresh camera angles. The character TRAVELS through space across the shots — he is NOT pinned to the exact spot or framing of the input image. Do not reproduce the reference frame 1:1. ACTION — SHOT 1 (0:00–0:02) — EXTREME CLOSE, chin to forehead. >> raises >> to lips, deep pull. Adam's apple descends then rises twice. Hard cut. SHOT 2 (0:02–0:04) — Same close framing. Can lowers fast. Eyes snap open, sharp and alert (natural skin and iris detail, no glow). Zygomaticus major fires — wide genuine grin builds in under a second, 2mm to 8mm at mouth corners. Sternum rises with a full deep inhale. He is electric. Hard cut. SHOT 3 (0:04–0:06) — WIDE, low. >> EXPLODES forward into a sprint at 30 km/h, continuing in the SAME direction he was already facing — straight ahead down the avenue, NOT turning around. First strides — sneakers slam cracked asphalt, dust puffs eject 10cm. Hard cut. SHOT 4 (0:06–0:08) — Tracking the sprint. Arms drive at 90°, track jacket trails 15cm, grin held, brow raised, jaw relaxed open. Stride extends, ground contacts shorten. Building speed. Hard cut. SHOT 5 (0:08–0:10) — Final two strides, knees compress to 110°. Right foot drives into asphalt — surface fractures 20cm radius. He LAUNCHES upward at 65 km/h. Hard cut. SHOT 6 (0:10–0:13) — Mid-air in the >> building canyon, 24fps FULL SPEED, NO slow-motion. He is NOT frozen — his whole body keeps moving through the jump: torso rotates, legs cycle and tuck, arms swing out and reach, head turns to track the incoming armor. As he flies and twists through the air, armor pieces rip in from the city geometry and converge onto his moving body — chest plate LOCKS mid-rotation, pauldrons CLAMP as his shoulders swing, gauntlets SNAP onto forearms while his arms extend, shin plates clamp onto cycling legs, helmet descends and visor SLAMS shut as his head turns forward — one violent fluid sequence, body in constant motion throughout. Sharp metallic CRACK at visor seal. Hard cut. SHOT 7 (0:13–0:15) — >> SLAMS into >> in a hard crouch, both armored feet hit asphalt simultaneously, dust ring erupts 80cm. He rises, snaps gauntleted fists open and shut — left then right — glances down at the chest plate. Muffled laugh from behind the sealed visor. Hard cut to black. CAMERA — SUPER-DYNAMIC HANDHELD, never static, never locked off. Strong aggressive operator shake on every shot — 4–8cm bounce, organic human jitter, no stabilization. Pronounced DUTCH ANGLE throughout — frame canted 12–20° and shifting between shots, tilting harder as energy builds. SHOT 1: FOV 18° extreme close, hard dutch 15°, violent micro-shake. SHOT 2: FOV 18°, dutch flips to opposite side ~18°, jittery. SHOT 3: FOV 84° low, dutch 12°, whip into the sprint. SHOT 4: FOV 84°, camera physically chasing the run, dutch 16°, heavy stride bounce, motion blur on pans. SHOT 5: FOV 63°, dutch 20°, snap-tilt up tracking the launch. SHOT 6: FOV 63°, AGGRESSIVE HANDHELD CAMERA MOVEMENT — operator arcs and swings around the airborne body, 6–8cm bounce, fast reframes chasing each armor piece as it locks onto his moving form, strong DUTCH ANGLE swinging 18°→8°, no slow-mo. SHOT 7: FOV 63°, hard jolt on landing impact, dutch 14°, camera recoils 8cm then keeps breathing on the suit inspection. STYLE — Dominant: bleached asphalt grey and building canyon 60%. Secondary: navy/cream outfit transitioning to chrome armor 30%. Accent: grin, dust, chrome flash and visor LED blue on landing 10%. WB 6500K. NO glowing eyes — eyes lit by natural sunlight only. CONSTRAINTS — Cut to black at end. 16:9. MULTISHOT with hard cuts per the timecodes above. NO slow-motion anywhere — everything 24fps full speed, especially the transformation. Camera always handheld with strong shake and dutch angle — NEVER static or locked. NO eye glow. >> is a reference for style only — do NOT lock to the input frame or keep him in the same spot; the world extends and he moves through it. Grin must be fully established before the sprint begins.$

