https://tannhausergate.saisur.info/Notes on NLP, complexity theory, re-engineering, and computational search.https://tannhausergate.saisur.info/light/demosaichttps://tannhausergate.saisur.info/light/demosaicSat, 05 Sep 2026 04:21:28 GMThttps://tannhausergate.saisur.info/light/forwardhttps://tannhausergate.saisur.info/light/forwardSat, 05 Sep 2026 04:21:28 GMThttps://tannhausergate.saisur.info/light/gammahttps://tannhausergate.saisur.info/light/gammaSat, 05 Sep 2026 04:21:28 GMThttps://tannhausergate.saisur.info/light/lumenhttps://tannhausergate.saisur.info/light/lumenSat, 05 Sep 2026 04:21:28 GMThttps://tannhausergate.saisur.info/light/protobuffhttps://tannhausergate.saisur.info/light/protobuffSat, 05 Sep 2026 04:21:28 GMThttps://tannhausergate.saisur.info/notes/Amazon-Adshttps://tannhausergate.saisur.info/notes/Amazon-AdsWed, 02 Sep 2026 00:00:00 GMTWhy this, why now Amazon’s advertising business did $56.2 billion in revenue in 2024 [26] - third behind only Google and Meta, and growing faster than either. Unlike the other two, almost all of it is sold one search at a time: a shopper types “mascara for volume”, and in the fewhttps://tannhausergate.saisur.info/reading/amazon-ads/01-mechanismhttps://tannhausergate.saisur.info/reading/amazon-ads/01-mechanismWed, 02 Sep 2026 00:00:00 GMT← Day index · Block 1 of 10 · ← Previous · Next → Time budget ~90 minutes. Read Edelman-Ostrovsky-Schwarz [1] first and work the pricing formula by hand. Skim Varian [2] for the symmetric-equilibrium bounds and Lahaie-Pennock [3] for squashing. Spend a full twenty minutes on thehttps://tannhausergate.saisur.info/reading/amazon-ads/02-autobidding-worldhttps://tannhausergate.saisur.info/reading/amazon-ads/02-autobidding-worldWed, 02 Sep 2026 00:00:00 GMT← Day index · Block 2 of 10 · ← Previous · Next → Time budget ~90 minutes. Read Aggarwal-Badanidiyuru-Mehta [1] first - it is short and everything else is a reaction to it. Then the value-vs-utility paper [2] and Liaw-Mehta-Perlroth [6] for the price-of-anarchy landscape. Skim thttps://tannhausergate.saisur.info/reading/amazon-ads/03-prediction-stackhttps://tannhausergate.saisur.info/reading/amazon-ads/03-prediction-stackWed, 02 Sep 2026 00:00:00 GMT← Day index · Block 3 of 10 · ← Previous · Next → Time budget ~90 minutes. Read the calibration section and its worked example first - it is the part that costs money. Then read SIM and HSTU for the two directions the ranker is being pulled in, and MTGR for the caution. Skim thehttps://tannhausergate.saisur.info/reading/amazon-ads/04-rl-biddinghttps://tannhausergate.saisur.info/reading/amazon-ads/04-rl-biddingWed, 02 Sep 2026 00:00:00 GMT← Day index · Block 4 of 10 · ← Previous · Next → Time budget ~90 minutes. Read Cai et al. (2017), Wu et al. (2018), and the AIGB paper (2024) in full - together they are the spine of the field. Skim USCB, SORL, and the AuctionNet paper for their formulations and headline numberhttps://tannhausergate.saisur.info/reading/amazon-ads/05-brown-bridgehttps://tannhausergate.saisur.info/reading/amazon-ads/05-brown-bridgeWed, 02 Sep 2026 00:00:00 GMT← Day index · Block 5 of 10 · ← Previous: RL for bidding · Next: Learning agents in auctions → Time budget ~75 minutes. Read the Libratus and ReBeL abstracts and the piKL paper’s Section 3 first (35 min); skim Safe and Nested Subgame Solving and Depth-Limited Solving for the defhttps://tannhausergate.saisur.info/reading/amazon-ads/06-equilibrium-learninghttps://tannhausergate.saisur.info/reading/amazon-ads/06-equilibrium-learningWed, 02 Sep 2026 00:00:00 GMT← Day index · Block 6 of 10 · ← Previous: