Loading video...

Video Failed to Load

Go Home

Self Attention vs Cross Attention by hand ✍️ Resize the matrices yourself 👉 Two attention mechanisms, side by side. Both project X into queries; both compute attention via S = Kᵀ × Q and F = V × A. The only difference is the source of K and V....

ProfTomYeh's profile picture

Tom Yeh

56,031 subscribers

61,300 views • 4 months ago •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Single vs Multi-hand Attention by hand ✍️ Resize matrices yourself 👉 The most important fact about multi-head attention: it has the same parameter count as single-head attention. The difference is purely structural — same total Wqkv weights, partitioned into smaller q–k–v triples. Look at the two diagrams below. Both Wqkv matrices have the same height — same number of weight rows, same number of parameters. What changes is how that single tall block is sliced. • Left. One head. The full Wqkv produces one big QKV: a tall Q (36 rows), a tall K, a tall V. One scoring computation runs over those full-width tensors. • Right. 3 heads. The same-height Wqkv is sliced into 3 smaller q–k–v triples — each 12 rows tall. 3 scoring computations run in parallel, each a thinner version of the left. The compute trade-off — kind of. Same Wqkv weights. Multi-head runs the attention scoring S = Kᵀ × Q once per head, so the dot-product count multiplies by H. • Single-head: seq × seq = 40² = 1600 dot products • Multi-head: seq × seq × H = 40² × 3 = 4800 dot products (3×) But each multi-head dot product is narrower — its inner dimension is head_dim instead of H × head_dim. So when you count actual scalar multiplications, the totals are equal: • Single-head: seq² × (H × head_dim) = 40² × 36 = 57600 • Multi-head: seq² × H × head_dim = 40² × 3 × 12 = 57600 Same FLOPs. Multi-head buys you H independent attention patterns at no extra weight cost and no extra arithmetic cost — it's the same total compute, sliced into H finer-grained heads.
0:14

Single vs Multi-hand Attention by hand ✍️ Resize matrices yourself 👉 The most important fact about multi-head attention: it has the same parameter count as single-head attention. The difference is purely structural — same total Wqkv weights, partitioned into smaller q–k–v triples. Look at the two diagrams below. Both Wqkv matrices have the same height — same number of weight rows, same number of parameters. What changes is how that single tall block is sliced. • Left. One head. The full Wqkv produces one big QKV: a tall Q (36 rows), a tall K, a tall V. One scoring computation runs over those full-width tensors. • Right. 3 heads. The same-height Wqkv is sliced into 3 smaller q–k–v triples — each 12 rows tall. 3 scoring computations run in parallel, each a thinner version of the left. The compute trade-off — kind of. Same Wqkv weights. Multi-head runs the attention scoring S = Kᵀ × Q once per head, so the dot-product count multiplies by H. • Single-head: seq × seq = 40² = 1600 dot products • Multi-head: seq × seq × H = 40² × 3 = 4800 dot products (3×) But each multi-head dot product is narrower — its inner dimension is head_dim instead of H × head_dim. So when you count actual scalar multiplications, the totals are equal: • Single-head: seq² × (H × head_dim) = 40² × 36 = 57600 • Multi-head: seq² × H × head_dim = 40² × 3 × 12 = 57600 Same FLOPs. Multi-head buys you H independent attention patterns at no extra weight cost and no extra arithmetic cost — it's the same total compute, sliced into H finer-grained heads.

Tom Yeh

35,807 views • 4 months ago

Transformer by hand ✍️ ~ 6 steps walkthrough below Open the hood of a transformer and the parts list is overwhelming: embeddings, positional encoding, attention weighting, self-attention, cross-attention, multi-head attention, layer norm, skip connections, softmax, linear, Nx, shifted right, query, key, value, masking. Which of those actually make the car run? Two of them. Attention weighting and the feed-forward network. Everything else is an enhancement to make it run faster and longer, which is how we got from a car to a truck, and to the word "large" in large language model. So I drew and calculated those two parts entirely by hand. Goal: push five features through one transformer block, filling in every cell yourself. 1. Given Five positions of input features, arriving from the previous block. 2. Attention matrix Let us feed all five features to a query-key module (QK) and read back an attention weight matrix, A. The details of that module are a post of their own. 3. Attention weighting We multiply the input features by A to get the attention weighted features, Z. Still five positions. The effect is to combine features *across positions*, horizontally: X1 becomes X1 + X2, X2 becomes X2 + X3, and so on. 4. First layer Let us feed all five weighted features into the first layer of the FFN. Multiply by the weights and biases. This time the combining happens *across feature dimensions*, vertically, and each feature grows from 3 numbers to 4. Note that every position goes through the same weight matrix. That is what "position-wise" means. 5. ReLU We cross out the negatives. They become zeros. 6. Second layer Let us bring it back down: 4 dimensions to 3. The output feeds the next block, which has a completely separate set of parameters, and the whole thing runs again. You have just calculated a transformer block by hand. ✍️ The takeaway: the two parts are doing two different jobs, and neither one alone is enough. Attention mixes *across positions*, so a feature can see its neighbours. The FFN mixes *across feature dimensions*, so each position can think about itself. Horizontal, then vertical. Then that pattern repeats N times, each block with its own separate set of weights. That is the Nx from the list up top, and that is what makes the transformer run. 💾 Save this post! #AIbyHand #Transformers #DeepLearning
0:12

Transformer by hand ✍️ ~ 6 steps walkthrough below Open the hood of a transformer and the parts list is overwhelming: embeddings, positional encoding, attention weighting, self-attention, cross-attention, multi-head attention, layer norm, skip connections, softmax, linear, Nx, shifted right, query, key, value, masking. Which of those actually make the car run? Two of them. Attention weighting and the feed-forward network. Everything else is an enhancement to make it run faster and longer, which is how we got from a car to a truck, and to the word "large" in large language model. So I drew and calculated those two parts entirely by hand. Goal: push five features through one transformer block, filling in every cell yourself. 1. Given Five positions of input features, arriving from the previous block. 2. Attention matrix Let us feed all five features to a query-key module (QK) and read back an attention weight matrix, A. The details of that module are a post of their own. 3. Attention weighting We multiply the input features by A to get the attention weighted features, Z. Still five positions. The effect is to combine features *across positions*, horizontally: X1 becomes X1 + X2, X2 becomes X2 + X3, and so on. 4. First layer Let us feed all five weighted features into the first layer of the FFN. Multiply by the weights and biases. This time the combining happens *across feature dimensions*, vertically, and each feature grows from 3 numbers to 4. Note that every position goes through the same weight matrix. That is what "position-wise" means. 5. ReLU We cross out the negatives. They become zeros. 6. Second layer Let us bring it back down: 4 dimensions to 3. The output feeds the next block, which has a completely separate set of parameters, and the whole thing runs again. You have just calculated a transformer block by hand. ✍️ The takeaway: the two parts are doing two different jobs, and neither one alone is enough. Attention mixes *across positions*, so a feature can see its neighbours. The FFN mixes *across feature dimensions*, so each position can think about itself. Horizontal, then vertical. Then that pattern repeats N times, each block with its own separate set of weights. That is the Nx from the list up top, and that is what makes the transformer run. 💾 Save this post! #AIbyHand #Transformers #DeepLearning