I love that Higgsfield is making its best AI commercial workflows completely open-source. Seedance 2.0 on Higgsfield AI Full open-sourced prompts & assets below. Prompts : Style: 8K IMAX. Photorealistic — no 3D render, no game engine, no game-cutscene aesthetic. Cinematography: Emmanuel Lubezki × Roger Deakins. Lighting: Natural light only — contre-jour backlight, camera on shadow side, atmospheric haze throughout. Key light from sky and windows only. Color: 60:30:10 — dominant / secondary / accent. Camera: Physical cine lens. 180° shutter motion blur. Skin: Pore-level realism — vellus hair, asymmetric moles, capillary flush, pore-shadow matching on-set light. Acting: Hollywood — micro-pauses before reactions, precise eye-line, wet living eyes with catch-lights, visible breath and chest rise. Physics: Gravity and inertia respected — mass has real weight, correct contact shadows. No floating props. Composition: Rule of thirds + golden ratio. Every person moving from frame one. Continuity: Characters, props, environment identical across every cut. No identity drift. Technical: 24fps smooth motion. 8K detail. No jitter. Audio: Environmental SFX only. No music. No subtitles. SUBJECT — >>, character matches input 100%, >> can in right hand. By end fully transformed into >> — armor assembled mid-air in real-time, >> matches input 100%. WB 6500K. MULTISHOT — fast cinematic cutting, escalating energy across the full 15s. LOCATION — >> is a STYLE REFERENCE ONLY, not a fixed keyframe. Use it for the look, mood, architecture and lighting of the environment — a deserted NYC-type avenue in hard midday sun. Seedance should freely generate and extend the world: new buildings, new stretches of avenue, fresh camera angles. The character TRAVELS through space across the shots — he is NOT pinned to the exact spot or framing of the input image. Do not reproduce the reference frame 1:1. ACTION — SHOT 1 (0:00–0:02) — EXTREME CLOSE, chin to forehead. >> raises >> to lips, deep pull. Adam's apple descends then rises twice. Hard cut. SHOT 2 (0:02–0:04) — Same close framing. Can lowers fast. Eyes snap open, sharp and alert (natural skin and iris detail, no glow). Zygomaticus major fires — wide genuine grin builds in under a second, 2mm to 8mm at mouth corners. Sternum rises with a full deep inhale. He is electric. Hard cut. SHOT 3 (0:04–0:06) — WIDE, low. >> EXPLODES forward into a sprint at 30 km/h, continuing in the SAME direction he was already facing — straight ahead down the avenue, NOT turning around. First strides — sneakers slam cracked asphalt, dust puffs eject 10cm. Hard cut. SHOT 4 (0:06–0:08) — Tracking the sprint. Arms drive at 90°, track jacket trails 15cm, grin held, brow raised, jaw relaxed open. Stride extends, ground contacts shorten. Building speed. Hard cut. SHOT 5 (0:08–0:10) — Final two strides, knees compress to 110°. Right foot drives into asphalt — surface fractures 20cm radius. He LAUNCHES upward at 65 km/h. Hard cut. SHOT 6 (0:10–0:13) — Mid-air in the >> building canyon, 24fps FULL SPEED, NO slow-motion. He is NOT frozen — his whole body keeps moving through the jump: torso rotates, legs cycle and tuck, arms swing out and reach, head turns to track the incoming armor. As he flies and twists through the air, armor pieces rip in from the city geometry and converge onto his moving body — chest plate LOCKS mid-rotation, pauldrons CLAMP as his shoulders swing, gauntlets SNAP onto forearms while his arms extend, shin plates clamp onto cycling legs, helmet descends and visor SLAMS shut as his head turns forward — one violent fluid sequence, body in constant motion throughout. Sharp metallic CRACK at visor seal. Hard cut. SHOT 7 (0:13–0:15) — >> SLAMS into >> in a hard crouch, both armored feet hit asphalt simultaneously, dust ring erupts 80cm. He rises, snaps gauntleted fists open and shut — left then right — glances down at the chest plate. Muffled laugh from behind the sealed visor. Hard cut to black. CAMERA — SUPER-DYNAMIC HANDHELD, never static, never locked off. Strong aggressive operator shake on every shot — 4–8cm bounce, organic human jitter, no stabilization. Pronounced DUTCH ANGLE throughout — frame canted 12–20° and shifting between shots, tilting harder as energy builds. SHOT 1: FOV 18° extreme close, hard dutch 15°, violent micro-shake. SHOT 2: FOV 18°, dutch flips to opposite side ~18°, jittery. SHOT 3: FOV 84° low, dutch 12°, whip into the sprint. SHOT 4: FOV 84°, camera physically chasing the run, dutch 16°, heavy stride bounce, motion blur on pans. SHOT 5: FOV 63°, dutch 20°, snap-tilt up tracking the launch. SHOT 6: FOV 63°, AGGRESSIVE HANDHELD CAMERA MOVEMENT — operator arcs and swings around the airborne body, 6–8cm bounce, fast reframes chasing each armor piece as it locks onto his moving form, strong DUTCH ANGLE swinging 18°→8°, no slow-mo. SHOT 7: FOV 63°, hard jolt on landing impact, dutch 14°, camera recoils 8cm then keeps breathing on the suit inspection. STYLE — Dominant: bleached asphalt grey and building canyon 60%. Secondary: navy/cream outfit transitioning to chrome armor 30%. Accent: grin, dust, chrome flash and visor LED blue on landing 10%. WB 6500K. NO glowing eyes — eyes lit by natural sunlight only. CONSTRAINTS — Cut to black at end. 16:9. MULTISHOT with hard cuts per the timecodes above. NO slow-motion anywhere — everything 24fps full speed, especially the transformation. Camera always handheld with strong shake and dutch angle — NEVER static or locked. NO eye glow. >> is a reference for style only — do NOT lock to the input frame or keep him in the same spot; the world extends and he moves through it. Grin must be fully established before the sprint begins.