Noam Brown’s program · Next: Measurement → Time budget ~60 minutes. Read Banchio & Skrzypacz and Kolumbus & Nisan in full (25 min); read the abstracts of NPGA, SODA, and Gaitonde et al. (15 min); skim the collusion section and keep the chttps://tannhausergate.saisur.info/reading/amazon-ads/07-measurementhttps://tannhausergate.saisur.info/reading/amazon-ads/07-measurementWed, 02 Sep 2026 00:00:00 GMT← Day index · Block 7 of 10 · ← Previous · Next → Time budget ~90 minutes. Read Lewis & Rao and Blake & Coey first - together they say the signal is tiny and the naive experiment is biased. Then Li-Zhao-Johari-Weintraub for which way to be wrong instead, and the off-policy sectihttps://tannhausergate.saisur.info/reading/amazon-ads/08-amazon-in-practicehttps://tannhausergate.saisur.info/reading/amazon-ads/08-amazon-in-practiceWed, 02 Sep 2026 00:00:00 GMT← Day index · Block 8 of 10 · ← Previous · Next → Time budget ~75 minutes. Read Amazon’s auction help page and the placement-adjustment page first (10 minutes; they are short and load-bearing). Then read the FTC press release and Amazon’s response side by side. Skim the Amazon Shttps://tannhausergate.saisur.info/notes/Looped-Transformershttps://tannhausergate.saisur.info/notes/Looped-TransformersWed, 02 Sep 2026 00:00:00 GMT The idea in one minute A looped transformer reuses the same block of layers to refine a hidden state. More iterations buy more computation without adding model weights. The useful question is whether that extra computation improves the answer, and how to train it without storinghttps://tannhausergate.saisur.info/reading/amazon-ads/10-open-problemshttps://tannhausergate.saisur.info/reading/amazon-ads/10-open-problemsWed, 02 Sep 2026 00:00:00 GMT← Day index · Block 10 of 10 · ← Previous · Back to the day index → The day ends where the literature does. Everything before this page is settled enough to teach; everything on it is a place where two careful papers disagree, where the theory has a guarantee the practice cannot https://tannhausergate.saisur.info/reading/amazon-ads/09-bibliographyhttps://tannhausergate.saisur.info/reading/amazon-ads/09-bibliographyWed, 02 Sep 2026 00:00:00 GMT← Day index · Block 9 of 10 · ← Previous · Next → This is the card catalogue for the day. Every paper mentioned across the ten blocks appears once, filed under the theme where it does the most work, with a one-line idea and a one-line reason it matters for a Sponsored Products bihttps://tannhausergate.saisur.info/notes/firstcontacthttps://tannhausergate.saisur.info/notes/firstcontactTue, 18 Feb 2025 00:00:00 GMT The first contact of intelligence with the objective intelligible world is its encounter with its own underlying structure. By rendering intelligible this structure, intelligence enables itself to intervene in its own structure and, in doing so, not only to transform itself but https://tannhausergate.saisur.info/notes/weaponizesaehttps://tannhausergate.saisur.info/notes/weaponizesaeFri, 11 Oct 2024 00:00:00 GMT Welcome to the age of turbo maya-induced schizoprenia -sponsored by golden-gate adtech Couple of months ago you might have noticed Golden Gate claude. Researchers effectively found that there appears to be a some features which are highly abstract, multilingual and multimodal. https://tannhausergate.saisur.info/notes/codesearchhttps://tannhausergate.saisur.info/notes/codesearchFri, 21 Jun 2024 00:00:00 GMTProgram Synthesis is the task of automatically finding programs from the underlying programming language that satisfy user intent expressed in some form of constraints. This is very