Latch

latch

If you're planning to join a remote team in a tech role, wouldn't it be fun to imagine how your future life would look like? You would wake up in your bed, and walk to your coffee machine, and walk to your laptop, and then... Would you have a daily standup and what time? Would you write words to a Slack channel, or should you physically stand up on a Zoom call? (meaning, should you have your clothes on at this point?) I'm writing a and asked remote companies about their daily standup habits. daily newsletter about remote work Most teams are posting what they're working on to the daily #standup channel thread in Slack. Some do video calls. Here are some comments from remote teams about their daily standup habits: 🦊 GitLab "Don't talk about what you did yesterday, this is not a reporting moment where everyone tries to look busy. Rather, kickstart the day with some bonding, solve anything blocking and share future plans so people can plan and act and ultimately save time." 🚀 Stanwood "We have 2 different kind of daily stand ups: 1. "Project office hours" twice a day at 10am and 4pm: PM opens an appear.in room and everyone working on that project can quickly and easily join and discuss questions or roadblocks. The PM summarises the results and posts it to the project channel in slack. 2. Daily Check ins in our slack standup channel: Everyone posts once a day: How many tickets are assigned to me? How many tickets did I close in this sprint so far? Can I take more tickets? Are there any roadblocks ahead of me?" 🏠 Latchel "We do Weekly Kickoff meetings and daily standups. In each daily standup we'll also schedule out regular check-ins throughout the day with other team members that we're working with. For example, you may schedule a check in with product in the middle of the day to make sure you understand all the requirements for a feature you're working on." 👂 Canny "Post what you're working on to the daily #standup channel thread in Slack." Starting a day with a standup is a good way to share what's happening in a team and encourage conversations about tasks in progress. It's even more so in a distributed team, although this can be sometimes tricky with members spreading through a lot of different time zones.

How GitLab, Stanwood, Latchel and Canny Execute their Daily Standups with Remote Teams