yopiwhs

11,743 просмотров • 1 месяц назад

i spent $26,600 on cloud GPU rentals over 14 months before i found a NVIDIA DGX Spark at $2,999 (founder's edition) or $3,999 (shipping price) it paid for itself in 6 weeks i run 200B parameter models locally now and my old cloud provider keeps sending me loyalty discount emails the math on that $26,600 is embarrassing to type out loud $1,900/month for 14 months, H100 instances on a specialist cloud provider, because anything bigger than a 70B model simply would not fit anywhere else i paid the invoices like they were a utility bill and told myself it was just the cost of doing serious AI work it took me over a year to find out it wasn't 14 months, broken down: → months 1-4: $1,400-1,600/month - felt like manageable infrastructure overhead → months 5-9: crept to $1,900-2,100 as i started running DeepSeek-class experiments, costs tracking directly with model size → months 10-12: one agent loop ran for 36 hours against a 130B model while i slept, that month hit $2,400 → month 13: ran the cumulative total for the first time, saw $23,800, felt physically sick → month 14: another $2,800 month while i waited for the hardware to ship the box is the NVIDIA DGX Spark - roughly the footprint of a large mac mini, powered by a GB10 Grace Blackwell chip with 128GB of unified LPDDR5X memory that unified memory is the whole thing an RTX 4090 has 24GB of VRAM, which means a 70B model in full BF16 precision physically does not fit, you're quantizing down or you're renting cloud, those are your options this box loads a 200B parameter model quantized and serves it through vLLM over localhost, same API interface the cloud endpoint used the migration took one line of code - i changed the base URL from the provider's endpoint to 127.0.0.1:8000 and everything just worked electricity to run continuous 200B inference locally comes out to about $12/month the payback arithmetic is almost too clean: $2,999 hardware cost against $1,900/month saved, the box paid for itself before i'd owned it two months what i didn't account for was how completely the cost model changes your behavior when there's no hourly meter running, you greenlight experiments you'd never approve on cloud - agent loops that churn for hours, running 10,000 documents through a reasoning pass at 3am, speculative fine-tuning jobs you'd normally skip because the cost felt unjustifiable i ran more experiments in the first 30 days after the box arrived than in the four months before it the loyalty discount email landed about 8 weeks after i cancelled the cloud subscription 15% off my next three months, valued customer, we'd love to have you back i didn't reply the box was already running