different from the traditional typical compiler which takes in high-level code and converts it to https://tannhausergate.saisur.info/notes/fetesizehttps://tannhausergate.saisur.info/notes/fetesizeTue, 18 Jun 2024 00:00:00 GMT superficial blurring the lines between intution and intellect, among us and machine, between the real and the virtual. This work is inspired by works of Sarvepalli Radhakrishnan, François Chollet, Stanford Encyclopedia of Philosophy, works of Reza Negarestani and my inner conflhttps://tannhausergate.saisur.info/notes/informationarchhttps://tannhausergate.saisur.info/notes/informationarchSun, 09 Jun 2024 00:00:00 GMT Looking for good content in a world where machines can mass produce content In this page, i have some disparate notes on information indexing and how smarter LMs break the traditional search engine paradigm. I have recently been puzzled how would search look in a world where inhttps://tannhausergate.saisur.info/notes/quotes_randomhttps://tannhausergate.saisur.info/notes/quotes_randomSun, 09 Jun 2024 00:00:00 GMTLanguage, as the domain of the intelligible, is presented as an intrinsically computational process of syntactic structuring and semantic ascenthttps://tannhausergate.saisur.info/notes/brottlhttps://tannhausergate.saisur.info/notes/brottlThu, 09 May 2024 00:00:00 GMTParts of the paper This choice is motivated by the fact that the distribution of the frequency of a symbol is strongly influenced by its kind (whether it is a length, a distance, or a literal), its context (i.e., its neighboring symbols), and its position in the text. Briefly (shttps://tannhausergate.saisur.info/notes/gpt2codehttps://tannhausergate.saisur.info/notes/gpt2codeThu, 02 May 2024 00:00:00 GMTLayer Norm Code class LayerNorm(nn.Module): def __init__(self, cfg): super().__init__() self.cfg = cfg self.w = nn.Parameter(t.ones(cfg.d_model)) self.b = nn.Parameter(t.zeros(cfg.d_model)) def forward(self, residual): # residual: https://tannhausergate.saisur.info/notes/conditionnumberhttps://tannhausergate.saisur.info/notes/conditionnumberSat, 20 Apr 2024 00:00:00 GMTGradient converges exponentially fast to a the global minimum with the rate controlled by the condition number of the function. The condition number (k_f) is the ratio of the largest eigenvalue of the Hessian matrix(H) and the smallest eigenvalue of the tangent kernel (K) of thehttps://tannhausergate.saisur.info/notes/optimizershttps://tannhausergate.saisur.info/notes/optimizersSat, 20 Apr 2024 00:00:00 GMTGradient Descent If wee initialize the parameters at some w_0 \in \mathbb{R}^d, the gradient descent algorithm updates the parameters as follows: w_{t+1} = w_t - \eta \nabla L(w_t) where \eta is the learning rate and \nabla L(w_t) is the gradient of the loss function L at w_t. Ithttps://tannhausergate.saisur.info/notes/tangentkernelhttps://tannhausergate.saisur.info/notes/tangentkernelFri, 19 Apr 2024 00:00:00 GMTA tangent kernel for a function F(w)=y, where w \in \mathbb{R}^m and y \in \mathbb{R}^n. A tanget kernel is a matrix K \in \mathbb{R}^{n \times n} is: K(w) = \nabla F(w) \cdot \nabla F(w)^T Essential, it is gradient of the function F at w multiplied with its transpose. One imporhttps://tannhausergate.saisur.info/notes/batchnormhttps://tannhausergate.saisur.info/notes/batchnormSun, 14 Apr 2024 00:00:00 GMTBatch normalization is a technique used to normalize the inputs of each layer in such a way that they have a mean of zero and a standard deviation of one. The algorithm is relatively simple to implement: Values of x over a mini-batch B = \{x_1, x_2, \ldots, x_m\} Outputs: \{y_i=https://tannhausergate.saisur.info/notes/codechunkshttps://tannhausergate.saisur.info/notes/codechunksSun, 14 Apr 2024 00:00:00 GMTUMT code snippet import torch import torch.nn as nn import torch.optim as optim from sklearn.model_selection import train_test_split from torch.utils.data import TensorDataset, DataLoader # Set the random seed for reproducibility torch.manual_seed(42) # Generate sample data frohttps://tannhausergate.saisur.info/notes/mathtermshttps://tannhausergate.saisur.info/notes/mathtermsSun, 14 Apr 2024 00:00:00 GMTIn this page, we will be discussing some math terms that are used in machine learning and deep learning that i have no idea about. Lipschitz continuity Given two metric spaces (X, d_X) and (Y, d_Y), a function f: X \rightarrow Y is said to be Lipschitz continuous if there exists https://tannhausergate.saisur.info/notes/backloghttps://tannhausergate.saisur.info/notes/backlogSun, 14 Apr 2024 00:00:00 GMTReading Backlog Read but not summarized: Intelligence and Spirit by Reza Negarestani. Combines philosophy, mathematics, logic, and computer science Fanged Noumena by Nick Land A collection of essays on philosophy, cybernetics, and accelerationism Neuroplasticity by Moheb Chttps://tannhausergate.saisur.info/notes/ffhttps://tannhausergate.saisur.info/notes/ffFri, 12 Apr 2024 00:00:00 GMTDeep Neural Networks (DNNs) form the backbone of modern machine learning. One of the strength of DNNs is its high expressivity power. This capability to capture any flexible data representation allows deep neural networks to have widespread use from biology or weather prediction.https://tannhausergate.saisur.info/notes/compressionhttps://tannhausergate.saisur.info/notes/compressionThu, 11 Apr 2024 00:00:00 GMT My 🥰 favorite part of information theory 🥰 The main engineering goal of compression is to represent a given any sequence x_i, x_2, \ldots, x_n with fewer bits than the original sequence. The main idea is to remove redundancy in the data. Reducing the number of bits is generalhttps://tannhausergate.saisur.info/notes/curveshttps://tannhausergate.saisur.info/notes/curvesThu, 11 Apr 2024 00:00:00 GMTPrecision and Recall Risk and Complexity Vapnik–Chervonenkis dimension Bias-Variance Tradeoffhttps://tannhausergate.saisur.info/notes/inftheoryhttps://tannhausergate.saisur.info/notes/inftheoryMon, 08 Apr 2024 00:00:00 GMTIn this webpage, we will be talking about the basics behind information theory. Before you start, you need to have basic understanding of probability theory and distributions. This page is crucious to understand the basics of how any neural network can model any distribution. Enthttps://tannhausergate.saisur.info/notes/distributionhttps://tannhausergate.saisur.info/notes/distributionSun, 07 Apr 2024 00:00:00 GMTBernoulli distribution The Bernoulli distribution is a discrete probability distribution for a random variable which takes the value 1 with probability p and the value 0 with probability q = 1 - p. f(k;p) = p^k(1-p)^{1-k} \text{ for } k \in \{0,1\} Which is basiically, at k=1 (suhttps://tannhausergate.saisur.info/notes/lossfunchttps://tannhausergate.saisur.info/notes/lossfuncSun, 07 Apr 2024 00:00:00 GMTCross-Entropy Cross-entropy measures the average number of bits needed to encode data from one distribution (the true distribution) using the wrong distribution (the model distribution). It’s used frequently in machine learning to quantify the difference between the true distribuhttps://tannhausergate.saisur.info/notes/layerexplainhttps://tannhausergate.saisur.info/notes/layerexplainSun, 07 Apr 2024 