Tom Yeh

26,089 views • 1 month ago

Softmax vs Sigmoid ✍️ Interact 👉 = Softmax = Softmax is how deep networks turn raw scores into a probability distribution — the final layer of every classifier, and the core of every attention head in a transformer. To see what it does, picture five boba tea shops on the same block, all competing for your dollar. Five candidates: a, b, c, d, e — different chains, different brewing styles, different pearls. A boba reviewer hands you a 𝘤𝘩𝘦𝘸𝘪𝘯𝘦𝘴𝘴 𝘴𝘤𝘰𝘳𝘦 for each — higher means perfectly chewy "QQ" pearls with the right bite (ask a Taiwanese friend to find out what QQ means). Negative scores are real: mushy bobas, overcooked pearls, a batch left sitting too long. How do you turn five chewiness scores into an allocation that adds to a whole dollar? You could spend everything at the chewiest shop, but that ignores how good the runners-up are. Softmax is the smooth alternative. Read the diagram left to right. First, raise each score to e^{x} — this does two things: it turns negative chewiness into small positives, and it stretches the gaps between scores exponentially. Then sum all five into a single total Z. Finally, divide each e^{x} by Z to get a probability. The five probabilities add up to one, so you can read them as percentages of your dollar. The chewiest shop gets the biggest slice — but never the whole dollar. That's the point of softmax: it ranks confidently while still leaving room for the others. = Sigmoid = Sigmoid squashes any real number into a probability between 0 and 1 — the classic activation for binary classification, and still the gating function inside LSTMs and GRUs. Same boba block as the previous Softmax example, narrowed to just two contenders — a hot new shop `a` with chewiness score x, and your usual go-to `b` whose score is pinned at zero (the neutral baseline you've come to expect). Sigmoid is just softmax with two players, one of them pinned to zero. Read the diagram left to right. First, raise each score to e^{x} — for the usual shop `b` whose score is zero, this is just e^0 = 1 (the constant baseline). Then sum the two into a total Z. Finally, divide each e^{x} by Z to get a probability. The two probabilities add up to one — the new shop wins more of your dollar when its pearls get chewier, and your usual keeps the rest. That's the point of sigmoid: it turns a single chewiness score into a clean 0-to-1 chance you'll try the new place over your usual. --- AI Math, Algorithms, Architectures by hand ✍️ Subscribe to my 60K+ reader newsletter 👉
0:14

Softmax vs Sigmoid ✍️ Interact 👉 = Softmax = Softmax is how deep networks turn raw scores into a probability distribution — the final layer of every classifier, and the core of every attention head in a transformer. To see what it does, picture five boba tea shops on the same block, all competing for your dollar. Five candidates: a, b, c, d, e — different chains, different brewing styles, different pearls. A boba reviewer hands you a 𝘤𝘩𝘦𝘸𝘪𝘯𝘦𝘴𝘴 𝘴𝘤𝘰𝘳𝘦 for each — higher means perfectly chewy "QQ" pearls with the right bite (ask a Taiwanese friend to find out what QQ means). Negative scores are real: mushy bobas, overcooked pearls, a batch left sitting too long. How do you turn five chewiness scores into an allocation that adds to a whole dollar? You could spend everything at the chewiest shop, but that ignores how good the runners-up are. Softmax is the smooth alternative. Read the diagram left to right. First, raise each score to e^{x} — this does two things: it turns negative chewiness into small positives, and it stretches the gaps between scores exponentially. Then sum all five into a single total Z. Finally, divide each e^{x} by Z to get a probability. The five probabilities add up to one, so you can read them as percentages of your dollar. The chewiest shop gets the biggest slice — but never the whole dollar. That's the point of softmax: it ranks confidently while still leaving room for the others. = Sigmoid = Sigmoid squashes any real number into a probability between 0 and 1 — the classic activation for binary classification, and still the gating function inside LSTMs and GRUs. Same boba block as the previous Softmax example, narrowed to just two contenders — a hot new shop `a` with chewiness score x, and your usual go-to `b` whose score is pinned at zero (the neutral baseline you've come to expect). Sigmoid is just softmax with two players, one of them pinned to zero. Read the diagram left to right. First, raise each score to e^{x} — for the usual shop `b` whose score is zero, this is just e^0 = 1 (the constant baseline). Then sum the two into a total Z. Finally, divide each e^{x} by Z to get a probability. The two probabilities add up to one — the new shop wins more of your dollar when its pearls get chewier, and your usual keeps the rest. That's the point of sigmoid: it turns a single chewiness score into a clean 0-to-1 chance you'll try the new place over your usual. --- AI Math, Algorithms, Architectures by hand ✍️ Subscribe to my 60K+ reader newsletter 👉

Tom Yeh

74,410 views • 3 months ago

Numerous road traffic offences being committed here by the driver of BT70KWR at White Ford Cherry Picker in Tamworth, Staffs. Not least of all driving without due care and attention, and driving other than on a road. There maybe a case to answer for the more serious offence dangerous driving given he drove into the path of oncoming vehicles. Is it time the standards applied to all other road uses are applied to these flag-shagging simpletons?
0:12

Numerous road traffic offences being committed here by the driver of BT70KWR at White Ford Cherry Picker in Tamworth, Staffs. Not least of all driving without due care and attention, and driving other than on a road. There maybe a case to answer for the more serious offence dangerous driving given he drove into the path of oncoming vehicles. Is it time the standards applied to all other road uses are applied to these flag-shagging simpletons?

The Rev. Anton Mittens 🌹👮🎓

17,172 views • 6 months ago

[Discrete Fourier Transform] by Hand ✍️ In signal processing, the Discrete Fourier Transform (DFT) is no doubt the most important method. But the math involved is extremely complex, literally, involving a summation over a complex number term e^(-iwt). I developed this exercise to demonstrate that underneath such complexity, DFT is just a series of matrix multiplications you can calculate by hand. ✍️ Once you see that, it should not surprise you that a deep neural network, which is also a series of matrix multiplications, with activation functions in-between, can learn to perform DFT to process and analyze signals so effectively. How does DFT work? [1] Given ↳ Signals A, B, and C in the 🟧 frequency domain: ◦ A = cos(w) + 2cos(2w) ◦ B = cos(w) + cos(3w) + cos(4w) ◦ C = -cos(2w) + cos(3w) ◦ Each signal is a weighed sum of four cosine waves at frequencies 1w, 2w, 3w, and 4w. ◦ We will apply Inverse DFT to convert the signals to time domain representations, and then demonstrate DFT can convert back to their original frequency domain representations. ↳ Signal X in the 🟩 time domain. X is sampled at 10 time points 1t, 2t, …, 10t: ◦ X = [-2.5, -1.8, 3, -0.7, -1.0, -0.7, 3, -1.8, -2.5, 5] ◦ Suppose X is also a weighted sum of the same four cosine waves, but we don’t already know their weights. We will apply DFT to discover them. [2] 🟧 Frequency Matrix (F) ↳ Write the coefficients of A, B, C as a matrix F. Each signal is a row. Each frequency is a column. ↳ A → [1, 2, 0, 0] ↳ B → [1, 0, 1, 1] ↳ C → [0, 1-, 1, 0] [3] Cosine → Discrete ↳ Sample from the continuous cosine waves at discrete time points 1t, 2t, 3t, to 10t. [4] Cosine Matrix (W) ↳ Write the samples as a matrix, Each frequency is a row. Each time point is a column. [5] Inverse DFT: 🟧 Frequency → 🟩 Time ↳ Multiply the frequency matrix F and the cosine matrix W. ↳ The meaning of this multiplication is to linearly combine the four cosine waves (rows in W) into time-domain signals (rows in T) using the weights specified in F. ↳ The result is matrix T, which are signals A, B, C converted to the time domain. Each signal is a row. Each time point is a column. [6] Transpose ↳ Transpose T, converting each signal’s time domain representation from a row to a column. [7] DFT: 🟩 Time → 🟧 Frequency ↳ Multiply the cosine matrix W with the transpose of matrix T. ↳ The purpose of this multiplication is to take a dot-product between each time-domain signal (columns in the transpose of T) and each cosine wave (rows in W), which has the effect of projecting the signal onto a cosine wave to determine how much they are correlated. Zero means not correlated at all. ↳ The result is an intermediate version of the “recovered” frequency matrix where each column corresponds to a signal and each row corresponds to a frequency. ↳ Compared to the original frequency matrix F, this intermediate matrix has non-zero weights in the correct places, but scaled up by a factor of 5 (n/2, n=10). For example, signal A, originally [1,2,0,0], is recovered at [5,10,0,0]. [8] Scale ↳ Multiply each value by 2/n = 1/5 to scale down the intermediate matrix to match the magnitude of the original frequency matrix F. [9] Transpose ↳ Transpose the recovered frequency matrix back to the same orientation of the original frequency matrix F. ↳ Like magic 🪄, the result is identical to the original F, which means DFT successfully recovered the frequency components of signals A, B, C. [10] Apply DFT to X: 🟩 Time → 🟧 Frequency ↳ Now that we have some confidence in DFT’s ability to recover frequency components, we apply DFT to X’s time-domain representation by multiplying W with X. ↳ The result is the an intermediate matrix. [11] Scale ↳ Similarly, we scale down by a factor of 5 to obtain the recovered frequency components of X (a column). [12] Transpose ↳ Similarly, we transpose the recovered column to row to match the orientation of the frequency matrix. ↳ Using the coefficients [0,0,3,2], we can write the equation of X as 3cos(3w) + 2cos(4w). Notes: I hope this by hand exercise helps you understand the essence of DFT. But there is more technical details, such as: • Sine: The complete DFT math also includes sine waves that follow a similar calculation process. • Phase: Here, we assume all the cosine waves are aligned at the origin, namely, phase is 0. If a phase p is added, for example, cos(w+p), we will need to calculate the sine component and use their ratio to figure out what p is. • Magnitude: If phase is not zero, the magnitude will need to be calculated by combining both cosine and sine terms.
0:12