i spent $26,600 on cloud GPU rentals over 14 months before i found a NVIDIA DGX Spark at $2,999 (founder's edition) or $3,999 (shipping price) it paid for itself in 6 weeks i run 200B parameter models locally now and my old cloud provider keeps sending me loyalty discount emails the math on that $26,600 is embarrassing to type out loud $1,900/month for 14 months, H100 instances on a specialist cloud provider, because anything bigger than a 70B model simply would not fit anywhere else i paid the invoices like they were a utility bill and told myself it was just the cost of doing serious AI work it took me over a year to find out it wasn't 14 months, broken down: → months 1-4: $1,400-1,600/month - felt like manageable infrastructure overhead → months 5-9: crept to $1,900-2,100 as i started running DeepSeek-class experiments, costs tracking directly with model size → months 10-12: one agent loop ran for 36 hours against a 130B model while i slept, that month hit $2,400 → month 13: ran the cumulative total for the first time, saw $23,800, felt physically sick → month 14: another $2,800 month while i waited for the hardware to ship the box is the NVIDIA DGX Spark - roughly the footprint of a large mac mini, powered by a GB10 Grace Blackwell chip with 128GB of unified LPDDR5X memory that unified memory is the whole thing an RTX 4090 has 24GB of VRAM, which means a 70B model in full BF16 precision physically does not fit, you're quantizing down or you're renting cloud, those are your options this box loads a 200B parameter model quantized and serves it through vLLM over localhost, same API interface the cloud endpoint used the migration took one line of code - i changed the base URL from the provider's endpoint to 127.0.0.1:8000 and everything just worked electricity to run continuous 200B inference locally comes out to about $12/month the payback arithmetic is almost too clean: $2,999 hardware cost against $1,900/month saved, the box paid for itself before i'd owned it two months what i didn't account for was how completely the cost model changes your behavior when there's no hourly meter running, you greenlight experiments you'd never approve on cloud - agent loops that churn for hours, running 10,000 documents through a reasoning pass at 3am, speculative fine-tuning jobs you'd normally skip because the cost felt unjustifiable i ran more experiments in the first 30 days after the box arrived than in the four months before it the loyalty discount email landed about 8 weeks after i cancelled the cloud subscription 15% off my next three months, valued customer, we'd love to have you back i didn't reply the box was already running

Argona

22,355 просмотров • 3 месяцев назад

#Patriots Voluntary OTA Notes [Non-Padded] (6/2): 𝗢𝗳𝗳𝗲𝗻𝘀𝗲 • QB Drake Maye connected with WR Kayshon Boutte for the play of the day — Maye perfectly placed the ball deep along the left side of the field and Boutte made a fingertip catch. • QB Drake Maye had no turnovers today and he had his best day of open OTAs by far — Maye completed 9 straight passes at one point. • Full QB Stats: Maye 14/16 Dobbs 11/13 Wooldridge 4/8, INT • WRs DeMario Douglas, Kayshon Boutte, Kyle Williams and Efton Chism spent the most time working with QB Drake Maye today. • WR Efton Chism III had 7 receptions on 7 targets and looked to always be open — the undrafted rookie took several snaps with QB Drake Maye and the top offensive unit. • RB Rhamondre Stevenson made an extremely impressive finger-tip grab for a 25-yard gain on a dot from QB Drake Maye. • OL Cole Strange and Wes Schweitzer were the two primary left guards today — that spot on the offensive line is truly a competition. • TE Jaheim Bell had a rough day — the former seventh-round pick had 2 bad drops. • It was a bit of a quiet day for rookie WR Kyle Williams who caught 1 of his 2 targets. • WR Stefon Diggs was limited today working with WR Ja’Lynn Polk off to the side at times — Diggs closely observed the offense towards the end of practice as they practiced plays. • Rookie TE CJ Dippre caught a nice pass thrown by QB Ben Wooldridge up the seam. • Veteran OT Morgan Moses once again stayed after practice to help and give advice to the young offensive lineman. • RG Mike Onwenu says he feels good getting back to an offense he is familiar with — Onwenu adds that he is rehabbing from a hand injury. • RB Rhamondre Stevenson says RBs coach Tony Dews flew to visit him in Las Vegas while Stevenson was dealing with his father’s passing. • Limited: WR Stefon Diggs WR Ja'Lynn Polk OL Mike Onwenu OL Jared Wilson OT Vederian Lowe TE Austin Hooper • Absent: WR Mack Hollins (Present But DNP) WR Kendrick Bourne (Unknown) RB Trayveon Williams (Unknown) OL Tyrese Robinson (Unknown) 𝗗𝗲𝗳𝗲𝗻𝘀𝗲 • DE Keion White was disruptive today — White sacked both QB Drake Maye and QB Josh Dobbs during team drills. • Rookie S Josh Minkins intercepted QB Ben Wooldridge during 11-on—11s • CB Christian Gonzalez, S Jaylinn Hawkins, and LB Monty Rice each had a pass break up. • DT Joshua Farmer batted another pass down during team drills — Farmer has been able to bat passes down consistently during OTAs. • CB Christian Gonzalez gave up a catch to WR Efton Chism — Chism got open after running an impressive dig route. • LB Jahlani Tavai went down during team period and was grabbing his left knee — the veteran had to be helped off the field after staying down for a couple minutes. • CBs Alex Austin and Marcellas Dial spent time in the slot — Austin has consistently taken reps in the slot since the start of OTAs. • The defensive line continues to disrupt the offense. • Limited: DL Jaquelin Roy DL Wilfried Pene • Absent: CB Carlton Davis (Unknown) CB Marcus Jones (Unknown) ED Anfernee Jennings (Unknown) 𝗦𝗽𝗲𝗰𝗶𝗮𝗹 𝗧𝗲𝗮𝗺𝘀 • Field Goals (33-50 Yards) Andy Borregales: 4/4 John Parker Romo: 2/4, missed wide right Borregales has yet to miss during open OTAs. #NEPats