00:00:00 GMTLayer code Basically represent the equation: Wx+b where W is the weight matrix and b is the bias. class Layer: def __init__(M,N): self.weight = self.random.random((M,N)) self.bias = np.zeros(M) def __forward(x): return np.inner(W,x) + selhttps://tannhausergate.saisur.info/notes/transformerdeephttps://tannhausergate.saisur.info/notes/transformerdeepSun, 07 Apr 2024 00:00:00 GMTIn this page, we will implement Attention is all you need, from scratch. This is a very simple implementation and is not optimized for speed. The goal is to understand the architecture of the transformer. If you are a basic understanding of linear algrebra and python, you should https://tannhausergate.saisur.info/notes/distanemetricshttps://tannhausergate.saisur.info/notes/distanemetricsSat, 06 Apr 2024 00:00:00 GMTHellinger Distance The Hellinger distance is a measure of the similarity between two probability distributions. It is defined as: H^2(P,Q) = \frac{1}{2} \sum_{i=1}^{n} (\sqrt{p_i} - \sqrt{q_i})^2 where P and Q are two probability distributions over the same set of n events. In thhttps://tannhausergate.saisur.info/notes/geometrymeaninghttps://tannhausergate.saisur.info/notes/geometrymeaningSat, 06 Apr 2024 00:00:00 GMT In this book, we will be covering the main points of the book. We will be drawing parallels between the cognitive interpretation and computational representation of language models. My aim here is to better understand representation learning in language models but from a chttps://tannhausergate.saisur.info/notes/jpdfhttps://tannhausergate.saisur.info/notes/jpdfSat, 06 Apr 2024 00:00:00 GMTThe Joint Probability Density Function (JPDF) is a fundamental concept in the study of random matrices, particularly in understanding the statistical properties of their eigenvalues. It describes how likely it is to find the eigenvalues of a random matrix within a particular ranghttps://tannhausergate.saisur.info/notes/linearalghttps://tannhausergate.saisur.info/notes/linearalgSat, 06 Apr 2024 00:00:00 GMTLinear Algebra study In this webpage, i will be writing notes about my linear algebra study. Introduction of Random Matrix Theory Random matrix theory is a branch of mathematical physics that studies the statistical properties of matrices. The theory is motivated by the observatihttps://tannhausergate.saisur.info/notes/quaterionhttps://tannhausergate.saisur.info/notes/quaterionSat, 06 Apr 2024 00:00:00 GMTQuaternion A quaternion is a type of hypercomplex number that extends complex numbers. It is typically represented as: q = a + bi + cj + dk where: a, b, c, and d are real numbers, i, j, and k are the fundamental quaternion units. Quaternions have properties that make them usefuhttps://tannhausergate.saisur.info/notes/wigersurmisehttps://tannhausergate.saisur.info/notes/wigersurmiseSat, 06 Apr 2024 00:00:00 GMTConsider a 2 x 2 GOE matrix H_s = \left( \begin{array}{cc} x_1 & x_3 \\ x_3 & x_2 \\ \end{array}\right) with x_1, x_2 \sim \mathcal{N}(0, 1) and x_3 \sim \mathcal{N}(0, 1/2). What is the pdf \rho(s) of the spacing s = \lambda_2 - \lambda_1 between its two eigenvalues (\lambdahttps://tannhausergate.saisur.info/light/10bithttps://tannhausergate.saisur.info/light/10bitSun, 17 Mar 2024 00:00:00 GMTA 10-bit image refers to the bit depth of an image, which is a measure of the number of bits used to represent the color of a single pixel. In a 10-bit image, each color channel of a pixel is represented with 10 bits of data. This means that for a standard RGB (Red, Green, Blue) https://tannhausergate.saisur.info/light/bayerhttps://tannhausergate.saisur.info/light/bayerSun, 