[Discrete Fourier Transform] by Hand ✍️ In signal processing, the Discrete Fourier Transform (DFT) is no doubt the most important method. But the math involved is extremely complex, literally, involving a summation over a complex number term e^(-iwt). I developed this exercise to demonstrate that underneath such complexity, DFT is just a series of matrix multiplications you can calculate by hand. ✍️ Once you see that, it should not surprise you that a deep neural network, which is also a series of matrix multiplications, with activation functions in-between, can learn to perform DFT to process and analyze signals so effectively. How does DFT work? [1] Given ↳ Signals A, B, and C in the 🟧 frequency domain: ◦ A = cos(w) + 2cos(2w) ◦ B = cos(w) + cos(3w) + cos(4w) ◦ C = -cos(2w) + cos(3w) ◦ Each signal is a weighed sum of four cosine waves at frequencies 1w, 2w, 3w, and 4w. ◦ We will apply Inverse DFT to convert the signals to time domain representations, and then demonstrate DFT can convert back to their original frequency domain representations. ↳ Signal X in the 🟩 time domain. X is sampled at 10 time points 1t, 2t, …, 10t: ◦ X = [-2.5, -1.8, 3, -0.7, -1.0, -0.7, 3, -1.8, -2.5, 5] ◦ Suppose X is also a weighted sum of the same four cosine waves, but we don’t already know their weights. We will apply DFT to discover them. [2] 🟧 Frequency Matrix (F) ↳ Write the coefficients of A, B, C as a matrix F. Each signal is a row. Each frequency is a column. ↳ A → [1, 2, 0, 0] ↳ B → [1, 0, 1, 1] ↳ C → [0, 1-, 1, 0] [3] Cosine → Discrete ↳ Sample from the continuous cosine waves at discrete time points 1t, 2t, 3t, to 10t. [4] Cosine Matrix (W) ↳ Write the samples as a matrix, Each frequency is a row. Each time point is a column. [5] Inverse DFT: 🟧 Frequency → 🟩 Time ↳ Multiply the frequency matrix F and the cosine matrix W. ↳ The meaning of this multiplication is to linearly combine the four cosine waves (rows in W) into time-domain signals (rows in T) using the weights specified in F. ↳ The result is matrix T, which are signals A, B, C converted to the time domain. Each signal is a row. Each time point is a column. [6] Transpose ↳ Transpose T, converting each signal’s time domain representation from a row to a column. [7] DFT: 🟩 Time → 🟧 Frequency ↳ Multiply the cosine matrix W with the transpose of matrix T. ↳ The purpose of this multiplication is to take a dot-product between each time-domain signal (columns in the transpose of T) and each cosine wave (rows in W), which has the effect of projecting the signal onto a cosine wave to determine how much they are correlated. Zero means not correlated at all. ↳ The result is an intermediate version of the “recovered” frequency matrix where each column corresponds to a signal and each row corresponds to a frequency. ↳ Compared to the original frequency matrix F, this intermediate matrix has non-zero weights in the correct places, but scaled up by a factor of 5 (n/2, n=10). For example, signal A, originally [1,2,0,0], is recovered at [5,10,0,0]. [8] Scale ↳ Multiply each value by 2/n = 1/5 to scale down the intermediate matrix to match the magnitude of the original frequency matrix F. [9] Transpose ↳ Transpose the recovered frequency matrix back to the same orientation of the original frequency matrix F. ↳ Like magic 🪄, the result is identical to the original F, which means DFT successfully recovered the frequency components of signals A, B, C. [10] Apply DFT to X: 🟩 Time → 🟧 Frequency ↳ Now that we have some confidence in DFT’s ability to recover frequency components, we apply DFT to X’s time-domain representation by multiplying W with X. ↳ The result is the an intermediate matrix. [11] Scale ↳ Similarly, we scale down by a factor of 5 to obtain the recovered frequency components of X (a column). [12] Transpose ↳ Similarly, we transpose the recovered column to row to match the orientation of the frequency matrix. ↳ Using the coefficients [0,0,3,2], we can write the equation of X as 3cos(3w) + 2cos(4w). Notes: I hope this by hand exercise helps you understand the essence of DFT. But there is more technical details, such as: • Sine: The complete DFT math also includes sine waves that follow a similar calculation process. • Phase: Here, we assume all the cosine waves are aligned at the origin, namely, phase is 0. If a phase p is added, for example, cos(w+p), we will need to calculate the sine component and use their ratio to figure out what p is. • Magnitude: If phase is not zero, the magnitude will need to be calculated by combining both cosine and sine terms.

Tom Yeh

116,622 views • 2 years ago

Nibiru has moved from 4 o’clock position to 3 o’clock position. Moving up towards the elliptic (center of Sun - circle) and seen closer to the Sun. Nibiru is moving infront of the Sun, just as predicted by the Zetas. 10/24/25 vs 11/04/25 ☄️ #Nibiru #Atlas “November is when the Nibiru Complex will move from the left of the Sun to center in front of the Sun.” “• Planet X, finding itself tilting along the flow lines while close to the Sun, does a slow roll to align with the Sun at its middle, the S Pole slung away from the Sun's S Pole, this direction and momentum continuing until it rolls 270° to click into a side-by-side alignment with the Sun. This is implied in Eastfield, a 270° roll both before and after the timeline. • Planet X does more than one 270° roll, as the fastest way to align itself again with magnetic flow lines after piercing the Ecliptic, as implied in Eastfield, the second roll culminating in the passage where the visible presence of Planet X increases enormously during the week of rotation stoppage. The timeline implies a position of Planet X prior to the passage that include angles of 45° on either side and a 180° reversal in between. • Planet X first approaches creating wobbles in the orientation of the Earth as it passes the Sun's S Pole, then a tight lock while rising through the flow lines to the Ecliptic, then slinging the Earth in a radically different orientation (denoted by the beak) during a 270° roll, and then halting the rotation of the Earth as the Atlantic Rift is gripped in line with Planet X.”
0:12