Next week, the US will hold a presidential election. At that time, we will vote not for a presidential candidate but for an impression of the candidate we see on TV and at campaign events, read about in newspapers and blogs, and hear about from our friends and family. Like Plato’s shadows on the cave wall, our impressions will reflect only whatever somebody chooses to show us. Modern politics overwhelms us with images crafted and cultivated to fit what we want to see rather than the truth. We latch onto the fragment of reality that we perceive. We never see the full picture. Outside of a few insiders, nobody has enough information to make an informed decision about anybody they vote for. I certainly don’t. Who’s good? Who’s bad? You may think the answers are obvious. Often, it’s not so easy to tell. The devil you don’t know or the other devil you don’t know For example, Congressman Tom Emmer, one of crypto’s most powerful advocates in Congress. He’s the guy who yelled at SEC Chairman Gary Gensler for missing the FTX fraud, among other grievances. Did you know FTX was one of the biggest donors to Tom Emmer’s campaign and gave almost $3 million to the GOP’s Congressional Leadership Fund while Emmer was in charge of it? Did you also know that Emmer was one of several Congresspeople who wrote a letter to Gensler asking him to stop investigating FTX (among other crypto entities)? If you support crypto and politicians, have you ever asked Emmer how he feels about protecting a scam? Taking money from a fraudster? Did he give the money back? Or, do you not care because Emmer says nice things about crypto and criticizes people you dislike? Because he shares your views? If you don’t support crypto and politicians, do you support Gensler’s nonsensical application of obsolete financial rules to stifle a new technology with arbitrary lawsuits and bizarre standards? Or, do you not care about people trying to build or use crypto, so you ignore it? Presidents don’t matter for your financial markets Why does market performance always seem disconnected from politics? The crypto market crashed more than 50% several times under Biden. It crashed more than 70% several times under Trump. Obama was even worse — multiple crashes of 80% or more. On the flip side, it went up 100% after Biden took office, 2,500% after Trump took office, and 5,500% after Obama took office. Seems fishy until you compare that with legacy assets. People make a big deal about the economy under one party or the other, but the economy — like crypto — seems to do whatever it wants regardless of who’s in office. Look at the stock market performance under Republicans and Democrats, adjusted for GDP. Red is Republican, blue is Democrat. Can you see any trend? Neither can I. You can extend this to the wider economy. Let’s do GDP. I counted the number of negative half-years in white: What about inflation? Same. What about employment? Same. What about budget deficits? For statistical consistency, we’ll calculate it as a percentage of GDP. Same. What about oil prices? Same. From a purely investing perspective, you get no advantage or disadvantage under Democrats and Republicans. On the whole, the US economy and its major assets perform as well or as poorly under one as the other. You can take these charts and move the boxes one or two years forward to account for the lag between a change of party and the results of their actions. You’ll get the same result. Know what matters Am I saying presidential elections don’t matter? No. They matter a lot. Every election carries high stakes for your business, employment opportunities, living standards, moral or religious beliefs, mental health, and many aspects of your life. But presidents can’t save your marriage, get you to lose that extra 10 pounds, or fix your financial problems. For people who work in crypto, you won’t find a better friend than Trump. That’s important for people trying to run businesses, build crypto apps, and create new tokens. What about the rest of us? Doesn’t matter. Sure, government policies can tinker around the edges, but crypto is a global asset. Prices come from buyers, sellers, and circumstances beyond any president’s control. The past two years saw a wave of innovation from non-US talent and US companies working abroad. Thousands of new projects have been launched. Hundreds more are waiting for market conditions to improve. Oddly enough, crypto is bigger under Biden now than Trump and Obama. Bitcoin’s hit three new all-time highs! For all those reasons, vote your conscience, not your crypto portfolio. If neither presidential candidate resonates with you, write somebody in. You can scribble “Bitcoin” on your ballot. Most likely, you live in a state where your presidential vote doesn't matter — any of the 43 states painted red or blue in the image below. Partisanship has already made your decision irrelevant. Why not vote for somebody or something you want? You have nothing to lose. I dive more deeply into this topic in a recent post, The US Presidential Election Matters for You, Not Your Crypto Portfolio. https://cryptoiseasy.beehiiv.com/p/the-us-presidential-election-matters-for-you-not-your-crypto-portfolio?embedable=true Mark Helfman publishes the Crypto is Easy newsletter. He is also the author of three books and a top Bitcoin writer on Medium and Hacker Noon. Learn more about him in his bio and connect with him on Tealfeed. Next week, the US will hold a presidential election. At that time, we will vote not for a presidential candidate but for an impression of the candidate we see on TV and at campaign events, read about in newspapers and blogs, and hear