17 Mar 2024 00:00:00 GMTAll of these investigations are from @gennyble, Please go say hi! The BayerJPEG is a strange format used by the Light L16… sometimes. We don’t yet know when it switches from it’s normal packed 10-bit raw format to this, or why. BayerJPEG Header https://tannhausergate.saisur.info/lighl16https://tannhausergate.saisur.info/lighl16Sun, 17 Mar 2024 00:00:00 GMTIn early 2018, Light.co released a groundbreaking camera. It was claimed to be engineering marvel. It takes 16 different images and using computational photogrammetry combines the 16 different images into one. Unfortunately the cameras software was badly executed. In this blog pohttps://tannhausergate.saisur.info/light/lrihttps://tannhausergate.saisur.info/light/lriSun, 17 Mar 2024 00:00:00 GMTAll of these investigations are from @gennyble, Please go say hi! Anatomy of an LRI The file is made up of many blocks, usually 10 or 11 but cases of 40 have occurred. Blocks start with a header and contain some data. There is always a protobuf message within that data, and somethttps://tannhausergate.saisur.info/light/ransachttps://tannhausergate.saisur.info/light/ransacSun, 17 Mar 2024 00:00:00 GMTThe RANSAC (Random Sample Consensus) algorithm is a robust estimation method used to fit a mathematical model to a dataset that contains outliers. It is widely used in computer vision for tasks such as estimating the fundamental matrix in stereo vision, finding homographies betwehttps://tannhausergate.saisur.info/light/sifthttps://tannhausergate.saisur.info/light/siftSun, 17 Mar 2024 00:00:00 GMTThe SIFT (Scale-Invariant Feature Transform) algorithm in OpenCV is a feature detection algorithm used in computer vision tasks to detect and describe local features in images. It’s useful for tasks like object recognition, image stitching, and tracking because it’s invariant to https://tannhausergate.saisur.info/light/tofhttps://tannhausergate.saisur.info/light/tofSun, 17 Mar 2024 00:00:00 GMTTime-of-Flight Calibration, relates to the calibration process for Time-of-Flight (ToF) cameras or sensors. ToF technology measures the time it takes for light to travel from the sensor to the subject and back, allowing the device to accurately map the distance or depth of objecthttps://tannhausergate.saisur.info/light/vignettinghttps://tannhausergate.saisur.info/light/vignettingSun, 17 Mar 2024 00:00:00 GMTVignetting characterization refers to the process of identifying and measuring vignetting, which is a reduction in image brightness or saturation at the periphery compared to the image center. Vignetting is a common optical phenomenon that can occur for various reasons, such as thttps://tannhausergate.saisur.info/https://tannhausergate.saisur.info/Sun, 10 Mar 2024 00:00:00 GMT A fault in the space-time continuum where two normally distant points of space touch one another Welcome to my blog, I will mainly talk about the NLP, Complexity theory, re-engineering products, lack of meaning in a world approaching AGI and computational search problemshttps://tannhausergate.saisur.info/galleryhttps://tannhausergate.saisur.info/gallerySun, 10 Mar 2024 00:00:00 GMTCollection of my favorite pictures Ong-Ard Satrabhandhu is a Thai architect whose work is known for its masterful combination of classical principles and traditional Thai qualities Puppies, by Maruyama Ōkyo, ca. 1790 Elephant Family, by Marc Allante https://tannhausergate.saisur.info/notes/Startups-to-Followhttps://tannhausergate.saisur.info/notes/Startups-to-FollowSat, 27 Jan 2024 00:00:00 GMTBorrowed from Ishan’s blog post