Nibiru has moved from 4 o’clock position to 3 o’clock position. Moving up towards the elliptic (center of Sun - circle) and seen closer to the Sun. Nibiru is moving infront of the Sun, just as predicted by the Zetas. 10/24/25 vs 11/04/25 ☄️ #Nibiru #Atlas “November is when the Nibiru Complex will move from the left of the Sun to center in front of the Sun.” “• Planet X, finding itself tilting along the flow lines while close to the Sun, does a slow roll to align with the Sun at its middle, the S Pole slung away from the Sun's S Pole, this direction and momentum continuing until it rolls 270° to click into a side-by-side alignment with the Sun. This is implied in Eastfield, a 270° roll both before and after the timeline. • Planet X does more than one 270° roll, as the fastest way to align itself again with magnetic flow lines after piercing the Ecliptic, as implied in Eastfield, the second roll culminating in the passage where the visible presence of Planet X increases enormously during the week of rotation stoppage. The timeline implies a position of Planet X prior to the passage that include angles of 45° on either side and a 180° reversal in between. • Planet X first approaches creating wobbles in the orientation of the Earth as it passes the Sun's S Pole, then a tight lock while rising through the flow lines to the Ecliptic, then slinging the Earth in a radically different orientation (denoted by the beak) during a 270° roll, and then halting the rotation of the Earth as the Atlantic Rift is gripped in line with Planet X.”

ZetaTalkLive👽

11,452 views • 9 months ago

Back when we were developing GEN3C, we often imagined a Holodeck-like future: a simulator where multiple agents can enter the same generated world, act independently, and learn to collaborate. Gamma-World makes this feel more concrete. It is a generative multi-agent world model that takes synchronized observations and actions, then rolls out what each agent will see next in the same evolving world — action-responsive at 24 FPS. For me, the key challenge is going beyond two players. As more agents enter, identity cannot be tied to fixed slots, interaction cannot rely on dense pairwise attention, and independent actions still need to resolve into one shared state. Two ideas make this work: 1⃣ Simplex RoPE Distinct agent identities without slot bias — unique, but permutation-equivalent. 2⃣ Sparse Hub Attention Agents communicate through learnable hubs instead of dense all-to-all attention: agent → hub → agent This keeps cross-agent communication scalable. The exciting part: training on two-player data can generalize to four-player rollouts without additional training, and the same formulation extends to real-world bimanual robot coordination. A step toward populated world models: many agents, one shared world. Congrats to the team on Gamma-World! Project:
0:11

Back when we were developing GEN3C, we often imagined a Holodeck-like future: a simulator where multiple agents can enter the same generated world, act independently, and learn to collaborate. Gamma-World makes this feel more concrete. It is a generative multi-agent world model that takes synchronized observations and actions, then rolls out what each agent will see next in the same evolving world — action-responsive at 24 FPS. For me, the key challenge is going beyond two players. As more agents enter, identity cannot be tied to fixed slots, interaction cannot rely on dense pairwise attention, and independent actions still need to resolve into one shared state. Two ideas make this work: 1⃣ Simplex RoPE Distinct agent identities without slot bias — unique, but permutation-equivalent. 2⃣ Sparse Hub Attention Agents communicate through learnable hubs instead of dense all-to-all attention: agent → hub → agent This keeps cross-agent communication scalable. The exciting part: training on two-player data can generalize to four-player rollouts without additional training, and the same formulation extends to real-world bimanual robot coordination. A step toward populated world models: many agents, one shared world. Congrats to the team on Gamma-World! Project:

Xuanchi Ren

304,242 views • 3 months ago

Crypto narratives tend to move in cycles. 2020 was DeFi. 2021 became NFTs. 2023 turned into the AI boom. 2024–2025 were dominated by memecoins and attention tokens. But markets eventually rotate back to something simple: real revenue. That’s why some people are starting to look at iGaming tokens as a potential emerging narrative in 2026. Unlike many hype driven tokens, the iGaming sector already runs large cash flow businesses. Many platforms generate hundreds of millions of dollars in monthly revenue, yet their tokens often trade with far less volume than projects that barely produce revenue at all. In other words, there’s a visible mismatch between actual business activity and token market valuation. One ecosystem that sits right inside this discussion is 1win Token, which already operates as one of the top 10 online casinos globally by scale and user activity. The upcoming $1win Token is designed to connect that existing business with on chain incentives. Its token model includes buybacks and burns funded directly from casino revenue, tying token supply mechanics to real cash flow. There’s also an interesting structural difference compared to previous gaming tokens. For example, $RLB (Rollbit) saw a massive post launch rally, but the product and revenue scale at launch were significantly smaller than what 1win operates today. Another notable point is the launch design: instead of only farming an airdrop, 1win Token plans a public sale model, allowing broader participation from the start. If Web3 narratives are indeed shifting away from pure attention cycles and back toward revenue generating platforms, sectors like iGaming may start attracting more analytical focus and 1win could emerge as the biggest winner.
0:17

Crypto narratives tend to move in cycles. 2020 was DeFi. 2021 became NFTs. 2023 turned into the AI boom. 2024–2025 were dominated by memecoins and attention tokens. But markets eventually rotate back to something simple: real revenue. That’s why some people are starting to look at iGaming tokens as a potential emerging narrative in 2026. Unlike many hype driven tokens, the iGaming sector already runs large cash flow businesses. Many platforms generate hundreds of millions of dollars in monthly revenue, yet their tokens often trade with far less volume than projects that barely produce revenue at all. In other words, there’s a visible mismatch between actual business activity and token market valuation. One ecosystem that sits right inside this discussion is 1win Token, which already operates as one of the top 10 online casinos globally by scale and user activity. The upcoming $1win Token is designed to connect that existing business with on chain incentives. Its token model includes buybacks and burns funded directly from casino revenue, tying token supply mechanics to real cash flow. There’s also an interesting structural difference compared to previous gaming tokens. For example, $RLB (Rollbit) saw a massive post launch rally, but the product and revenue scale at launch were significantly smaller than what 1win operates today. Another notable point is the launch design: instead of only farming an airdrop, 1win Token plans a public sale model, allowing broader participation from the start. If Web3 narratives are indeed shifting away from pure attention cycles and back toward revenue generating platforms, sectors like iGaming may start attracting more analytical focus and 1win could emerge as the biggest winner.

BitBull

20,972 views • 5 months ago

M E S S I E R | Tokenomics Messier is a #utility ecosystem delivering a range of Web3 services for multiple use cases. The success of the project is driven by two core pillars: the usage of these dApps, which fuels revenue growth within the ecosystem, and the performance of the #DAO investments. How does this work? Usage fees flow into the DAO treasury to fund new investments. Stakers earn rewards after each successful investment proposal. After every investment cycle, all profits exceeding 87 $ETH are allocated to buy back and burn the #M87 token, strengthening the deflationary model. This creates a dual positive impact on the token price while positioning holders as true stakeholders in the ecosystem’s success. Messier continues to grow and evolve with unique market features designed to further expand its revenue streams. ⚫️
0:15

M E S S I E R | Tokenomics Messier is a #utility ecosystem delivering a range of Web3 services for multiple use cases. The success of the project is driven by two core pillars: the usage of these dApps, which fuels revenue growth within the ecosystem, and the performance of the #DAO investments. How does this work? Usage fees flow into the DAO treasury to fund new investments. Stakers earn rewards after each successful investment proposal. After every investment cycle, all profits exceeding 87 $ETH are allocated to buy back and burn the #M87 token, strengthening the deflationary model. This creates a dual positive impact on the token price while positioning holders as true stakeholders in the ecosystem’s success. Messier continues to grow and evolve with unique market features designed to further expand its revenue streams. ⚫️

