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

This masterpiece piece traveling with us for four decades 😍😍😍 Same tune for Four different situations, Four diff arrangements & orchestrations Four different singers Four different movies 1.Thumbi vaa - Olangal (1982) 2.Sangathil Padatha - Auto Raja (1982) 3.Monday Tho Utkar - Aur Ek Prem Kahani (1996) 4.Gumm Summ... show more

DJR

1,652 subscribers

22,161 просмотров • 1 год назад •via X (Twitter)

Музыка Искусство

Anya Rossi• Live Now

Private livecam show

Комментарии: 11

Фото профиля pachaiyammal

pachaiyammal1 год назад

this is the telugu version

Фото профиля DJR

DJR1 год назад

wow this is new to me 😍😍🙏🏻

Фото профиля Superior Titanium Products, Inc.

Superior Titanium Products, Inc.3 лет назад

"I was given one of your VIPER money clip last weekend… I am amazed! My friend had 100 bills folded- threw it at me and I was instantly SOLD. He’s ordering another one…I kept his….minus the money! I’ve owned “many” over the years….this one ROCKS!" - Jeff C.

Фото профиля Throttle

Throttle1 год назад

6th version

Фото профиля DJR

DJR1 год назад

really 😍😍 this is nice like a pathos version 😍😍

Фото профиля Throttle

Throttle1 год назад

7th version

Фото профиля DJR

DJR1 год назад

semma semma this one i knew 😍😍

Фото профиля Partha Krishnaswamy

Partha Krishnaswamy1 год назад

There could have been one more version had Bhagyaraj accepted this tune for Thooral Ninnu Pochu. When Bhagyaraj said enna saami namaku ellam intha mathiri tune kuduka maatraru...Pandiarajan responded by saying mudhala namaku thaan kodatharu, neenga venamnu sollitinga 🙂.

Фото профиля DJR

DJR1 год назад

can guess that, Thooral Ninnu Pochu also in same year 1982 👍🏻👍🏻😊😊

Фото профиля Thundering typhoon

Thundering typhoon1 год назад

You maybe surprised.

Фото профиля DJR

DJR1 год назад

😍😍😍

Похожие видео

Fibonacci numbers are the sequence where each one is the sum of the two before it: 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233… Every so often one of them is prime these are called Fibonacci primes, and only 21 are known. Here's the new discovery, take any Fibonacci prime (past the first few small ones) and reduce it to modulo 144. You'd expect the leftover remainder could be almost anything , 144 different possibilities. It never is. It only ever lands on one of just four numbers: 1, 5, 13, or 89. Mod 144... This just means finding the remainder after dividing by 144. Here's the three-step process, using the actual Fibonacci prime 1597 as an example. 1. Divide the number by 144. 1597÷144=11.09. 2. Keep only the whole number part, then multiply back by 144. 11×144=1584 3. Subtract to find what's left over. 1597 - 1584 = 13 So 1597 "reduced mod 144" is 13 one of exactly the four allowed numbers. Try it with any of the other 20 known Fibonacci primes and you'll always land on 1, 5, 13, or 89. Never anything else. That's a 97% reduction. Out of 144 possible remainders, 140 of them are simply forbidden to Fibonacci primes. We proved this happens every single time, for every Fibonacci prime known, with zero exceptions. It gets stranger. Those four allowed remainders 1, 5, 13, 89 are themselves smaller numbers from the same Fibonacci sequence. Every Fibonacci prime, when you shrink it down this way, lands back on another Fibonacci prime. The sequence points back at itself. And the positions that produce those four numbers the 1st, 5th, 7th, and 11th spots in the sequence turn out to be exactly the same four numbers that mark critical boundaries in a completely separate math system,(PLCT) one based on multiplying the numbers 2 and 3 instead of adding golden-ratio powers. Two totally different number systems, built from different operations, share the same four "checkpoint" numbers. The animation traces 233 different ways to build the number 144 purely out of powers of the golden ratio (phi ≈ 1.618, the number where a whole equals its bigger part divided by its smaller part). 233 is itself a Fibonacci number the sequence even shows up in how many ways you can build it. Every flash you see is one valid combination, spiraling and glowing as it cycles through all of them. The Golden Lattice ϕ-Power Representations, the General Count Law, and the Fibonacci Prime Mod-144 Signature Theorem

Fibonacci numbers are the sequence where each one is the sum of the two before it: 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233… Every so often one of them is prime these are called Fibonacci primes, and only 21 are known. Here's the new discovery, take any Fibonacci prime (past the first few small ones) and reduce it to modulo 144. You'd expect the leftover remainder could be almost anything , 144 different possibilities. It never is. It only ever lands on one of just four numbers: 1, 5, 13, or 89. Mod 144... This just means finding the remainder after dividing by 144. Here's the three-step process, using the actual Fibonacci prime 1597 as an example. 1. Divide the number by 144. 1597÷144=11.09. 2. Keep only the whole number part, then multiply back by 144. 11×144=1584 3. Subtract to find what's left over. 1597 - 1584 = 13 So 1597 "reduced mod 144" is 13 one of exactly the four allowed numbers. Try it with any of the other 20 known Fibonacci primes and you'll always land on 1, 5, 13, or 89. Never anything else. That's a 97% reduction. Out of 144 possible remainders, 140 of them are simply forbidden to Fibonacci primes. We proved this happens every single time, for every Fibonacci prime known, with zero exceptions. It gets stranger. Those four allowed remainders 1, 5, 13, 89 are themselves smaller numbers from the same Fibonacci sequence. Every Fibonacci prime, when you shrink it down this way, lands back on another Fibonacci prime. The sequence points back at itself. And the positions that produce those four numbers the 1st, 5th, 7th, and 11th spots in the sequence turn out to be exactly the same four numbers that mark critical boundaries in a completely separate math system,(PLCT) one based on multiplying the numbers 2 and 3 instead of adding golden-ratio powers. Two totally different number systems, built from different operations, share the same four "checkpoint" numbers. The animation traces 233 different ways to build the number 144 purely out of powers of the golden ratio (phi ≈ 1.618, the number where a whole equals its bigger part divided by its smaller part). 233 is itself a Fibonacci number the sequence even shows up in how many ways you can build it. Every flash you see is one valid combination, spiraling and glowing as it cycles through all of them. The Golden Lattice ϕ-Power Representations, the General Count Law, and the Fibonacci Prime Mod-144 Signature Theorem

CTFTHEORY

12,946 просмотров • 1 месяц назад

Self Attention by hand ✍️ ~ 9 steps walkthrough below Self-attention is what enables LLMs to understand context. How does it work? So I drew and calculated one entirely by hand. Goal: turn four 6D features into four 3D attention weighted features, filling in every cell yourself. = 1. Given = Four feature vectors, six dimensions each, one per position. = 2. Query, key, value = Let us multiply the features by WQ, WK and WV. Queries, keys and values all come out of the same four features, and that is what the word "self" is doing in self-attention. = 3. Prepare for MatMul = We copy the queries across the top and the transposed keys down the side. Lining the two up is half the work. = 4. MatMul = Let us multiply K transpose by Q. Every cell is the dot product of one key with one query, which we use as a matching score. That works because the dot product is the numerator of cosine similarity: it is how alike two vectors are, before anyone divides by their lengths. = 5. Scale = We divide by the square root of dk, the dimension of a key vector, here 3. Without it the scores grow with the dimension and a 64-wide head would swamp the softmax. To keep the page doable in pen, the drawing approximates dividing by root 3 with halving. = 6. e to the power = Let us raise e to the power of each score. This is the first half of softmax, and the drawing uses 3 in place of e, which is close enough to do in your head. = 7. Sum = We add up each column: 16, 6, 7 and 12. = 8. Normalize = Let us divide every cell by its column sum. That gives the attention weight matrix in yellow, and each of its four columns is now a probability distribution over the four positions. The decimals are nudged as they are rounded, so every column still sums to exactly 1. = 9. MatMul = We multiply the value vectors by those weights. Each output is a blend of all four values, mixed in the proportion the attention matrix just decided, and it goes to the position-wise feed forward network in the next layer: the FFN box at the bottom of the page. The outputs: Attention weights (A), by column = [.2, .6, 0, .2], [.2, .4, .2, .2], [.4, .2, 0, .4], [.1, .7, .1, .1] Attention weighted features (Z) = [8, 2, 6], [8, 4, 4], [16, 4, 2], [4, 2, 7] The takeaway: attention is a weighted average, and everything before step 9 exists to decide the weights. Compare every position with every other, turn the scores into one distribution per position, then blend. 💾 Save this post!