Foundation Models OpenAI Anthropic Adept AI Cohere Sakana xAI Robotics Figure 1X Clone Prosper Reason Open Models Nous Research Mistral ggml Skunkworks Alignment Conjecture Apollo Research Finetuning Glaive Hardware Atomic Semi Etched https://tannhausergate.saisur.info/notes/Researchhttps://tannhausergate.saisur.info/notes/ResearchTue, 12 Dec 2023 00:00:00 GMT TitleLinkDynamoLinkMapReduceLinkTAOLinkThe Google File SystemLinkBigtableLinkCAP TheoremLinkKafkaLinkChubbyLinkLSM TreeLinkSpannerLinkConsistent HashingLinkOut of the Tar PitLinkRaft ConsensusLinkScaling Memcachhttps://tannhausergate.saisur.info/notes/Systems-Whitepapershttps://tannhausergate.saisur.info/notes/Systems-WhitepapersTue, 12 Dec 2023 00:00:00 GMT TitleLinkDynamoLinkMapReduceLinkTAOLinkThe Google File SystemLinkBigtableLinkCAP TheoremLinkKafkaLinkChubbyLinkLSM TreeLinkSpannerLinkConsistent HashingLinkOut of the Tar PitLinkRaft ConsensusLinkScaling Memcachhttps://tannhausergate.saisur.info/notes/Podcastshttps://tannhausergate.saisur.info/notes/PodcastsMon, 30 Oct 2023 00:00:00 GMT Stanford MLSys Seminar https://tannhausergate.saisur.info/notes/Productivity-Stackhttps://tannhausergate.saisur.info/notes/Productivity-StackMon, 30 Oct 2023 00:00:00 GMTMy current productivity stack: Roam Research https://tannhausergate.saisur.info/notes/Textbookshttps://tannhausergate.saisur.info/notes/TextbooksMon, 30 Oct 2023 00:00:00 GMT Elements of Statistical Learning https://tannhausergate.saisur.info/notes/Bookshttps://tannhausergate.saisur.info/notes/BooksSat, 28 Oct 2023 00:00:00 GMTSome pages of books Geometry of Meaning by Peter Gardenfors https://tannhausergate.saisur.info/notes/Internet-Gemshttps://tannhausergate.saisur.info/notes/Internet-GemsMon, 23 Oct 2023 00:00:00 GMTHardware https://siboehm.com/ https://leimao.github.io/ https://highscalability.com/ Software https://kipp.ly/ Life lessons https://www.benkuhn.net/ https://tannhausergate.saisur.info/notes/Manifestohttps://tannhausergate.saisur.info/notes/ManifestoFri, 20 Oct 2023 00:00:00 GMTWORK IN PROGRESS My whole phil: Complex systems penalizes massive step functions Most changes around you have increase in variance but mode stays the same Complexity increases with time but not due to the mean but due to the variance Computational beats analytical solutions Nevehttps://tannhausergate.saisur.info/notes/Libraryhttps://tannhausergate.saisur.info/notes/LibraryFri, 06 Oct 2023 00:00:00 GMTHave a nice read, hope you find something you like :) Recs Podcasts Programming Cool Tools Studying Textbooks Courses https://tannhausergate.saisur.info/notes/Topics-to-Explorehttps://tannhausergate.saisur.info/notes/Topics-to-ExploreFri, 06 Oct 2023 00:00:00 GMT Terreance Tao Blog Dominic Cumming substack Grey Mirror ( haven’t read in a year) https://tannhausergate.saisur.info/notes/Transformershttps://tannhausergate.saisur.info/notes/TransformersFri, 06 Oct 2023 00:00:00 GMTTransformers have been all the rage in the NLP community ever since GPT-3 was released and have recently become more well-known to the public after ChatGPT was released. I’m going to keep track of my favorite ways to learn about the Transformer architecture here. Papers Attentiohttps://tannhausergate.saisur.info/notes/Idea-Listhttps://tannhausergate.saisur.info/notes/Idea-ListWed, 04 Oct 2023 00:00:00 GMTOpen ideas, if you would want to collaborate on any of these, please let me know. I have done some progress in them but not detailed enoughs Light L16 stiching algorithm Replace the method of laplacian blending with a more advanced method. Specifically try Alpha blending and cy