about from our friends and family. Like Plato’s shadows on the cave wall, our impressions will reflect only whatever somebody chooses to show us. Modern politics overwhelms us with images crafted and cultivated to fit what we want to see rather than the truth. We latch onto the fragment of reality that we perceive. We never see the full picture. Outside of a few insiders, nobody has enough information to make an informed decision about anybody they vote for. I certainly don’t. Who’s good? Who’s bad? You may think the answers are obvious. Often, it’s not so easy to tell. The devil you don’t know or the other devil you don’t know For example, Congressman Tom Emmer, one of crypto’s most powerful advocates in Congress. He’s the guy who yelled at SEC Chairman Gary Gensler for missing the FTX fraud, among other grievances. Did you know FTX was one of the biggest donors to Tom Emmer’s campaign and gave almost $3 million to the GOP’s Congressional Leadership Fund while Emmer was in charge of it? Did you also know that Emmer was one of several Congresspeople who wrote a letter to Gensler asking him to stop investigating FTX (among other crypto entities)? If you support crypto and politicians, have you ever asked Emmer how he feels about protecting a scam? Taking money from a fraudster? Did he give the money back? Or, do you not care because Emmer says nice things about crypto and criticizes people you dislike? Because he shares your views? If you don’t support crypto and politicians, do you support Gensler’s nonsensical application of obsolete financial rules to stifle a new technology with arbitrary lawsuits and bizarre standards? Or, do you not care about people trying to build or use crypto, so you ignore it? Presidents don’t matter for your financial markets Why does market performance always seem disconnected from politics? The crypto market crashed more than 50% several times under Biden. It crashed more than 70% several times under Trump. Obama was even worse — multiple crashes of 80% or more. On the flip side, it went up 100% after Biden took office, 2,500% after Trump took office, and 5,500% after Obama took office. Seems fishy until you compare that with legacy assets. People make a big deal about the economy under one party or the other, but the economy — like crypto — seems to do whatever it wants regardless of who’s in office. Look at the stock market performance under Republicans and Democrats, adjusted for GDP. Red is Republican, blue is Democrat. Can you see any trend? Neither can I. You can extend this to the wider economy. Let’s do GDP. I counted the number of negative half-years in white: What about inflation? Same. What about employment? Same. What about budget deficits? For statistical consistency, we’ll calculate it as a percentage of GDP. Same. What about oil prices? Same. From a purely investing perspective, you get no advantage or disadvantage under Democrats and Republicans. On the whole, the US economy and its major assets perform as well or as poorly under one as the other. You can take these charts and move the boxes one or two years forward to account for the lag between a change of party and the results of their actions. You’ll get the same result. Know what matters Am I saying presidential elections don’t matter? No. They matter a lot. Every election carries high stakes for your business, employment opportunities, living standards, moral or religious beliefs, mental health, and many aspects of your life. But presidents can’t save your marriage, get you to lose that extra 10 pounds, or fix your financial problems. For people who work in crypto, you won’t find a better friend than Trump. That’s important for people trying to run businesses, build crypto apps, and create new tokens. What about the rest of us? Doesn’t matter. Sure, government policies can tinker around the edges, but crypto is a global asset. Prices come from buyers, sellers, and circumstances beyond any president’s control. The past two years saw a wave of innovation from non-US talent and US companies working abroad. Thousands of new projects have been launched. Hundreds more are waiting for market conditions to improve. Oddly enough, crypto is bigger under Biden now than Trump and Obama. Bitcoin’s hit three new all-time highs! For all those reasons, vote your conscience, not your crypto portfolio. If neither presidential candidate resonates with you, write somebody in. You can scribble “Bitcoin” on your ballot. Most likely, you live in a state where your presidential vote doesn't matter — any of the 43 states painted red or blue in the image below. Partisanship has already made your decision irrelevant. Why not vote for somebody or something you want? You have nothing to lose. I dive more deeply into this topic in a recent post, The US Presidential Election Matters for You, Not Your Crypto Portfolio . The US Presidential Election Matters for You, Not Your Crypto Portfolio The US Presidential Election Matters for You, Not Your Crypto Portfolio The US Presidential Election Matters for You, Not Your Crypto Portfolio https://cryptoiseasy.beehiiv.com/p/the-us-presidential-election-matters-for-you-not-your-crypto-portfolio?embedable=true https://cryptoiseasy.beehiiv.com/p/the-us-presidential-election-matters-for-you-not-your-crypto-portfolio?embedable=true Mark Helfman publishes the Crypto is Easy newsletter. He is also the author of three books and a top Bitcoin writer on Medium and Hacker Noon . Learn more about him in his bio and connect with him on Tealfeed . Crypto is Easy Crypto is Easy Crypto is Easy Medium Medium Medium Hacker Noon Hacker Noon Hacker Noon bio bio Tealfeed Tealfeed