MESSIER | M87

10,222 views • 6 months ago

🔐 Tokenomics Update: G Coin Lock Mechanism G Coin uses a time-based token lock mechanism to manage circulating supply and support long-term ecosystem stability. The mechanism applies to both gameplay activity and unsold tokens. How it works: 🎮 Gameplay token lock G Coin lost during gameplay is removed from active circulation for 12 months ⏱️ After one year, those tokens are released back based on the original loss date. 🧰 Unsold tokens at TGE All tokens not sold before the Token Generation Event (TGE) are locked under a 12-month cliff, followed by 24-month linear vesting 📊 What this means for holders: ▫️ Predictable and controlled supply release 🛡️ ▫️ Reduced post-TGE supply pressure ⚖️ ▫️ Growth driven by real usage, not sudden unlocks 📈 ▫️ Strong long-term alignment across the ecosystem 🤝 This mechanism isn’t about restriction. It’s about responsible distribution, transparency, and long-term confidence. Strong ecosystems start with disciplined tokenomics 🏗️
0:10

🔐 Tokenomics Update: G Coin Lock Mechanism G Coin uses a time-based token lock mechanism to manage circulating supply and support long-term ecosystem stability. The mechanism applies to both gameplay activity and unsold tokens. How it works: 🎮 Gameplay token lock G Coin lost during gameplay is removed from active circulation for 12 months ⏱️ After one year, those tokens are released back based on the original loss date. 🧰 Unsold tokens at TGE All tokens not sold before the Token Generation Event (TGE) are locked under a 12-month cliff, followed by 24-month linear vesting 📊 What this means for holders: ▫️ Predictable and controlled supply release 🛡️ ▫️ Reduced post-TGE supply pressure ⚖️ ▫️ Growth driven by real usage, not sudden unlocks 📈 ▫️ Strong long-term alignment across the ecosystem 🤝 This mechanism isn’t about restriction. It’s about responsible distribution, transparency, and long-term confidence. Strong ecosystems start with disciplined tokenomics 🏗️

Playnance

4,048,283 views • 7 months ago

MLP in PyTorch by hand ✍️ ~ 7 steps walkthrough below Goal: fill in every blank in the PyTorch code to build a multi-layer perceptron. 1. Given Let us start with a code template on the left and the network it is supposed to build on the right. Every blank in the code can be worked out from the picture. 2. Linear layer We count: 3 features in, 4 features out. So the weight matrix is 4 by 3. There is an extra column for the biases, which means bias = T. 3. ReLU Let us apply the activation. ReLU crosses out the negatives, so -1 becomes 0. 4. Linear layer The input size is 4, because that is what the previous layer put out. The output size is 2. A 2 by 4 weight matrix, and this time no extra column, so bias = F. 5. ReLU We cross out the negatives again. 6. Linear layer Two features in, five out. A 5 by 2 weight matrix, with a bias column, so bias = T. 7. Sigmoid Let us finish. Sigmoid squashes the raw scores (3, 0, -2, 5, -5) into probabilities between 0 and 1. You have just implemented a three-layer deep neural network by hand. ✍️ == Story == Three years ago I gave this exercise to my students, to connect the code to the math. They found it odd. Every other AI course they were taking lived inside a Jupyter notebook, and here I was handing out paper. Three years later, my colleagues are the ones rushing to move their materials to paper. The exercise has not changed. Paper still asks the one thing a notebook lets you skip: do you actually understand what the code is doing? If you can tell me why the weight matrix is 4 by 3, and why bias is F on the second layer, you understand nn.Linear better than someone who has been copy-pasting it for a year. 💾 Save this post! #AIbyHand #PyTorch #DeepLearning
0:14

MLP in PyTorch by hand ✍️ ~ 7 steps walkthrough below Goal: fill in every blank in the PyTorch code to build a multi-layer perceptron. 1. Given Let us start with a code template on the left and the network it is supposed to build on the right. Every blank in the code can be worked out from the picture. 2. Linear layer We count: 3 features in, 4 features out. So the weight matrix is 4 by 3. There is an extra column for the biases, which means bias = T. 3. ReLU Let us apply the activation. ReLU crosses out the negatives, so -1 becomes 0. 4. Linear layer The input size is 4, because that is what the previous layer put out. The output size is 2. A 2 by 4 weight matrix, and this time no extra column, so bias = F. 5. ReLU We cross out the negatives again. 6. Linear layer Two features in, five out. A 5 by 2 weight matrix, with a bias column, so bias = T. 7. Sigmoid Let us finish. Sigmoid squashes the raw scores (3, 0, -2, 5, -5) into probabilities between 0 and 1. You have just implemented a three-layer deep neural network by hand. ✍️ == Story == Three years ago I gave this exercise to my students, to connect the code to the math. They found it odd. Every other AI course they were taking lived inside a Jupyter notebook, and here I was handing out paper. Three years later, my colleagues are the ones rushing to move their materials to paper. The exercise has not changed. Paper still asks the one thing a notebook lets you skip: do you actually understand what the code is doing? If you can tell me why the weight matrix is 4 by 3, and why bias is F on the second layer, you understand nn.Linear better than someone who has been copy-pasting it for a year. 💾 Save this post! #AIbyHand #PyTorch #DeepLearning

Tom Yeh

13,318 views • 1 month ago

🔋 One more push ahead, and even the safe and cozy Yaniv can already be seen. Some starving howls are following along, closer and louder every second. The backpack is full of precious habar from a foray, though no breath to keep it further. Whoever was there knows… In moments like this, you should start to say goodbyes to the habar or your life, depending on one’s values. For that kind of mishap, experienced stalkers got an ace in the can, the same as 16 years ago. Reliable. Tasty. And refreshing, how it should be. Be NON STOP by nature, but keep a few in your stash. Non Stop🤝S.T.A.L.K.E.R. OFFICIAL — just like on the first journey, officially together again! #Stalker2 x #NonStopEnergy x #NonStoppable
0:12

🔋 One more push ahead, and even the safe and cozy Yaniv can already be seen. Some starving howls are following along, closer and louder every second. The backpack is full of precious habar from a foray, though no breath to keep it further. Whoever was there knows… In moments like this, you should start to say goodbyes to the habar or your life, depending on one’s values. For that kind of mishap, experienced stalkers got an ace in the can, the same as 16 years ago. Reliable. Tasty. And refreshing, how it should be. Be NON STOP by nature, but keep a few in your stash. Non Stop🤝S.T.A.L.K.E.R. OFFICIAL — just like on the first journey, officially together again! #Stalker2 x #NonStopEnergy x #NonStoppable

S.T.A.L.K.E.R. OFFICIAL

988,645 views • 3 years ago

AI token usage is up 10x in 7 months, compounding 40%/MONTH! There is NO BUBBLE when demand is STILL accelerating And this is just OpenRouter, it doesn't count the labs direct token usage and APIs But here's what's interesting about these numbers, the demand is coming from everywhere at once US models (OpenAI, Anthropic, Google) keep growing, while Chinese open weight models (DeepSeek, Tencent, Xiaomi, Minimax) grew even faster and now drive over 60% of usage on OpenRouter Closed source and open source both compounding at the same time. This is literally the best case scenario for AI Infra investors It means both frontier model tokens and cheaper tokens have product market fit. This means the application layer is finding ways to use both and generate ROI with both types Demand for tokens IS demand for compute. This is why SpaceX is looking to build 10GW of compute by next year, because the demand is clearly here Now combine this demand set up, with NVIDIA yesterday announcing financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize over $500 billion of third party capital for AI infrastructure And Jensen has said publicly he expects $3 to $4 TRILLION of AI infrastructure spend by 2030 The build out will have to continue for a lot longer than the market is expecting, that is very clear to me. Don't let this consolidation period in AI infra stocks shake you out, they will have their moment again and take their next leg higher p.s. if you want to see how im investing in this, you can track my real-time portfolio and the research of all 5 Milk Road PRO analysts with live trade notifications, and it's just $1 to try it out (insane price just to check it out). Learn more here: Good luck out there!
0:19