#Patriots Voluntary OTA Notes [Non-Padded] (6/2): 𝗢𝗳𝗳𝗲𝗻𝘀𝗲 • QB Drake Maye connected with WR Kayshon Boutte for the play of the day — Maye perfectly placed the ball deep along the left side of the field and Boutte made a fingertip catch. • QB Drake Maye had no turnovers today and he had his best day of open OTAs by far — Maye completed 9 straight passes at one point. • Full QB Stats: Maye 14/16 Dobbs 11/13 Wooldridge 4/8, INT • WRs DeMario Douglas, Kayshon Boutte, Kyle Williams and Efton Chism spent the most time working with QB Drake Maye today. • WR Efton Chism III had 7 receptions on 7 targets and looked to always be open — the undrafted rookie took several snaps with QB Drake Maye and the top offensive unit. • RB Rhamondre Stevenson made an extremely impressive finger-tip grab for a 25-yard gain on a dot from QB Drake Maye. • OL Cole Strange and Wes Schweitzer were the two primary left guards today — that spot on the offensive line is truly a competition. • TE Jaheim Bell had a rough day — the former seventh-round pick had 2 bad drops. • It was a bit of a quiet day for rookie WR Kyle Williams who caught 1 of his 2 targets. • WR Stefon Diggs was limited today working with WR Ja’Lynn Polk off to the side at times — Diggs closely observed the offense towards the end of practice as they practiced plays. • Rookie TE CJ Dippre caught a nice pass thrown by QB Ben Wooldridge up the seam. • Veteran OT Morgan Moses once again stayed after practice to help and give advice to the young offensive lineman. • RG Mike Onwenu says he feels good getting back to an offense he is familiar with — Onwenu adds that he is rehabbing from a hand injury. • RB Rhamondre Stevenson says RBs coach Tony Dews flew to visit him in Las Vegas while Stevenson was dealing with his father’s passing. • Limited: WR Stefon Diggs WR Ja'Lynn Polk OL Mike Onwenu OL Jared Wilson OT Vederian Lowe TE Austin Hooper • Absent: WR Mack Hollins (Present But DNP) WR Kendrick Bourne (Unknown) RB Trayveon Williams (Unknown) OL Tyrese Robinson (Unknown) 𝗗𝗲𝗳𝗲𝗻𝘀𝗲 • DE Keion White was disruptive today — White sacked both QB Drake Maye and QB Josh Dobbs during team drills. • Rookie S Josh Minkins intercepted QB Ben Wooldridge during 11-on—11s • CB Christian Gonzalez, S Jaylinn Hawkins, and LB Monty Rice each had a pass break up. • DT Joshua Farmer batted another pass down during team drills — Farmer has been able to bat passes down consistently during OTAs. • CB Christian Gonzalez gave up a catch to WR Efton Chism — Chism got open after running an impressive dig route. • LB Jahlani Tavai went down during team period and was grabbing his left knee — the veteran had to be helped off the field after staying down for a couple minutes. • CBs Alex Austin and Marcellas Dial spent time in the slot — Austin has consistently taken reps in the slot since the start of OTAs. • The defensive line continues to disrupt the offense. • Limited: DL Jaquelin Roy DL Wilfried Pene • Absent: CB Carlton Davis (Unknown) CB Marcus Jones (Unknown) ED Anfernee Jennings (Unknown) 𝗦𝗽𝗲𝗰𝗶𝗮𝗹 𝗧𝗲𝗮𝗺𝘀 • Field Goals (33-50 Yards) Andy Borregales: 4/4 John Parker Romo: 2/4, missed wide right Borregales has yet to miss during open OTAs. #NEPats