Self Attention by hand ✍️ ~ 9 steps walkthrough below Self-attention is what enables LLMs to understand context. How does it work? So I drew and calculated one entirely by hand. Goal: turn four 6D features into four 3D attention weighted features, filling in every cell yourself. = 1. Given = Four feature vectors, six dimensions each, one per position. = 2. Query, key, value = Let us multiply the features by WQ, WK and WV. Queries, keys and values all come out of the same four features, and that is what the word "self" is doing in self-attention. = 3. Prepare for MatMul = We copy the queries across the top and the transposed keys down the side. Lining the two up is half the work. = 4. MatMul = Let us multiply K transpose by Q. Every cell is the dot product of one key with one query, which we use as a matching score. That works because the dot product is the numerator of cosine similarity: it is how alike two vectors are, before anyone divides by their lengths. = 5. Scale = We divide by the square root of dk, the dimension of a key vector, here 3. Without it the scores grow with the dimension and a 64-wide head would swamp the softmax. To keep the page doable in pen, the drawing approximates dividing by root 3 with halving. = 6. e to the power = Let us raise e to the power of each score. This is the first half of softmax, and the drawing uses 3 in place of e, which is close enough to do in your head. = 7. Sum = We add up each column: 16, 6, 7 and 12. = 8. Normalize = Let us divide every cell by its column sum. That gives the attention weight matrix in yellow, and each of its four columns is now a probability distribution over the four positions. The decimals are nudged as they are rounded, so every column still sums to exactly 1. = 9. MatMul = We multiply the value vectors by those weights. Each output is a blend of all four values, mixed in the proportion the attention matrix just decided, and it goes to the position-wise feed forward network in the next layer: the FFN box at the bottom of the page. The outputs: Attention weights (A), by column = [.2, .6, 0, .2], [.2, .4, .2, .2], [.4, .2, 0, .4], [.1, .7, .1, .1] Attention weighted features (Z) = [8, 2, 6], [8, 4, 4], [16, 4, 2], [4, 2, 7] The takeaway: attention is a weighted average, and everything before step 9 exists to decide the weights. Compare every position with every other, turn the scores into one distribution per position, then blend. 💾 Save this post!

Tom Yeh

27,211 просмотров • 1 месяц назад

Generative Adversarial Network (GAN) by hand ✍️ ~ 9 steps walkthrough below The Gen in GenAI came from this landmark paper by Ian Goodfellow et al., 12 years ago. The paper showed that a neural network can not only classify but also turn upside down to generate realistic looking images. The secret? We pit two of them against each other: a Generator turns noise into fake data, and a Discriminator learns to tell fake from real, pushing the Generator to keep doing better. One runs upside down, the other right way up. I drew and calculated one entirely by hand. Goal: generate realistic 4D data out of 2D noise, filling in every cell yourself. = 1. Given = Four noise vectors in 2D, and four real data vectors in 4D. = 2. Generator, first layer = Let us multiply the noise by weights and biases to get new features. = 3. ReLU = We apply the activation, and -1 and -2 are crossed out and set to 0. = 4. Generator, second layer = Let us multiply again. ReLU applies here too, but every value is already positive, so nothing changes. What comes out is the fake data F, made by a two-layer generator out of nothing but noise. = 5. Discriminator, first layer = We feed it both, the four fakes and the four real vectors, through the same weights. It never learns which is which from the layout, only from the numbers. = 6. Discriminator, second layer = Let us reduce each data vector to a single feature Z. Eight vectors in, eight numbers out. = 7. Sigmoid = We turn each Z into a probability Y. A 1 means the discriminator is certain the data is real, a 0 means certain it is fake. = 8. Training the Discriminator = Let us take the gradients as Y minus YD, where YD is what the discriminator should have said: 0 for the four fakes, 1 for the four real. Why so simple? Because pairing sigmoid with binary cross entropy loss makes the math collapse to exactly this subtraction. Its loss uses both halves of the page. = 9. Training the Generator = We do it again, as Y minus YG, and YG is [1, 1, 1, 1]: the generator wants the discriminator to call every fake real. Same predictions, different target, opposite goal. Its loss uses only the fakes. The outputs: Fake data F = [1, 2, 3, 1], [1, 1, 2, 1], [2, 2, 4, 2], [1, 0, 1, 1] Predictions on fakes = [.7, .5, .9, .3] Predictions on real = [.7, .9, .9, 1] Discriminator gradients = [.7, .5, .9, .3] and [-.3, -.1, -.1, 0] Generator gradients = [-.3, -.5, -.1, -.7] The takeaway: the adversarial part is one subtraction done twice. The same eight predictions, scored against two opposite targets, send one set of gradients back through the blue weights and another back through the green ones. 💾 Save this post!

Generative Adversarial Network (GAN) by hand ✍️ ~ 9 steps walkthrough below The Gen in GenAI came from this landmark paper by Ian Goodfellow et al., 12 years ago. The paper showed that a neural network can not only classify but also turn upside down to generate realistic looking images. The secret? We pit two of them against each other: a Generator turns noise into fake data, and a Discriminator learns to tell fake from real, pushing the Generator to keep doing better. One runs upside down, the other right way up. I drew and calculated one entirely by hand. Goal: generate realistic 4D data out of 2D noise, filling in every cell yourself. = 1. Given = Four noise vectors in 2D, and four real data vectors in 4D. = 2. Generator, first layer = Let us multiply the noise by weights and biases to get new features. = 3. ReLU = We apply the activation, and -1 and -2 are crossed out and set to 0. = 4. Generator, second layer = Let us multiply again. ReLU applies here too, but every value is already positive, so nothing changes. What comes out is the fake data F, made by a two-layer generator out of nothing but noise. = 5. Discriminator, first layer = We feed it both, the four fakes and the four real vectors, through the same weights. It never learns which is which from the layout, only from the numbers. = 6. Discriminator, second layer = Let us reduce each data vector to a single feature Z. Eight vectors in, eight numbers out. = 7. Sigmoid = We turn each Z into a probability Y. A 1 means the discriminator is certain the data is real, a 0 means certain it is fake. = 8. Training the Discriminator = Let us take the gradients as Y minus YD, where YD is what the discriminator should have said: 0 for the four fakes, 1 for the four real. Why so simple? Because pairing sigmoid with binary cross entropy loss makes the math collapse to exactly this subtraction. Its loss uses both halves of the page. = 9. Training the Generator = We do it again, as Y minus YG, and YG is [1, 1, 1, 1]: the generator wants the discriminator to call every fake real. Same predictions, different target, opposite goal. Its loss uses only the fakes. The outputs: Fake data F = [1, 2, 3, 1], [1, 1, 2, 1], [2, 2, 4, 2], [1, 0, 1, 1] Predictions on fakes = [.7, .5, .9, .3] Predictions on real = [.7, .9, .9, 1] Discriminator gradients = [.7, .5, .9, .3] and [-.3, -.1, -.1, 0] Generator gradients = [-.3, -.5, -.1, -.7] The takeaway: the adversarial part is one subtraction done twice. The same eight predictions, scored against two opposite targets, send one set of gradients back through the blue weights and another back through the green ones. 💾 Save this post!