AI token usage is up 10x in 7 months, compounding 40%/MONTH! There is NO BUBBLE when demand is STILL accelerating And this is just OpenRouter, it doesn't count the labs direct token usage and APIs But here's what's interesting about these numbers, the demand is coming from everywhere at once US models (OpenAI, Anthropic, Google) keep growing, while Chinese open weight models (DeepSeek, Tencent, Xiaomi, Minimax) grew even faster and now drive over 60% of usage on OpenRouter Closed source and open source both compounding at the same time. This is literally the best case scenario for AI Infra investors It means both frontier model tokens and cheaper tokens have product market fit. This means the application layer is finding ways to use both and generate ROI with both types Demand for tokens IS demand for compute. This is why SpaceX is looking to build 10GW of compute by next year, because the demand is clearly here Now combine this demand set up, with NVIDIA yesterday announcing financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize over $500 billion of third party capital for AI infrastructure And Jensen has said publicly he expects $3 to $4 TRILLION of AI infrastructure spend by 2030 The build out will have to continue for a lot longer than the market is expecting, that is very clear to me. Don't let this consolidation period in AI infra stocks shake you out, they will have their moment again and take their next leg higher p.s. if you want to see how im investing in this, you can track my real-time portfolio and the research of all 5 Milk Road PRO analysts with live trade notifications, and it's just $1 to try it out (insane price just to check it out). Learn more here: Good luck out there!

Kyle Reidhead | Milk Road

28,320 views • 18 days ago

Can you beat the top score? Get the top score on Swipe the cat and win an Apple Gift Card (link on my profile) 🥇 Get top score: $300 Apple Gift Card Best video submission: $200 Apple Gift Card Funniest submission: $100 Apple Gift Card Rules: 1) Post your score to X with an attached screenshot (required) or video (optional) of your best score (NOT the leaderboard). See example. To be eligible you must either tag me in your post or reply to this post. Reposts are optional for visibility and do not affect eligibility 2) Entries must be posted to X before December 10, 2025 @ 11:59pm AEDT (Australian Eastern Daylight time). 3) Your score must be listed on the ranked leaderboard. You may be asked to validate your score by providing your Game Center username. Be sure your profile is public, that you are playing on the latest version of the game and are logged into Game Center. 4) The highest score posted wins the top score prize. In the event that multiple winners get the same top score the winnings will be divided amongst them. So 2 people getting the same top score means prize will be $150 each. 5) Best video submission will be judged by me and my decision will be final. Criteria: I will be looking for creativity, entertainment value, style and originality. The entry that stands out the most from other video submissions. 6) Best funniest submission will be judged by me and my decision will be final. You may include an optional video for extra visibility. Criteria will be: what makes me laugh the most. Only age appropriate entries will be considered (no crude humour). 7) Gift cards are in USD. Gift card value may be subject to currency conversion for my local currency (Australia) and your local currency. Depending on the arrangement we find to send you the Gift card. Apple Gift Cards must be redeemable in your region. If not we will arrange an alternative of equivalent value. 8) Winners will be notified by direct message on X and publicly on my X profile on or before December 15. Winners must respond within 7 days to claim their prize. 9) You may enter the competition as many times as you like. Only your highest valid score will be considered for the Top Score prize. 10) You must be 13+ or meet your region’s minimum age for using X and participating in online competitions. 11) I reserve the right to disqualify suspicious or unverifiable scores at my discretion (e.g., if I suspect hacked/modded devices/assisting devices). Important disclaimers: This competition is not affiliated with Apple or X By entering into this competition you give me permission to use the contents of your post (name, text, image, video) for use in promotional material for the game. This content will be used on my YouTube channel and to promote my game not only on X but on other platforms too. If you do not wish to give permission to use your post text, images and/or video please opt-out on your submitted post. "I do not give permission for my post to be used in promotional material". Opting out of promotional use does not disqualify you from winning.
0:13

Can you beat the top score? Get the top score on Swipe the cat and win an Apple Gift Card (link on my profile) 🥇 Get top score: $300 Apple Gift Card Best video submission: $200 Apple Gift Card Funniest submission: $100 Apple Gift Card Rules: 1) Post your score to X with an attached screenshot (required) or video (optional) of your best score (NOT the leaderboard). See example. To be eligible you must either tag me in your post or reply to this post. Reposts are optional for visibility and do not affect eligibility 2) Entries must be posted to X before December 10, 2025 @ 11:59pm AEDT (Australian Eastern Daylight time). 3) Your score must be listed on the ranked leaderboard. You may be asked to validate your score by providing your Game Center username. Be sure your profile is public, that you are playing on the latest version of the game and are logged into Game Center. 4) The highest score posted wins the top score prize. In the event that multiple winners get the same top score the winnings will be divided amongst them. So 2 people getting the same top score means prize will be $150 each. 5) Best video submission will be judged by me and my decision will be final. Criteria: I will be looking for creativity, entertainment value, style and originality. The entry that stands out the most from other video submissions. 6) Best funniest submission will be judged by me and my decision will be final. You may include an optional video for extra visibility. Criteria will be: what makes me laugh the most. Only age appropriate entries will be considered (no crude humour). 7) Gift cards are in USD. Gift card value may be subject to currency conversion for my local currency (Australia) and your local currency. Depending on the arrangement we find to send you the Gift card. Apple Gift Cards must be redeemable in your region. If not we will arrange an alternative of equivalent value. 8) Winners will be notified by direct message on X and publicly on my X profile on or before December 15. Winners must respond within 7 days to claim their prize. 9) You may enter the competition as many times as you like. Only your highest valid score will be considered for the Top Score prize. 10) You must be 13+ or meet your region’s minimum age for using X and participating in online competitions. 11) I reserve the right to disqualify suspicious or unverifiable scores at my discretion (e.g., if I suspect hacked/modded devices/assisting devices). Important disclaimers: This competition is not affiliated with Apple or X By entering into this competition you give me permission to use the contents of your post (name, text, image, video) for use in promotional material for the game. This content will be used on my YouTube channel and to promote my game not only on X but on other platforms too. If you do not wish to give permission to use your post text, images and/or video please opt-out on your submitted post. "I do not give permission for my post to be used in promotional material". Opting out of promotional use does not disqualify you from winning.