Carlos A. Lopez

60,241 просмотров • 1 год назад

Facial reconstruction of a 25,000-year-old Gravettian This man was buried together with two teenage girls—his close relatives. Barma Grande is a cave belonging to the famous Grimaldi cave complex in Liguria. The cave was first explored in 1872. Later, Abbo, a local entrepreneur, purchased it for limestone extraction. In 1884, the Frenchman Louis Julien obtained permission from the owner to conduct excavations and, after removing about 8 meters of deposits at the cave entrance, discovered the burial of a man covered with red ochre, accompanied by flint blades. The skeleton was prepared for transport to a museum; however, a dispute over the rights to the finds broke out, resulting in most of the skeleton being destroyed. Further excavations were undertaken by the cave’s owner himself, with very practical goals - to quarry stone for building a terrace under his vineyard. In February 1892, one of Abbo’s children, playing in the cave, came across human bones. The find turned out to be a triple burial of ancient humans. The discovery attracted scholars, who determined that the skeletons had likely been placed in an artificial pit. This conclusion was confirmed by further excavations in the cave. Later, the English patron Thomas Hanbury built a small museum in Grimaldi, where the three discovered skeletons were put on public display. This is the only skeletal material from Grimaldi that has remained in Italy. Barma Grande 2 (the man) lay on his back, while the two adolescents, Barma Grande 3 and 4, lay side by side on their sides, facing the cave entrance. All three appear to have been buried at the same time. The skeletons were covered with ochre, and the grave contained numerous perforated marine shells, deer canines, fish vertebrae, beads and ivory pendants, as well as extraordinarily long flint blades (Gravettian culture). Beneath the skull of BG3, who lay in the middle, was the bone of a horned animal, and beneath the occiput of BG4 was a large flint tool. The ornaments apparently once adorned their clothing. The skeletons belonged to a mature man and two adolescents aged 12–13 and 14–15. Initially, anthropologists assumed that one adolescent was a girl and the other a boy. Nearly 100 years later, skeletal analysis confirmed that Barma Grande 3 was female, but uncertainty remained regarding the second adolescent. The question was finally resolved by recent DNA analysis: both adolescents were girls. Certain features of the skull structure suggested close kinship among all three individuals. This hypothesis was confirmed by mitochondrial DNA analysis of the skeletons in 2006. The man was very tall - 188 cm, with a massive skull and robust limbs. Marked asymmetry in his arms indicated that, due to an injury, he had been forced for a long time to rely almost exclusively on his right hand.

Facial reconstruction of a 25,000-year-old Gravettian This man was buried together with two teenage girls—his close relatives. Barma Grande is a cave belonging to the famous Grimaldi cave complex in Liguria. The cave was first explored in 1872. Later, Abbo, a local entrepreneur, purchased it for limestone extraction. In 1884, the Frenchman Louis Julien obtained permission from the owner to conduct excavations and, after removing about 8 meters of deposits at the cave entrance, discovered the burial of a man covered with red ochre, accompanied by flint blades. The skeleton was prepared for transport to a museum; however, a dispute over the rights to the finds broke out, resulting in most of the skeleton being destroyed. Further excavations were undertaken by the cave’s owner himself, with very practical goals - to quarry stone for building a terrace under his vineyard. In February 1892, one of Abbo’s children, playing in the cave, came across human bones. The find turned out to be a triple burial of ancient humans. The discovery attracted scholars, who determined that the skeletons had likely been placed in an artificial pit. This conclusion was confirmed by further excavations in the cave. Later, the English patron Thomas Hanbury built a small museum in Grimaldi, where the three discovered skeletons were put on public display. This is the only skeletal material from Grimaldi that has remained in Italy. Barma Grande 2 (the man) lay on his back, while the two adolescents, Barma Grande 3 and 4, lay side by side on their sides, facing the cave entrance. All three appear to have been buried at the same time. The skeletons were covered with ochre, and the grave contained numerous perforated marine shells, deer canines, fish vertebrae, beads and ivory pendants, as well as extraordinarily long flint blades (Gravettian culture). Beneath the skull of BG3, who lay in the middle, was the bone of a horned animal, and beneath the occiput of BG4 was a large flint tool. The ornaments apparently once adorned their clothing. The skeletons belonged to a mature man and two adolescents aged 12–13 and 14–15. Initially, anthropologists assumed that one adolescent was a girl and the other a boy. Nearly 100 years later, skeletal analysis confirmed that Barma Grande 3 was female, but uncertainty remained regarding the second adolescent. The question was finally resolved by recent DNA analysis: both adolescents were girls. Certain features of the skull structure suggested close kinship among all three individuals. This hypothesis was confirmed by mitochondrial DNA analysis of the skeletons in 2006. The man was very tall - 188 cm, with a massive skull and robust limbs. Marked asymmetry in his arms indicated that, due to an injury, he had been forced for a long time to rely almost exclusively on his right hand.