Tom Yeh

16,752 просмотров • 1 месяц назад

Mike Benz Sends A Direct Message To The FBI and DOJ on How To Immediately Fix The J6 Pipebomber Crisis and Deliver Transparency To The American People: 1) Upload all 39,000 raw files of the pipebomber from different angles 2) Upload all 400+ interviews the FBI conducted. 3) Upload all FBI and DOJ forensic assessments 4) Upload every internal DOJ email and text message relative to the case 5) Run an internal search on main FBI agents and DOJ officials in charge of the investigation. 6) Trace the physical chain of custody for the cameras and footage 7) Request all data from the pipe bomber's phone "Let us be your long arm extension to help with this. And that gives us something to do with real evidence and real facts to add to our public body of record while you're continuing to run the traps internally." "We're over four years past it. You can't hide behind ongoing criminal investigation anymore. This has been four years. Help us help you."

Mike Benz Sends A Direct Message To The FBI and DOJ on How To Immediately Fix The J6 Pipebomber Crisis and Deliver Transparency To The American People: 1) Upload all 39,000 raw files of the pipebomber from different angles 2) Upload all 400+ interviews the FBI conducted. 3) Upload all FBI and DOJ forensic assessments 4) Upload every internal DOJ email and text message relative to the case 5) Run an internal search on main FBI agents and DOJ officials in charge of the investigation. 6) Trace the physical chain of custody for the cameras and footage 7) Request all data from the pipe bomber's phone "Let us be your long arm extension to help with this. And that gives us something to do with real evidence and real facts to add to our public body of record while you're continuing to run the traps internally." "We're over four years past it. You can't hide behind ongoing criminal investigation anymore. This has been four years. Help us help you."

Benny Johnson

143,013 просмотров • 9 месяцев назад

She Wore a Saree Before She Wore a School Bag👌🏼👏🏼👏🏼😍🥰😘 A tiny girl, barely four years old dressed in a crisp beige saree, walks confidently onto one of India's grandest stages. No nerves. No hesitation. Just quiet, unshakeable grace. She collects the National Film Award for Best Child Artist from the President of India and in that single moment, rewrites history forever. Her name is Treesha Vivek Thosar. And she just broke a record that stood for over six decades. 🎬 She acted at three. She won at four. Treesha played Chimi in the Marathi film Naal 2 an innocent rural child, full of wonder. She wasn't performing. She was living the role. At an age when most children are learning alphabets, she was moving a national jury with raw, natural emotion. 🏆 She broke Kamal Haasan's 60-year-old record. The legend himself posted publicly "You've beaten my record! I was already six when I got my first award.

She Wore a Saree Before She Wore a School Bag👌🏼👏🏼👏🏼😍🥰😘 A tiny girl, barely four years old dressed in a crisp beige saree, walks confidently onto one of India's grandest stages. No nerves. No hesitation. Just quiet, unshakeable grace. She collects the National Film Award for Best Child Artist from the President of India and in that single moment, rewrites history forever. Her name is Treesha Vivek Thosar. And she just broke a record that stood for over six decades. 🎬 She acted at three. She won at four. Treesha played Chimi in the Marathi film Naal 2 an innocent rural child, full of wonder. She wasn't performing. She was living the role. At an age when most children are learning alphabets, she was moving a national jury with raw, natural emotion. 🏆 She broke Kamal Haasan's 60-year-old record. The legend himself posted publicly "You've beaten my record! I was already six when I got my first award.

🪷 MЄЄƝƛ ƖƳЄƦ 🪷

171,784 просмотров • 3 месяцев назад

The astonishing 12,000-year-old #IceAge #Ratnagiri Petroglyph provides insight into the astronomical knowledge of our ancestors. This petroglyph features four prominent figures: 1. A man enclosed in a rectangular bracket with six grooves on alternate sides within the rectangle. On the outer bracket, there are four semi-spheres representing the four directions. Waves can be seen at the bottom which could be the sea or a river, while the top displays alternating rectangular shapes, possibly indicating an early understanding of the solar year. 2. The second figure resembles a semi-sphere with a head containing thirteen grooves. Below, there are straight lines suggesting some form of axis, which may represent the reconciliation of the lunar year with the solar year. 3. The third figure portrays a female character with 7-8 tentacles and a distinct head. Near the figure's belly, there is a 'V' sign. 4. The last figure is a giant man with a large foot. Additionally, smaller figures throughout the petroglyph may represent stars or constellations. From an aerial view, these four figures demonstrate the advanced understanding of ancient Indians during the last glacial period, approximately 12,000 years ago. Compiling the various pictures from different sources to create this meaningful video took nearly a month. Please note that none of the pictures belong to me; I am solely responsible for the compilation and interpretation. The pictures are owned by MyTemplesMyGlory 🇮🇳 on Twitter and Appa Yadav on Pinterest Kevinsandage Please like and subscribe the video at and help growing Indic Channel #Archaeology #Iceage

The astonishing 12,000-year-old #IceAge #Ratnagiri Petroglyph provides insight into the astronomical knowledge of our ancestors. This petroglyph features four prominent figures: 1. A man enclosed in a rectangular bracket with six grooves on alternate sides within the rectangle. On the outer bracket, there are four semi-spheres representing the four directions. Waves can be seen at the bottom which could be the sea or a river, while the top displays alternating rectangular shapes, possibly indicating an early understanding of the solar year. 2. The second figure resembles a semi-sphere with a head containing thirteen grooves. Below, there are straight lines suggesting some form of axis, which may represent the reconciliation of the lunar year with the solar year. 3. The third figure portrays a female character with 7-8 tentacles and a distinct head. Near the figure's belly, there is a 'V' sign. 4. The last figure is a giant man with a large foot. Additionally, smaller figures throughout the petroglyph may represent stars or constellations. From an aerial view, these four figures demonstrate the advanced understanding of ancient Indians during the last glacial period, approximately 12,000 years ago. Compiling the various pictures from different sources to create this meaningful video took nearly a month. Please note that none of the pictures belong to me; I am solely responsible for the compilation and interpretation. The pictures are owned by MyTemplesMyGlory 🇮🇳 on Twitter and Appa Yadav on Pinterest Kevinsandage Please like and subscribe the video at and help growing Indic Channel #Archaeology #Iceage

GemsOfINDOLOGY

95,399 просмотров • 2 лет назад

Marc Andreessen says the four most dangerous words in investing are "this time is different." But this time it's actually different. He's referring to the AI boom when he mentioned this. For 80 years, the industry has followed the same pattern. 1) Excitement builds 2) Money pours in 3) The technology fails to deliver 4) The industry collapses. Here are 4 examples of when this happened in AI: • In 1943, the original neural network paper was written. It proved that simple mathematical models of brain cells could perform any logical computation. It was a foundational breakthrough but didn't takeoff. • In 1955, scientists got a grant to spend 10 weeks at Dartmouth and believed they could achieve AGI by the end of the summer. They did not. • In the 1980s, corporations poured over a billion dollars into "expert systems," AI programs designed to replicate the decision-making of human specialists in medicine, finance, and engineering. By 1987, this entire market crashed. • In 2016-17, machine learning hype surged again. It faded quickly. Each time, the technology underdelivered and the money dried up. But here's what really different about this time. AI has made four breakthroughs back to back. 1) Language models 2) Reasoning 3) Coding agents 4) Self-improvement And ALL are generating revenue in production right now. That's why old Nvidia chips are gaining value instead of depreciating. And this has literally never happened in computing before. Every GPU is sold out for the next 3-4 years. He calls betting against this "essentially suicidal" and "an invitation to get your face ripped off." He calls it "an 80-year overnight success." Researchers worked their entire lives on this, and many passed away without seeing it work. Now it is all arriving at once. — Marc Andreessen (Marc Andreessen 🇺🇸), co-founder at a16z (a16z) on the Latent Space podcast (Latent.Space)