Adam Lyttle

159,986 views • 8 months ago

Full Fine-tuning vs. Freezing Layers. Interact 👉 and == Full Fine-tuning == A real network has many — three layers in this example, billions of parameters in a production model. What does fine-tuning look like when you update all of them? That’s full fine-tuning: continue training every weight in the pretrained network on your new task. Every layer’s W gets its own ΔW. Nothing is frozen — every parameter is in play. Think of an MLP as a chain of prerequisites leading to an advanced course. Layer 1 might be Linear Algebra, layer 2 Probability, layer 3 Advanced Machine Learning — each one building on what came before. Fine-tuning is what happens during graduate study: the foundations are already there from undergrad, so you’re not re-learning. Full fine-tuning is reviewing every prerequisite to see what new topics have appeared and what discoveries the field has made since the last time you sat through them. Effective — but exhausting. This diagram shows the same three-layer MLP twice, side by side. On the left, the pretrained network runs on input X: three weight matrices W₁, W₂, W₃, each followed by a ReLU activation. Full fine-tuning gives the model the most freedom to specialize. Every parameter can move — and every parameter that can move must be stored. But not every prerequisite needs revisiting. The further you go back in the chain, the less the material has changed since pretraining — the linear-algebra basics under your computer-vision course are largely the same as they ever were. The next page does exactly that: freeze the prerequisites that haven’t moved, and only refresh the advanced one closest to your specialization. == Freezing Layers == Full fine-tuning reviewed every prerequisite — Linear Algebra, Probability, Advanced ML — to refresh each subject with the latest topics. Effective, but exhausting. Then you realize something. The prerequisites haven’t actually changed that much. Linear Algebra is still Linear Algebra; the matrix decompositions you learned still hold. Probability is still Probability; the distributions and Bayes’ rule haven’t moved. Almost all the new material — the new ideas, the recent discoveries — lives in the advanced layer at the top. That’s freezing layers: keep the prerequisite layers fixed at their pretrained state, and only update the advanced one. In the diagram below, W1​ and W2​ — the foundational prerequisites — stay frozen. Only W3​ — the layer closest to your task-specific output — gets a ΔW.
0:16

Full Fine-tuning vs. Freezing Layers. Interact 👉 and == Full Fine-tuning == A real network has many — three layers in this example, billions of parameters in a production model. What does fine-tuning look like when you update all of them? That’s full fine-tuning: continue training every weight in the pretrained network on your new task. Every layer’s W gets its own ΔW. Nothing is frozen — every parameter is in play. Think of an MLP as a chain of prerequisites leading to an advanced course. Layer 1 might be Linear Algebra, layer 2 Probability, layer 3 Advanced Machine Learning — each one building on what came before. Fine-tuning is what happens during graduate study: the foundations are already there from undergrad, so you’re not re-learning. Full fine-tuning is reviewing every prerequisite to see what new topics have appeared and what discoveries the field has made since the last time you sat through them. Effective — but exhausting. This diagram shows the same three-layer MLP twice, side by side. On the left, the pretrained network runs on input X: three weight matrices W₁, W₂, W₃, each followed by a ReLU activation. Full fine-tuning gives the model the most freedom to specialize. Every parameter can move — and every parameter that can move must be stored. But not every prerequisite needs revisiting. The further you go back in the chain, the less the material has changed since pretraining — the linear-algebra basics under your computer-vision course are largely the same as they ever were. The next page does exactly that: freeze the prerequisites that haven’t moved, and only refresh the advanced one closest to your specialization. == Freezing Layers == Full fine-tuning reviewed every prerequisite — Linear Algebra, Probability, Advanced ML — to refresh each subject with the latest topics. Effective, but exhausting. Then you realize something. The prerequisites haven’t actually changed that much. Linear Algebra is still Linear Algebra; the matrix decompositions you learned still hold. Probability is still Probability; the distributions and Bayes’ rule haven’t moved. Almost all the new material — the new ideas, the recent discoveries — lives in the advanced layer at the top. That’s freezing layers: keep the prerequisite layers fixed at their pretrained state, and only update the advanced one. In the diagram below, W1​ and W2​ — the foundational prerequisites — stay frozen. Only W3​ — the layer closest to your task-specific output — gets a ΔW.

Tom Yeh

27,587 views • 4 months ago

A 27-year-old in Chengdu has been pretending to go to work for 11 months. His mother irons the shirt every morning. Last Tuesday his Polymarket wallet crossed $69,800. He goes by kingofcoinflips. Huawei cut him in June along with the rest of the cloud team. Two months of severance, a stack of recommendation letters nobody opens, and a non-disclosure he keeps in the same drawer as his diploma. He never said a word at home. Every morning he pulls on the same shirt, takes the metro three stops past the Huawei tower, and sits in a co-working desk above a dumpling shop in Chunxi Lu. Six other guys in the same row are running the same routine. Nobody asks. What he brought to the table was 3,400 logged setups since August and a highlighted photocopy of a 1948 Bell Labs paper. Claude Shannon. The MIT professor who quietly compounded 28% a year for three decades and edged out Buffett in the process. He didn't pick stocks. He measured how much information his bets contained, in bits. This kid does the same thing on daily Bitcoin price markets. Every contract gets one number before he touches it: D_KL(P‖Q) = Σ p(x) · log2[p(x)/q(x)] Under 0.05 bits and the fees swallow you. Past 0.10 it's real signal. Past 0.30 your model is broken. Last Tuesday's hit: a BTC Above $80,000 contract quoted at 26.8¢ at sunrise. Order flow on Binance over the prior two hours gave his calculator 0.31 bits. True probability sat closer to 71%. The bot loaded half Kelly. Eight hours later the contract settled at a dollar. +$2,264 on a single click. He stacks a second number on top of every market. How much the Binance tape actually tells him about where Polymarket is heading: I(X;Y) = H(X) - H(X|Y) Under 0.10 bits and the noise wins. Past 0.18 the channel is open. A third loop measures the sharpness of every estimate before sizing: J(θ) = E[(∂/∂θ log f(x|θ))²] When Fisher climbs while mutual information falls, the market is sharpening on noise. He calls it the trap zone. Cost him $9,000 in the first six weeks before he started logging it. Hasn't lost there since. The whole thing exists because Polymarket's daily crypto markets trail the spot tape by roughly forty minutes on slow Asian sessions. Forty minutes is forever for a script and impossible for a person. Out of 3,400 entries, the calculator killed 89% before they ever reached the order book. The 11% that survived built the $69,800 position stack and the $27,500 cumulative since August. His mother sent him a picture of a suit yesterday. Said it's for the family dinner next month. He told her he'd wear it. Plans to tell her the truth at $100,000. His wallet: 99.9% scroll past and call it luck. 0.01% count bits.
0:19

A 27-year-old in Chengdu has been pretending to go to work for 11 months. His mother irons the shirt every morning. Last Tuesday his Polymarket wallet crossed $69,800. He goes by kingofcoinflips. Huawei cut him in June along with the rest of the cloud team. Two months of severance, a stack of recommendation letters nobody opens, and a non-disclosure he keeps in the same drawer as his diploma. He never said a word at home. Every morning he pulls on the same shirt, takes the metro three stops past the Huawei tower, and sits in a co-working desk above a dumpling shop in Chunxi Lu. Six other guys in the same row are running the same routine. Nobody asks. What he brought to the table was 3,400 logged setups since August and a highlighted photocopy of a 1948 Bell Labs paper. Claude Shannon. The MIT professor who quietly compounded 28% a year for three decades and edged out Buffett in the process. He didn't pick stocks. He measured how much information his bets contained, in bits. This kid does the same thing on daily Bitcoin price markets. Every contract gets one number before he touches it: D_KL(P‖Q) = Σ p(x) · log2[p(x)/q(x)] Under 0.05 bits and the fees swallow you. Past 0.10 it's real signal. Past 0.30 your model is broken. Last Tuesday's hit: a BTC Above $80,000 contract quoted at 26.8¢ at sunrise. Order flow on Binance over the prior two hours gave his calculator 0.31 bits. True probability sat closer to 71%. The bot loaded half Kelly. Eight hours later the contract settled at a dollar. +$2,264 on a single click. He stacks a second number on top of every market. How much the Binance tape actually tells him about where Polymarket is heading: I(X;Y) = H(X) - H(X|Y) Under 0.10 bits and the noise wins. Past 0.18 the channel is open. A third loop measures the sharpness of every estimate before sizing: J(θ) = E[(∂/∂θ log f(x|θ))²] When Fisher climbs while mutual information falls, the market is sharpening on noise. He calls it the trap zone. Cost him $9,000 in the first six weeks before he started logging it. Hasn't lost there since. The whole thing exists because Polymarket's daily crypto markets trail the spot tape by roughly forty minutes on slow Asian sessions. Forty minutes is forever for a script and impossible for a person. Out of 3,400 entries, the calculator killed 89% before they ever reached the order book. The 11% that survived built the $69,800 position stack and the $27,500 cumulative since August. His mother sent him a picture of a suit yesterday. Said it's for the family dinner next month. He told her he'd wear it. Plans to tell her the truth at $100,000. His wallet: 99.9% scroll past and call it luck. 0.01% count bits.