Ancestral Whispers

36,639 просмотров • 6 месяцев назад

[LSTM] by Hand ✍️ LSTMs have been the most effective architecture to process long sequences of data, until our world was taken over by the Transformers. LSTMs belong to the broader family of recurrent neural network (RNNs) that process data sequentially in a recurrent manner. Transformers, on the other hand, abandon recurrence and use self-attention instead to process data concurrently in parallel. Recently, there is renewed interest in recurrence as people realized self-attention doesn’t scale to extremely long sequences, like hundreds of thousands of tokens. Mamba is a good example to bring back recurrence. All of a sudden, it is cool to study LSTMs. How do LSTMs work? [1] Given ↳ 🟨 Input sequence X1, X2, X3 (d = 3) ↳ 🟩 Hidden state h (d = 2) ↳ 🟦 Memory C (d = 2) ↳ Weight matrices Wf, Wc, Wi, Wo Process t = 1 [2] Initialize ↳ Randomly set the previous hidden state h0 to [1, 1] and memory cells C0 to [0.3, -0.5] [3] Linear Transform ↳ Multiply the four weight matrices with the concatenation of current input (X1) and the previous hidden state (h0). ↳ The results are feature values, each is a linear combination of the current input and hidden state. [4] Non-linear Transform ↳ Apply sigmoid σ to obtain gate values (between 0 and 1). • Forget gate (f1): [-4, -6] → [0, 0] • Input gate (i1): [6, 4] → [1, 1] • Output gate (o1): [4, -5] → [1, 0] ↳ Apply tanh to obtain candidate memory values (between -1 and 1) • Candidate memory (C’1): [1, -6] → [0.8, -1] [5] Update Memory ↳ Forget (C0 .* f1): Element-wise multiply the current memory with forget gate values. ↳ Input (C’1 .* o1): Element-wise multiply the “candidate” memory with input gate values. ↳ Update the memory to C1 by adding the two terms above: C0 .* f1 + C’1 .* o1 = C1 [6] Candiate Output ↳ Apply tanh to the new memory C1 to obtain candidate output o’1. [0.8, -1] → [0.7, -0.8] [7] Update Hidden State ↳ Output (o’1 .* o1 → h1): Element-wise multiply the candidate output with the output gate. ↳ The result is updated hidden state h1 ↳ Also, it is the first output. Process t = 2 [8] Initialize ↳ Copy previous hidden state h1 and memory C1 [9] Linear Transform ↳ Repeat [3] [10] Update Memory (C2) ↳ Repeat [4] and [5] [11] Update Hidden State (h2) ↳ Repeat [6] and [7] Process t = 3 [12] Initialize ↳ Copy previous hidden state h2 and memory C2 [13] Linear Transform ↳ Repeat [3] [14] Update Memory (C3) ↳ Repeat [4] and [5] [15] Update Hidden State (h3) ↳ Repeat [6] and [7]

[LSTM] by Hand ✍️ LSTMs have been the most effective architecture to process long sequences of data, until our world was taken over by the Transformers. LSTMs belong to the broader family of recurrent neural network (RNNs) that process data sequentially in a recurrent manner. Transformers, on the other hand, abandon recurrence and use self-attention instead to process data concurrently in parallel. Recently, there is renewed interest in recurrence as people realized self-attention doesn’t scale to extremely long sequences, like hundreds of thousands of tokens. Mamba is a good example to bring back recurrence. All of a sudden, it is cool to study LSTMs. How do LSTMs work? [1] Given ↳ 🟨 Input sequence X1, X2, X3 (d = 3) ↳ 🟩 Hidden state h (d = 2) ↳ 🟦 Memory C (d = 2) ↳ Weight matrices Wf, Wc, Wi, Wo Process t = 1 [2] Initialize ↳ Randomly set the previous hidden state h0 to [1, 1] and memory cells C0 to [0.3, -0.5] [3] Linear Transform ↳ Multiply the four weight matrices with the concatenation of current input (X1) and the previous hidden state (h0). ↳ The results are feature values, each is a linear combination of the current input and hidden state. [4] Non-linear Transform ↳ Apply sigmoid σ to obtain gate values (between 0 and 1). • Forget gate (f1): [-4, -6] → [0, 0] • Input gate (i1): [6, 4] → [1, 1] • Output gate (o1): [4, -5] → [1, 0] ↳ Apply tanh to obtain candidate memory values (between -1 and 1) • Candidate memory (C’1): [1, -6] → [0.8, -1] [5] Update Memory ↳ Forget (C0 .* f1): Element-wise multiply the current memory with forget gate values. ↳ Input (C’1 .* o1): Element-wise multiply the “candidate” memory with input gate values. ↳ Update the memory to C1 by adding the two terms above: C0 .* f1 + C’1 .* o1 = C1 [6] Candiate Output ↳ Apply tanh to the new memory C1 to obtain candidate output o’1. [0.8, -1] → [0.7, -0.8] [7] Update Hidden State ↳ Output (o’1 .* o1 → h1): Element-wise multiply the candidate output with the output gate. ↳ The result is updated hidden state h1 ↳ Also, it is the first output. Process t = 2 [8] Initialize ↳ Copy previous hidden state h1 and memory C1 [9] Linear Transform ↳ Repeat [3] [10] Update Memory (C2) ↳ Repeat [4] and [5] [11] Update Hidden State (h2) ↳ Repeat [6] and [7] Process t = 3 [12] Initialize ↳ Copy previous hidden state h2 and memory C2 [13] Linear Transform ↳ Repeat [3] [14] Update Memory (C3) ↳ Repeat [4] and [5] [15] Update Hidden State (h3) ↳ Repeat [6] and [7]