Marc Andreessen says the four most dangerous words in investing are "this time is different." But this time it's actually different. He's referring to the AI boom when he mentioned this. For 80 years, the industry has followed the same pattern. 1) Excitement builds 2) Money pours in 3) The technology fails to deliver 4) The industry collapses. Here are 4 examples of when this happened in AI: • In 1943, the original neural network paper was written. It proved that simple mathematical models of brain cells could perform any logical computation. It was a foundational breakthrough but didn't takeoff. • In 1955, scientists got a grant to spend 10 weeks at Dartmouth and believed they could achieve AGI by the end of the summer. They did not. • In the 1980s, corporations poured over a billion dollars into "expert systems," AI programs designed to replicate the decision-making of human specialists in medicine, finance, and engineering. By 1987, this entire market crashed. • In 2016-17, machine learning hype surged again. It faded quickly. Each time, the technology underdelivered and the money dried up. But here's what really different about this time. AI has made four breakthroughs back to back. 1) Language models 2) Reasoning 3) Coding agents 4) Self-improvement And ALL are generating revenue in production right now. That's why old Nvidia chips are gaining value instead of depreciating. And this has literally never happened in computing before. Every GPU is sold out for the next 3-4 years. He calls betting against this "essentially suicidal" and "an invitation to get your face ripped off." He calls it "an 80-year overnight success." Researchers worked their entire lives on this, and many passed away without seeing it work. Now it is all arriving at once. — Marc Andreessen (Marc Andreessen 🇺🇸), co-founder at a16z (a16z) on the Latent Space podcast (Latent.Space)

GeniusThinking

51,298 просмотров • 4 месяцев назад

🚨 TODAY MAY BE THE MOST DANGEROUS MARKET OPEN OF 2026 Four separate macro risks are colliding at the same time, and markets are priced like none of them exist 1. US-Iran peace deal is reportedly on the verge of collapsing - oil shock risk is back on the table 2. Fed dot plot now explicitly prices a rate hike in 2026 - Warsh delivered that signal this week 3. China, Japan and Turkey are actively reducing US Treasury exposure -bond market stress is building 4. JP Morgan is set to rebalance $165B in US equity exposure Monday -that is mechanical selling regardless of sentiment Each of those is a standalone market mover All four landing together is a different conversation entirely Three ways this plays out: > Headlines stabilize fast - light shock, brief vol spike, risk recovers by Wednesday > Iran deal collapses - oil reprices hard, energy sector bids, everything else sells > Full breakdown - oil surges, yields spike, liquidity conditions tighten overnight, risk assets dump simultaneously and there's no orderly exit The third scenario is what keeps me up at night Not because it's certain, but because the system has no margin for error right now Fragile conditions don't need a large shock - they need just one I would consider watching oil, 10-year yield and credit spread closely Those three tell you which scenario you're in before equities do Follof + notifs on, I will keep you updated

🚨 TODAY MAY BE THE MOST DANGEROUS MARKET OPEN OF 2026 Four separate macro risks are colliding at the same time, and markets are priced like none of them exist 1. US-Iran peace deal is reportedly on the verge of collapsing - oil shock risk is back on the table 2. Fed dot plot now explicitly prices a rate hike in 2026 - Warsh delivered that signal this week 3. China, Japan and Turkey are actively reducing US Treasury exposure -bond market stress is building 4. JP Morgan is set to rebalance $165B in US equity exposure Monday -that is mechanical selling regardless of sentiment Each of those is a standalone market mover All four landing together is a different conversation entirely Three ways this plays out: > Headlines stabilize fast - light shock, brief vol spike, risk recovers by Wednesday > Iran deal collapses - oil reprices hard, energy sector bids, everything else sells > Full breakdown - oil surges, yields spike, liquidity conditions tighten overnight, risk assets dump simultaneously and there's no orderly exit The third scenario is what keeps me up at night Not because it's certain, but because the system has no margin for error right now Fragile conditions don't need a large shock - they need just one I would consider watching oil, 10-year yield and credit spread closely Those three tell you which scenario you're in before equities do Follof + notifs on, I will keep you updated

Pepesso

18,357 просмотров • 2 месяцев назад

At exactly 6am on 20th August 2019, four people entered a brand new studio and started a journey along an uncharted path. When I switched on the four microphones that morning I was excited at launching yet another station, but I also had butterflies in my stomach; would this work? Would the audience embrace four hours of non-stop talk during their busy morning commute? The four of us with CT Muga, Ndu Okoh and Njeri Thorne took on the new assignment with dedication and a true belief that we could do it. Outside of that studio was our producer Brian Evusa, the Programmes Controller Brian Obara and, although he wasn’t in the studio, the Radio General Manager Tom Japanni was present. It has been a great journey building The Situation Room to what it has become and we have very many people to thank, not least the team of producers, marketing, commercial, technical and other colleagues. The Situation Room has become the home of Kenya’s Biggest Conversations and it continues to host many thought leaders, community influencers, different professionals and duty bearers. The biggest part of this show has been the audience who tune in on radio, Tv or digital platforms. It’s humbling whenever people reach out on calls, text or approach us with compliments, comments and suggestions for the show. On this 6th day of January 2025, my heart is heavy as I bid kwaheri to the brilliant team of colleagues at Spice FM and the Standard Group. I am grateful for the opportunity to once again be a part of Kenya’s oldest private media house and I am proud of the teams that I leave behind in radio, TV and print. I love the thrill of creating and starting new things and I have been blessed to do just that. I was privileged to be there at the launch of Kenya’s first 24hr religious media house - Family FM and Family TV, the first 24hrs vernacular radio station - Kameme FM and the first 24hr news and information TV station - K24. Is there room for something new? Of course, always, and I sure hope to be a part of it soon. Wish me luck, I will keep you posted.

Eric Latiff

351,588 просмотров • 1 год назад

🌟Our latest LangChain Academy course – Deep Agents with LangGraph – is now live!🌟 Many agents today follow the same simple pattern: run in a loop, call tools. That architecture works well enough, but it breaks down as tasks get more complex. Today, companies of all sizes – from startups to large enterprises – are building their own Deep Agents. These agents dive deeper. They’re able to plan complex tasks and carry them out over longer time horizons. There are four key features that set Deep Agents apart from regular agents: 1. Planning – keeps agents on track 2. File system – allows agents to offload context 3. Sub-agents – act as focused specialists 4. Prompting – provides agents with detailed instructions Our latest LangChain Academy course, Deep Agents with LangGraph, shows you how to combine these pieces with LangGraph to orchestrate long-running, multi-agent workflows. Big thanks to community member Dmitry Labazkin for helping us shape this course with his contributions! Enroll for free ➡️

🌟Our latest LangChain Academy course – Deep Agents with LangGraph – is now live!🌟 Many agents today follow the same simple pattern: run in a loop, call tools. That architecture works well enough, but it breaks down as tasks get more complex. Today, companies of all sizes – from startups to large enterprises – are building their own Deep Agents. These agents dive deeper. They’re able to plan complex tasks and carry them out over longer time horizons. There are four key features that set Deep Agents apart from regular agents: 1. Planning – keeps agents on track 2. File system – allows agents to offload context 3. Sub-agents – act as focused specialists 4. Prompting – provides agents with detailed instructions Our latest LangChain Academy course, Deep Agents with LangGraph, shows you how to combine these pieces with LangGraph to orchestrate long-running, multi-agent workflows. Big thanks to community member Dmitry Labazkin for helping us shape this course with his contributions! Enroll for free ➡️