Lunar

22,351 views • 3 months ago

[SBS News] BTS V's self-composed song 'Into the Sun'... appears in G7 Summit welcome video The song "Into the Sun," co-written and composed by BTS V, is garnering attention after being used in a video related to the G7 Summit. On the 16th (local time), French President Emmanuel Macron released a video on his official social media welcoming the leaders attending the G7 Summit held in Evian, France. The video featured the leaders of each country appearing one by one, with music symbolizing their respective national images and cultures used as background music. In particular, BTS's "Into the Sun" played during the scene featuring President Lee Jae-myung, drawing significant attention. The National, a prominent English-language media outlet in the Middle East, reported that President Macron selected music tailored to the nationality, culture, and image of each world leader. According to the report, BTS's "Into the Sun" was used for South Korean President Lee Jae-myung's video; Tom Petty's "Love Is a Long Road" for U.S. President Donald Trump; the James Bond theme song "The World Is Not Enough" for British Prime Minister Keir Starmer; and Celine Dion's "Jueire Ou Tou Ira" for Canadian Prime Minister Mark Carney. "Into the Sun" is a track from BTS's studio album *Arirang*, with V participating in the main songwriting and composition. It is known that V completed the song based on a melody that came to him on his way back from a workout. He previously stated regarding the song, "Although it is a song I created, I thought I needed to make a cold, objective judgment on whether it fit the album," but it was ultimately included due to the active recommendations of the other members. Following the release of 'Arirang,' 'Into the Sun' received favorable reviews from major international media outlets. The U.S. daily newspaper The New York Times described it as "a hypnotic song that puts the mind at ease," while the music magazine Rolling Stone noted that "the falsetto harmonies and sparkling tempo demonstrate BTS's infinite potential." The BBC in the UK also introduced the track as "an experimental yet intriguing song that adds a mysterious atmosphere through digital effects." (Article: #BTSV #INTOTHESUN #ARIRANG
0:10

[SBS News] BTS V's self-composed song 'Into the Sun'... appears in G7 Summit welcome video The song "Into the Sun," co-written and composed by BTS V, is garnering attention after being used in a video related to the G7 Summit. On the 16th (local time), French President Emmanuel Macron released a video on his official social media welcoming the leaders attending the G7 Summit held in Evian, France. The video featured the leaders of each country appearing one by one, with music symbolizing their respective national images and cultures used as background music. In particular, BTS's "Into the Sun" played during the scene featuring President Lee Jae-myung, drawing significant attention. The National, a prominent English-language media outlet in the Middle East, reported that President Macron selected music tailored to the nationality, culture, and image of each world leader. According to the report, BTS's "Into the Sun" was used for South Korean President Lee Jae-myung's video; Tom Petty's "Love Is a Long Road" for U.S. President Donald Trump; the James Bond theme song "The World Is Not Enough" for British Prime Minister Keir Starmer; and Celine Dion's "Jueire Ou Tou Ira" for Canadian Prime Minister Mark Carney. "Into the Sun" is a track from BTS's studio album *Arirang*, with V participating in the main songwriting and composition. It is known that V completed the song based on a melody that came to him on his way back from a workout. He previously stated regarding the song, "Although it is a song I created, I thought I needed to make a cold, objective judgment on whether it fit the album," but it was ultimately included due to the active recommendations of the other members. Following the release of 'Arirang,' 'Into the Sun' received favorable reviews from major international media outlets. The U.S. daily newspaper The New York Times described it as "a hypnotic song that puts the mind at ease," while the music magazine Rolling Stone noted that "the falsetto harmonies and sparkling tempo demonstrate BTS's infinite potential." The BBC in the UK also introduced the track as "an experimental yet intriguing song that adds a mysterious atmosphere through digital effects." (Article: #BTSV #INTOTHESUN #ARIRANG

Taehyung Naver

25,147 views • 2 months ago

⚡ SONIC BONDS ARE NOW LIVE! ⚡⛓️ We’re thrilled to bring Bonds to Sonic — the fastest EVM chain powered by $S, combining speed, incentives, and world-class infrastructure. 🔥 In addition to the $S Bonds from Sonic Labs, we’re also launching two new partner Bonds from within the ecosystem: Shadow Exchange x(3,3) 💥! and むん兵衛@(マッド)ボンバーマン! 1️⃣ Sonic is the highest-performing EVM L1, combining speed, incentives, and world-class infrastructure, powering the next generation of DeFi applications. The chain provides 400,000 TPS and sub-second finality. The S token is Sonic's native token, used for paying transaction fees, staking, running validators, and participating in governance. Grab discounted $S now! → 2️⃣ Shadow is a Sonic-native concentrated liquidity exchange that offers deep liquidity, minimal slippage, and precise trading. Users can maximize returns by targeting active liquidity ranges and fine-tuning price bands. The platform rewards users with fees, vote incentives, and rebases, while its dynamic, customizable fee system adapts to market activity, powered by SHADOW and x33 tokens. Get $x33 tokens at a discount! → 3️⃣ MoonBay is a crypto project on the Sonic Network with a strong community and the $MOON token at its core. Blending meme culture with real utility, it embraces DeFi, NFTs, GameFi, and more. Focused on trends and innovation, MoonBay offers value, entertainment, and growth, making it a vibrant hub in the crypto space. Buy $MOON at a discount! → Get ready to Bond faster, better, and smarter — the Sonic way. 💨
0:12

⚡ SONIC BONDS ARE NOW LIVE! ⚡⛓️ We’re thrilled to bring Bonds to Sonic — the fastest EVM chain powered by $S, combining speed, incentives, and world-class infrastructure. 🔥 In addition to the $S Bonds from Sonic Labs, we’re also launching two new partner Bonds from within the ecosystem: Shadow Exchange x(3,3) 💥! and むん兵衛@(マッド)ボンバーマン! 1️⃣ Sonic is the highest-performing EVM L1, combining speed, incentives, and world-class infrastructure, powering the next generation of DeFi applications. The chain provides 400,000 TPS and sub-second finality. The S token is Sonic's native token, used for paying transaction fees, staking, running validators, and participating in governance. Grab discounted $S now! → 2️⃣ Shadow is a Sonic-native concentrated liquidity exchange that offers deep liquidity, minimal slippage, and precise trading. Users can maximize returns by targeting active liquidity ranges and fine-tuning price bands. The platform rewards users with fees, vote incentives, and rebases, while its dynamic, customizable fee system adapts to market activity, powered by SHADOW and x33 tokens. Get $x33 tokens at a discount! → 3️⃣ MoonBay is a crypto project on the Sonic Network with a strong community and the $MOON token at its core. Blending meme culture with real utility, it embraces DeFi, NFTs, GameFi, and more. Focused on trends and innovation, MoonBay offers value, entertainment, and growth, making it a vibrant hub in the crypto space. Buy $MOON at a discount! → Get ready to Bond faster, better, and smarter — the Sonic way. 💨

ApeBond

22,038 views • 1 year ago