Tom Yeh

72,966 просмотров • 2 лет назад

In 2024, Russia faced a 13-year record jump in gas prices after a series of Ukrainian drone attacks on refineries that caused fuel production to collapse by more than 10% in the first half of the year. According to Rosstat, from the beginning of the year to December 23, gasoline prices rose by 11% on average in Russia. At the same time, prices in remote regions of the Far East exceeded the Russian average by a quarter. By the end of the year, gasoline price growth will be the strongest since 2011. For the first time in 6 years, gasoline prices have risen significantly above the headline inflation rate, which the Russian ministry of economic development estimates at 9.7% a week before the end of the year. This year, the Russian government tried to curb gasoline prices by banning its exports: the restrictions were imposed shortly after two dozen major Russian refineries were attacked by Ukrainian UAVs and a number of them were forced to halt production. By the end of May, the decline in gasoline production in Russia reached 20% compared to December 2023, and diesel fuel - 11%. In response, the authorities classified fuel output statistics, citing geopolitics and the threat of market manipulation as reasons. In 2025, gasoline will continue to rise in price in Russia. In the best-case scenario, it will rise by 10-15%, and in the worst case - 20%, which, according to Rosstat, has not happened in Russia since 2004 (when the cost of fuel jumped by a record 31.3%). The reason will be an increase in Transneft's pipeline pumping tariffs, as well as a sharp increase in excise taxes: they will rise 14% for gasoline and 16% for diesel fuel, which is three times more than was provided for in the Tax Code (4.7%). According to government calculations, this will bring 170 billion rubles ($1,6 billion) to the treasury, of which 116 billion ($1,13 billion) will be due to the unscheduled increase. Rising gasoline prices will automatically lead to higher prices for everything else in Russia, as the cost of fuel is included in the delivery of all goods and automatically increases the cost of the end product for the buyer. Inflation in Russia will also increase.

In 2024, Russia faced a 13-year record jump in gas prices after a series of Ukrainian drone attacks on refineries that caused fuel production to collapse by more than 10% in the first half of the year. According to Rosstat, from the beginning of the year to December 23, gasoline prices rose by 11% on average in Russia. At the same time, prices in remote regions of the Far East exceeded the Russian average by a quarter. By the end of the year, gasoline price growth will be the strongest since 2011. For the first time in 6 years, gasoline prices have risen significantly above the headline inflation rate, which the Russian ministry of economic development estimates at 9.7% a week before the end of the year. This year, the Russian government tried to curb gasoline prices by banning its exports: the restrictions were imposed shortly after two dozen major Russian refineries were attacked by Ukrainian UAVs and a number of them were forced to halt production. By the end of May, the decline in gasoline production in Russia reached 20% compared to December 2023, and diesel fuel - 11%. In response, the authorities classified fuel output statistics, citing geopolitics and the threat of market manipulation as reasons. In 2025, gasoline will continue to rise in price in Russia. In the best-case scenario, it will rise by 10-15%, and in the worst case - 20%, which, according to Rosstat, has not happened in Russia since 2004 (when the cost of fuel jumped by a record 31.3%). The reason will be an increase in Transneft's pipeline pumping tariffs, as well as a sharp increase in excise taxes: they will rise 14% for gasoline and 16% for diesel fuel, which is three times more than was provided for in the Tax Code (4.7%). According to government calculations, this will bring 170 billion rubles ($1,6 billion) to the treasury, of which 116 billion ($1,13 billion) will be due to the unscheduled increase. Rising gasoline prices will automatically lead to higher prices for everything else in Russia, as the cost of fuel is included in the delivery of all goods and automatically increases the cost of the end product for the buyer. Inflation in Russia will also increase.

Anton Gerashchenko

81,314 просмотров • 1 год назад