LangChain

63,349 просмотров • 11 месяцев назад

Dad Films Hilarious Moment When Four Giant Dogs Realize They’re Headed to Their Favorite Place Do you remember the pure excitement of childhood surprises? One minute you think you’re heading to the grocery store for a boring errand, and the next your parents suddenly turn the car toward a theme park or the movies. That rush of unexpected joy makes your heart race and a huge smile spread across your face. As it turns out, this feeling isn’t limited to humans. Our four-legged friends—especially dogs—experience the same kind of delightful surprises, and sometimes they show it even more dramatically than we do. For most dogs, getting into the car usually signals one of two things: a trip to the vet (definitely not their favorite destination) or a fun outing to the park. Many dogs are surprisingly good at reading the route. A simple left or right turn after leaving the neighborhood can tell them exactly where they’re headed. Plenty of dog owners have stories of their childhood pets reacting instantly—right turn meant the dreaded vet, left turn meant adventure and freedom. This canine intuition shines perfectly in a viral video featuring four energetic German Shorthaired Pointers, or GSPs. These powerful hunting dogs are known for their sleek coats, athletic builds, and boundless energy. Packed into the car together, the four giant pups start off relatively calm. But the moment they realize the car isn’t heading toward the vet’s office, pure chaos erupts. They burst into loud, continuous howling that sounds exactly like a pack of wolves calling to one another. They jump up and down, tails wagging wildly, mouths open wide, and eyes sparkling with pure delight. Through all the noise, their dad stays impressively patient. He just smiles, clearly enjoying every second of his dogs’ explosive happiness. The howling isn’t complaining—it’s pure celebration and gratitude. It’s their way of saying, “Thank you for taking us to the best place ever!” Moments like this remind us how easy it is to create pure joy for our pets. A simple surprise trip to the park, a longer walk than usual, or even just taking a different route can turn an ordinary day into something special. Dogs don’t ask for much. They just want time with us, space to play, and the chance to feel loved. So the next time you load your furry friend into the car, consider adding a little surprise. Watch how quickly their excitement can light up the entire day—for both of you. Those four giant GSPs prove that sometimes the greatest happiness is waiting just around an unexpected turn. What’s the funniest or most exciting reaction your dog has ever had when realizing a trip to the park (or somewhere special) was happening? Share your story in the comments!

Dad Films Hilarious Moment When Four Giant Dogs Realize They’re Headed to Their Favorite Place Do you remember the pure excitement of childhood surprises? One minute you think you’re heading to the grocery store for a boring errand, and the next your parents suddenly turn the car toward a theme park or the movies. That rush of unexpected joy makes your heart race and a huge smile spread across your face. As it turns out, this feeling isn’t limited to humans. Our four-legged friends—especially dogs—experience the same kind of delightful surprises, and sometimes they show it even more dramatically than we do. For most dogs, getting into the car usually signals one of two things: a trip to the vet (definitely not their favorite destination) or a fun outing to the park. Many dogs are surprisingly good at reading the route. A simple left or right turn after leaving the neighborhood can tell them exactly where they’re headed. Plenty of dog owners have stories of their childhood pets reacting instantly—right turn meant the dreaded vet, left turn meant adventure and freedom. This canine intuition shines perfectly in a viral video featuring four energetic German Shorthaired Pointers, or GSPs. These powerful hunting dogs are known for their sleek coats, athletic builds, and boundless energy. Packed into the car together, the four giant pups start off relatively calm. But the moment they realize the car isn’t heading toward the vet’s office, pure chaos erupts. They burst into loud, continuous howling that sounds exactly like a pack of wolves calling to one another. They jump up and down, tails wagging wildly, mouths open wide, and eyes sparkling with pure delight. Through all the noise, their dad stays impressively patient. He just smiles, clearly enjoying every second of his dogs’ explosive happiness. The howling isn’t complaining—it’s pure celebration and gratitude. It’s their way of saying, “Thank you for taking us to the best place ever!” Moments like this remind us how easy it is to create pure joy for our pets. A simple surprise trip to the park, a longer walk than usual, or even just taking a different route can turn an ordinary day into something special. Dogs don’t ask for much. They just want time with us, space to play, and the chance to feel loved. So the next time you load your furry friend into the car, consider adding a little surprise. Watch how quickly their excitement can light up the entire day—for both of you. Those four giant GSPs prove that sometimes the greatest happiness is waiting just around an unexpected turn. What’s the funniest or most exciting reaction your dog has ever had when realizing a trip to the park (or somewhere special) was happening? Share your story in the comments!

Emotion & Music

698,969 просмотров • 8 дней назад

sonnet 5 vs sonnet 4.6 vs opus 4.8 vs glm 5.2 – frontend tasks dropped sonnet 5 into a quick test today. same three prompts to all four models, single-shot html/canvas, no edits: • objects falling on a trampoline • rockets playing tennis • a slingshot breaking bottles ranked by speed (total across the 3 tasks): 1. opus 4.8 – 15m 09s 2. sonnet 5 – 16m 05s 3. glm 5.2 – 27m 18s 4. sonnet 4.6 – 35m 06s ranked by code shortness (total loc): 1. sonnet 5 – 1794 2. opus 4.8 – 2063 3. sonnet 4.6 – 2182 4. glm 5.2 – 3285 sonnet 5 came out on top here – leanest code overall and a near-tie for fastest it was also the most creative. in every task it added something none of the others did: – kept the trampoline vibrating after the objects landed – drew a +1 next to the rocket that scored the point – turned the slingshot to face the next bottle before each shot opus 4.8 evaluated the code sonnet 5 produced. four things stood out: • the sphere is a fake, and that's the smart move. the cube and star are real 3d meshes with proper culling and shading, but the ball is just a flat shaded circle. a lit sphere looks identical from every angle, so building it in 3d would burn compute for zero visible payoff. knowing where not to bother is its own kind of skill • weight actually means something on the trampoline. the star is heavy, so it barely bounces and dents the mat hard. the ball is light, so it's lively and leaves a shallow dip. the three objects aren't just different shapes – they have different temperaments, and the physics is what gives them that • the slingshot is framed like a shot, not just drawn. the handle is anchored below the bottom of the screen and runs off-frame, so it reads as something you're holding rather than a sprite parked in the scene. that's a staging instinct, not a rendering one • the paddle ai forward-simulates the ball to predict where it'll land, then adds a deliberate error bias (roughly 1 in 5 shots is a real miss). that's why scoring looks natural instead of robotic – plus four distinct fault types with a catch-all so a rally never hangs without a result bottom line: sonnet 5 does more with less. fastest tier, leanest code, and the only one that added small touches nobody asked for follow thehype. for 24/7 ai news, analysis and breakdowns

sonnet 5 vs sonnet 4.6 vs opus 4.8 vs glm 5.2 – frontend tasks dropped sonnet 5 into a quick test today. same three prompts to all four models, single-shot html/canvas, no edits: • objects falling on a trampoline • rockets playing tennis • a slingshot breaking bottles ranked by speed (total across the 3 tasks): 1. opus 4.8 – 15m 09s 2. sonnet 5 – 16m 05s 3. glm 5.2 – 27m 18s 4. sonnet 4.6 – 35m 06s ranked by code shortness (total loc): 1. sonnet 5 – 1794 2. opus 4.8 – 2063 3. sonnet 4.6 – 2182 4. glm 5.2 – 3285 sonnet 5 came out on top here – leanest code overall and a near-tie for fastest it was also the most creative. in every task it added something none of the others did: – kept the trampoline vibrating after the objects landed – drew a +1 next to the rocket that scored the point – turned the slingshot to face the next bottle before each shot opus 4.8 evaluated the code sonnet 5 produced. four things stood out: • the sphere is a fake, and that's the smart move. the cube and star are real 3d meshes with proper culling and shading, but the ball is just a flat shaded circle. a lit sphere looks identical from every angle, so building it in 3d would burn compute for zero visible payoff. knowing where not to bother is its own kind of skill • weight actually means something on the trampoline. the star is heavy, so it barely bounces and dents the mat hard. the ball is light, so it's lively and leaves a shallow dip. the three objects aren't just different shapes – they have different temperaments, and the physics is what gives them that • the slingshot is framed like a shot, not just drawn. the handle is anchored below the bottom of the screen and runs off-frame, so it reads as something you're holding rather than a sprite parked in the scene. that's a staging instinct, not a rendering one • the paddle ai forward-simulates the ball to predict where it'll land, then adds a deliberate error bias (roughly 1 in 5 shots is a real miss). that's why scoring looks natural instead of robotic – plus four distinct fault types with a catch-all so a rally never hangs without a result bottom line: sonnet 5 does more with less. fastest tier, leanest code, and the only one that added small touches nobody asked for follow thehype. for 24/7 ai news, analysis and breakdowns

thehype.

14,518 просмотров • 1 месяц назад

DFlash2 makes Qwen3.8 27B up to 4x faster⚡️ We rented one RTX 6000 and ran the same Qwen3.8-27B four ways: baseline, MTP, DFlash, DFlash2. We picked 4 different tasks for the test: a music library as JSON, a Python rate limiter, a seating puzzle, a heist story. Outputs: Baseline: 47.4 tok/s MTP: 114.7 tok/s · 52% accepted DFlash: 99.3 tok/s · 34% accepted DFlash2: 140.6 tok/s · 56% accepted The diff between DFlash and DFlash2 is essentially this: DFlash only looks at the current token, makes one guess and sends it, which often results in poor acceptance rates. DFlash2 on the other hand looks at the previous and current tokens, creates 16 candidates and ranks each guess on two criteria: 1. how well it fits as the next token 2. does it connect with the previous token. After ranking, it picks the best candidate and submits it as the guess. This is how DFlash2 achieves much higher acceptance rates and as a result is often much faster than DFlash

DFlash2 makes Qwen3.8 27B up to 4x faster⚡️ We rented one RTX 6000 and ran the same Qwen3.8-27B four ways: baseline, MTP, DFlash, DFlash2. We picked 4 different tasks for the test: a music library as JSON, a Python rate limiter, a seating puzzle, a heist story. Outputs: Baseline: 47.4 tok/s MTP: 114.7 tok/s · 52% accepted DFlash: 99.3 tok/s · 34% accepted DFlash2: 140.6 tok/s · 56% accepted The diff between DFlash and DFlash2 is essentially this: DFlash only looks at the current token, makes one guess and sends it, which often results in poor acceptance rates. DFlash2 on the other hand looks at the previous and current tokens, creates 16 candidates and ranks each guess on two criteria: 1. how well it fits as the next token 2. does it connect with the previous token. After ranking, it picks the best candidate and submits it as the guess. This is how DFlash2 achieves much higher acceptance rates and as a result is often much faster than DFlash

atomic.chat

27,777 просмотров • 8 дней назад

🚨THE CRASH NOBODY'S READY FOR IS HERE For over a decade, Bitcoin has moved to the same beat: 2-3 years up, then a brutal ~1-year purge. Halving, rally, euphoria, collapse. Repeat. It's happened four times already 2013, 2017, 2021, 2025. People treat it like gravity. And that's exactly why the crowd is convinced this time is different that spot ETFs and Wall Street money finally broke the cycle for good. I don't think they broke it. I think they just changed who's holding the bag when it turns. Every past cycle was retail-driven forums, exchanges, leverage that unwound fast and violently, then got forgotten in a bull market two years later. This cycle got captured by institutions. Which means when the real move comes, it won't be degens getting liquidated on a random exchange. It'll be pension allocations, ETF flows, and corporate treasuries unwinding in public, all at once, in daylight. The warning lights are already on: → The softest drawdown in Bitcoin's history which historically means the real leg down hasn't even started → Corporate balance sheets (Strategy and the copycats that followed) now sitting on paper losses if price keeps sliding → "This time is different" louder than ever the exact phrase people used in 2017 and again in 2021 → Retail interest already fading while price is still well above the last cycle's bottom → A macro backdrop nobody priced in when the euphoria was building A pattern that's repeated four times doesn't need a fifth confirmation to keep working. It just needs one moment where everyone stops believing it will. I'm not writing this to scare anyone. I've watched this exact chart before obvious in hindsight, mocked in the moment. When the real move comes, it won't wait for the crowd to agree. It never does. Follow now, notifs on, so you're early instead of surprised.

🚨THE CRASH NOBODY'S READY FOR IS HERE For over a decade, Bitcoin has moved to the same beat: 2-3 years up, then a brutal ~1-year purge. Halving, rally, euphoria, collapse. Repeat. It's happened four times already 2013, 2017, 2021, 2025. People treat it like gravity. And that's exactly why the crowd is convinced this time is different that spot ETFs and Wall Street money finally broke the cycle for good. I don't think they broke it. I think they just changed who's holding the bag when it turns. Every past cycle was retail-driven forums, exchanges, leverage that unwound fast and violently, then got forgotten in a bull market two years later. This cycle got captured by institutions. Which means when the real move comes, it won't be degens getting liquidated on a random exchange. It'll be pension allocations, ETF flows, and corporate treasuries unwinding in public, all at once, in daylight. The warning lights are already on: → The softest drawdown in Bitcoin's history which historically means the real leg down hasn't even started → Corporate balance sheets (Strategy and the copycats that followed) now sitting on paper losses if price keeps sliding → "This time is different" louder than ever the exact phrase people used in 2017 and again in 2021 → Retail interest already fading while price is still well above the last cycle's bottom → A macro backdrop nobody priced in when the euphoria was building A pattern that's repeated four times doesn't need a fifth confirmation to keep working. It just needs one moment where everyone stops believing it will. I'm not writing this to scare anyone. I've watched this exact chart before obvious in hindsight, mocked in the moment. When the real move comes, it won't wait for the crowd to agree. It never does. Follow now, notifs on, so you're early instead of surprised.

Shelpid.WI3M

39,616 просмотров • 21 дней назад

qwen 3.8 max vs deepseek v4 flash 0731 vs kimi k3 vs gpt 5.6 sol – on rubik's cube and chess four frontier models built a rubik's cube stand and solved it, then built a chess board and played claude opus 5 on it the setup: Nous Research's hermes agent cli on OpenRouter tasks: 1. cube – build a 3d rubik's cube with a cli and a Three.js viewer, then solve an identical scrambled position on your own stand 2. chess – build a 3d chess stand, then play white against claude opus 5 as black, live, one move at a time. no engine, no solver, no opening book on either side. stockfish depth 14 grades every chess ply afterwards; neither player sees the score models: DeepSeek v4 flash 0731, OpenAI gpt-5.6 sol, Kimi.ai kimi k3, Qwen qwen 3.8 max gpt-5.6 sol and deepseek v4 flash solved their cubes – sol in 24 moves and seventeen seconds, deepseek in 32. qwen and kimi never got there, giving up at 96 and 207 moves then all four built chess stands and played white against claude opus 5 on them, and all four resigned: deepseek on move 13, sol on 19, kimi on 21, qwen holding out longest at 29 - build time, both stands #1 gpt-5.6 sol – 16m 43s #2 deepseek v4 flash – 97m 39s #3 kimi k3 – 166m 09s #4 qwen 3.8 max – 215m 08s - build attempts before a working stand #1 gpt-5.6 sol – 3 #2 qwen 3.8 max – 4 #3 kimi k3 – 4 #4 deepseek v4 flash – 5 - total tokens #1 gpt-5.6 sol – 6,713,754 #2 qwen 3.8 max – 17,272,507 #3 kimi k3 – 22,427,504 #4 deepseek v4 flash – 27,417,442 - total price #1 deepseek v4 flash – $0.557 #2 gpt-5.6 sol – $6.319 #3 qwen 3.8 max – $10.270 #4 kimi k3 – $16.667 observations: • deepseek v4 flash is the cheapest model here by a margin nobody else is near, and it got there while being the least efficient of the four. it burned 27.4m tokens – more than anyone, 5m more than kimi – and still finished both benchmarks for $0.557. that is $0.02 per million tokens against kimi's $0.74. it also needed the most passes to produce working stands, five, and that did not matter: all five deepseek passes together cost a thirtieth of kimi's two • so what deepseek cannot do is get it right the first time. what it can do is get it right the fifth time, for half a dollar. that is a different thing to be buying – not a good first draft, but the option to keep asking • gpt-5.6 sol is the opposite profile and the strongest of the four on pure efficiency. 16m 43s to build both stands, 6.7m tokens, three passes – under 40% of the next lowest token count and a quarter of deepseek's, on an eighth of qwen's clock. it also solved the cube fastest of anyone, 24 moves in seventeen seconds. sol is what you reach for when you want the answer now and can absorb $0.94 per million • sol's weakness is in what it does not check. its chess viewer deleted the capturing piece instead of the captured one, so pieces disappeared off the board mid-game – a defect the fifty-cent deepseek stand did not have. fast and terse turns out to be the same dial as fast and unverified • qwen 3.8 max is not the cheap open-weights option it gets treated as. $10.270 across the two benchmarks, second most expensive of the four, 18x deepseek, and by a distance the slowest – 215 minutes of build time, nearly thirteen times sol's. what the money buys is judgment: it played eighteen moves without a single error worth a hundredth of a pawn, then made exactly one bad move in the whole game, and averaged 44.6 centipawns lost across the longest game any of the four managed. it also could not solve a rubik's cube in 96 tries • kimi k3 is the one line with no reading that flatters it. most expensive at $16.667, last on the cube at 207 moves, last at chess at 478 centipawns lost per move. it is also the model that verified hardest – on the cube it wrote its own integrity check instead of trusting its output. that makes the result worse rather than better: the checking was real, and the reasoning underneath it still was not follow thehype. for 24/7 ai news, analysis and breakdowns

thehype.

84,595 просмотров • 23 дней назад

One of the best applications for agentic loops is SEO. Here's why it works: the outcome is "black or white." You moved up this month, you moved down, or you stayed the same. That's the entire requirement for a loop — a metric that grades the work for you. Build it in five parts: 1) Access. - Connect it to your real data, not your description of your data. - Google Search Console has an API. - DataForSEO shows you the four pages ranking above yours. 2) Action. - Let it audit and fix before it experiments. - Meta tags, JSON-LD, a missing sitemap. - Quick wins on most sites. 3) Memory. - A markdown file where it logs every change it made and why. 4) Judgment. - Every two weeks it reopens that file and checks its own work: I rewrote that description — did it rank us higher or lower? 5) Reversal. - A change that took you from position 20 to 30 gets undone. - Nothing here is set in stone. Then put it on a routine so it picks up where it left off without you. The catch: this takes months, not days. Months of nothing, then month four, bam bam bam, page one. So find the number in your business that can't be argued with. Build the loop around that one.

One of the best applications for agentic loops is SEO. Here's why it works: the outcome is "black or white." You moved up this month, you moved down, or you stayed the same. That's the entire requirement for a loop — a metric that grades the work for you. Build it in five parts: 1) Access. - Connect it to your real data, not your description of your data. - Google Search Console has an API. - DataForSEO shows you the four pages ranking above yours. 2) Action. - Let it audit and fix before it experiments. - Meta tags, JSON-LD, a missing sitemap. - Quick wins on most sites. 3) Memory. - A markdown file where it logs every change it made and why. 4) Judgment. - Every two weeks it reopens that file and checks its own work: I rewrote that description — did it rank us higher or lower? 5) Reversal. - A change that took you from position 20 to 30 gets undone. - Nothing here is set in stone. Then put it on a routine so it picks up where it left off without you. The catch: this takes months, not days. Months of nothing, then month four, bam bam bam, page one. So find the number in your business that can't be argued with. Build the loop around that one.

The Startup Ideas Podcast (SIP) 🧃

11,505 просмотров • 21 дней назад

Almost 20 years later, AWS is still the most popular cloud in the world. The reason is simple: it just works! They have four services focused on Generative AI: 1. Amazon Q 2. Amazon Bedrock 3. SageMaker JumpStart 4. PartyRock I've been using AWS for around 15 years (honestly, I don't remember well), and their Developer Center is a gold mine. If you open their Developer Center, you'll find a new learning path, "Generative AI for Developers." I'm linking to it below. This is not just a course. This is a collection of courses, examples, videos, tutorials, and code walkthroughs. They will teach you how to use Generative AI on AWS using the four services above. ↑ That right there is a huge selling point: These classes aren't just theoretical. You'll have a chance to learn using the same professional tools everyone else uses. By the way, there are many more resources in the Developer Center: • Machine Learning • Data Operations • DevOps All of these are free. Click, click, and start learning right away. One more thing before I forget: If you are building anything with AWS, check out Amazon Q, their coding assistant. This is the *best* coding assistant for AWS-related work, and it's not even close. It's a Visual Studio Code extension. Install it, and it works like any other code assistant, except this one knows a lot about AWS. Thanks to AWS for sponsoring a post about their free courses and learning resources. There's a special place in Developer Heaven for you.

Almost 20 years later, AWS is still the most popular cloud in the world. The reason is simple: it just works! They have four services focused on Generative AI: 1. Amazon Q 2. Amazon Bedrock 3. SageMaker JumpStart 4. PartyRock I've been using AWS for around 15 years (honestly, I don't remember well), and their Developer Center is a gold mine. If you open their Developer Center, you'll find a new learning path, "Generative AI for Developers." I'm linking to it below. This is not just a course. This is a collection of courses, examples, videos, tutorials, and code walkthroughs. They will teach you how to use Generative AI on AWS using the four services above. ↑ That right there is a huge selling point: These classes aren't just theoretical. You'll have a chance to learn using the same professional tools everyone else uses. By the way, there are many more resources in the Developer Center: • Machine Learning • Data Operations • DevOps All of these are free. Click, click, and start learning right away. One more thing before I forget: If you are building anything with AWS, check out Amazon Q, their coding assistant. This is the best coding assistant for AWS-related work, and it's not even close. It's a Visual Studio Code extension. Install it, and it works like any other code assistant, except this one knows a lot about AWS. Thanks to AWS for sponsoring a post about their free courses and learning resources. There's a special place in Developer Heaven for you.

Santiago

22,104 просмотров • 1 год назад

When it comes to the Lindsay Clancy trial, multiple things are true. 1. Doctors don’t take seriously enough the danger of SSRIs. Clancy was prescribed 13 different medications - taken sporadically - by six different doctors and nurses in a period of four months. Most of these medications came with the potential side effects of worsening depression, anxiety, and suicidal thoughts. Many of the medications had interactions with each other. 2. The drugs and her possible psychosis do not automatically absolve her from guilt. She carefully planned her children’s murders and she carried them out methodically. The strength and resolve required to strangle toddlers and an infant one by one may demonstrate a willful, knowing plan to kill. We cannot know for sure whether these medications are the primary or even secondary cause of her actions. 3. Most people who commit violent crimes likely suffer from some level of mental sickness. That doesn’t negate the need for justice. 4. We should take more seriously the effects of postpartum psychosis and/or depression. There was no reason for Lindsay ever to be left alone with her children. This likely could have been prevented. 5. Millions of women suffer from mental breakdowns postpartum and do not murder their children with exercise bands. What precedent does it set if she’s deemed legally irresponsible and is able to be released from psychiatric care for just a few years - after committing triple homicide? 6. Most important: the primary victims are the CHILDREN. They deserve most of our compassion. Their pain should be our primary focus. My full analysis:

When it comes to the Lindsay Clancy trial, multiple things are true. 1. Doctors don’t take seriously enough the danger of SSRIs. Clancy was prescribed 13 different medications - taken sporadically - by six different doctors and nurses in a period of four months. Most of these medications came with the potential side effects of worsening depression, anxiety, and suicidal thoughts. Many of the medications had interactions with each other. 2. The drugs and her possible psychosis do not automatically absolve her from guilt. She carefully planned her children’s murders and she carried them out methodically. The strength and resolve required to strangle toddlers and an infant one by one may demonstrate a willful, knowing plan to kill. We cannot know for sure whether these medications are the primary or even secondary cause of her actions. 3. Most people who commit violent crimes likely suffer from some level of mental sickness. That doesn’t negate the need for justice. 4. We should take more seriously the effects of postpartum psychosis and/or depression. There was no reason for Lindsay ever to be left alone with her children. This likely could have been prevented. 5. Millions of women suffer from mental breakdowns postpartum and do not murder their children with exercise bands. What precedent does it set if she’s deemed legally irresponsible and is able to be released from psychiatric care for just a few years - after committing triple homicide? 6. Most important: the primary victims are the CHILDREN. They deserve most of our compassion. Their pain should be our primary focus. My full analysis:

Allie Beth Stuckey

106,839 просмотров • 22 дней назад

U-Net by hand ✍️ ~ 17 steps walkthrough below I consider U-Net as a key milestone in deep learning, the first image-to-image model that really worked! It came out of medical imaging, an unusual place, not from NeurIPS or CVPR or ACL. Now it is the backbone of diffusion models, which you see in almost all modern image generation models. I drew the network as a C so the matrix multiplication flows naturally down. Tilt your head to the right and it is a U again. 🤣 Goal: push a 3 x 16 image down to a 2 x 4 bottleneck and back out again, filling in every cell yourself. = 1. Given = An image of three channels, R, G and B, sixteen pixels wide, and every kernel the network will use. = 2. Convolution 1 = Let us slide the first kernel over the image. Each output is one multiply-and-add over a 2 x 3 window, and the result is the green feature map. = 3. Find the maxima = We circle the largest value in each 1 x 2 window. Circling first is worth the extra step: it is the pooling decision, made before anything is written down. = 4. Max pool 1 = Let us copy those maxima down. Sixteen columns become eight, and half the detail is gone for good. = 5. Convolution 2 = We convolve again with the second kernel, deeper into the contracting path. The feature map is blue now. = 6. Find the maxima again = Same move as step 3, on the blue map. = 7. Max pool 2 = Eight columns become four. = 8. The bottleneck = Let us convolve once more. This is the bottom of the U, a 2 x 4 block that is everything the network kept. = 9. Spread it out = We start back up. The transposed convolution writes each bottleneck value into a wider grid, leaving gaps between them. = 10. Transposed convolution 1 = Let us fill those gaps by convolving over the spread-out grid. Four columns become eight. = 11. The first skip = We copy the encoder's matching row straight across. This is the skip connection, and it is the whole reason a U-Net can recover detail that pooling threw away. = 12. Convolution with the skip = Let us convolve the upsampled features together with the copied ones. = 13. Spread it out again = Same as step 9, one level up. = 14. Transposed convolution 2 = Eight columns become sixteen, back to the width we started at. = 15. The second skip = The encoder's first feature map comes across, the one made before any pooling happened. = 16. Convolution and ReLU = We convolve, then cross out every negative and set it to zero. = 17. Output convolution = Let us apply the last kernel. Out comes R', G' and B', an image the same size as the one we started with. The outputs: R' = [3, 0, 7, 0, 7, 0, 17, 0, 3, 0, 9, 0, 2, 0, 6, 0] G' = [1, 20, 1, 10, 1, 12, 1, 19, 2, 5, 1, 11, 1, 3, 1, 7] B' = [4, 20, 8, 10, 8, 12, 18, 19, 5, 5, 10, 11, 3, 3, 7, 7] Congrats! You just calculated a U-Net by hand. 💾 Save this post!

U-Net by hand ✍️ ~ 17 steps walkthrough below I consider U-Net as a key milestone in deep learning, the first image-to-image model that really worked! It came out of medical imaging, an unusual place, not from NeurIPS or CVPR or ACL. Now it is the backbone of diffusion models, which you see in almost all modern image generation models. I drew the network as a C so the matrix multiplication flows naturally down. Tilt your head to the right and it is a U again. 🤣 Goal: push a 3 x 16 image down to a 2 x 4 bottleneck and back out again, filling in every cell yourself. = 1. Given = An image of three channels, R, G and B, sixteen pixels wide, and every kernel the network will use. = 2. Convolution 1 = Let us slide the first kernel over the image. Each output is one multiply-and-add over a 2 x 3 window, and the result is the green feature map. = 3. Find the maxima = We circle the largest value in each 1 x 2 window. Circling first is worth the extra step: it is the pooling decision, made before anything is written down. = 4. Max pool 1 = Let us copy those maxima down. Sixteen columns become eight, and half the detail is gone for good. = 5. Convolution 2 = We convolve again with the second kernel, deeper into the contracting path. The feature map is blue now. = 6. Find the maxima again = Same move as step 3, on the blue map. = 7. Max pool 2 = Eight columns become four. = 8. The bottleneck = Let us convolve once more. This is the bottom of the U, a 2 x 4 block that is everything the network kept. = 9. Spread it out = We start back up. The transposed convolution writes each bottleneck value into a wider grid, leaving gaps between them. = 10. Transposed convolution 1 = Let us fill those gaps by convolving over the spread-out grid. Four columns become eight. = 11. The first skip = We copy the encoder's matching row straight across. This is the skip connection, and it is the whole reason a U-Net can recover detail that pooling threw away. = 12. Convolution with the skip = Let us convolve the upsampled features together with the copied ones. = 13. Spread it out again = Same as step 9, one level up. = 14. Transposed convolution 2 = Eight columns become sixteen, back to the width we started at. = 15. The second skip = The encoder's first feature map comes across, the one made before any pooling happened. = 16. Convolution and ReLU = We convolve, then cross out every negative and set it to zero. = 17. Output convolution = Let us apply the last kernel. Out comes R', G' and B', an image the same size as the one we started with. The outputs: R' = [3, 0, 7, 0, 7, 0, 17, 0, 3, 0, 9, 0, 2, 0, 6, 0] G' = [1, 20, 1, 10, 1, 12, 1, 19, 2, 5, 1, 11, 1, 3, 1, 7] B' = [4, 20, 8, 10, 8, 12, 18, 19, 5, 5, 10, 11, 3, 3, 7, 7] Congrats! You just calculated a U-Net by hand. 💾 Save this post!

Tom Yeh

17,510 просмотров • 1 месяц назад