Trump or Harris? Won’t Matter for Your Crypto Portfolio

Escape the Coordi(N)ation Hunger Games: The Matrix of Coordination & Optimistic Exit (Part 1)

Escape the Coordi(N)ation Manq’at awtjata anatt’awinaka: Matriz de Coordinación & Salida Optimista (1ri t’aqa)

Escape the Coordi(N)ation Hunger Games: The Matrix of Coordination & Optimistic Exit (Μέρος 1)

Escape de los Juegos del Hambre de la Coordinación: La Matriz de la Coordinación y la Salida Optimista (Parte 1)

Escape the Coordi(N)ation Hunger Games: Koordinazioaren matrizea eta irteera baikorra (1. zatia)

Escape the Coordi(N)ation Hunger Games: The Matrix of Coordination & Optimistic Exit (Parte 1)

Menekülés a Coordination(N)ation Hunger Games elől: A koordináció és az optimista kilépés mátrixa (1. rész)

コーディネーション・ハンガー・ゲームからの脱出：コーディネーションのマトリックスと楽観的な脱出（パート 1）

Escape the Coordi(N)ation Hunger Games: კოორდინაციის მატრიცა და ოპტიმისტური გასასვლელი (ნაწილი 1)

Melarikan diri dari Coordi(N)ation Hunger Games: Matriks Penyelarasan & Keluar Optimis (Bahagian 1)

Escape the Coordi(N)ation Hunger Games: Matrix of Coordination & Optimistic Exit (1. časť)

Фирор аз Коорди (N)ation Бозиҳои гуруснагӣ: Матритсаи ҳамоҳангсозӣ ва хуруҷи хушбин (Қисми 1)

逃离协调饥饿游戏：协调矩阵与乐观退出（第 1 部分）

Recollections of My Life: Part 3

Navigating the Cybersecurity Hiring Trenches: Challenges, Realities, and Paths Forward

Authors:
(1) Gonzalo J. Aniano Porcile, LinkedIn;
(2) Jack Gindi, LinkedIn;
(3) Shivansh Mundra, LinkedIn;
(4) James R. Verbus, LinkedIn;
(5) Hany Farid, LinkedIn and University of California, Berkeley. Table of Links Abstract and Intro
Data sets
Model
Results
Discussion, Acknowledgements, and References Abstract AI-based image generation has continued to rapidly improve, producing increasingly more realistic images with fewer obvious visual flaws. AI-generated images are being used to create fake online profiles which in turn are being used for spam, fraud, and disinformation campaigns. As the general problem of detecting any type of manipulated or synthesized content is receiving increasing attention, here we focus on a more narrow task of distinguishing a real face from an AI-generated face. This is particularly applicable when tackling inauthentic online accounts with a fake user profile photo. We show that by focusing on only faces, a more resilient and general-purpose artifact can be detected that allows for the detection of AI-generated faces from a variety of GAN- and diffusion-based synthesis engines, and across image resolutions (as low as 128 × 128 pixels) and qualities. 1. Introduction The past three decades have seen remarkable advances in the statistical modeling of natural images. The simplest power-spectral model [20] captures the 1/ω frequency magnitude fall-off typical of natural images, Figure 1(a). Because this model does not incorporate any phase information, it is unable to capture detailed structural information. By early 2000, new statistical models were able to capture the natural statistics of both magnitude and (some) phase [25], leading to breakthroughs in modeling basic texture patterns, Figure 1(b). While able to capture repeating patterns, these models are not able to capture the geometric properties of objects, faces, or complex scenes. Starting in 2017, and powered by large data sets of natural images, advances in deep learning, and powerful GPU clusters, generative models began to capture detailed properties of human faces and objects [16, 18]. Trained on a large number of images from a single category (faces, cars, cats, etc.), these generative adversarial networks (GANs) capture highly detailed properties of, for example, faces, Figure 1(c), but are constrained to only a single category. Most recently, diffusion-based models [2,26] have combined generative image models with linguistic prompts allowing for the synthesis of images from descriptive text prompts like “a beekeeper painting a self portrait”, Figure 1(d). Traditionally, the development of generative image models were driven by two primary goals: (1) understand the fundamental statistical properties of natural images; and (2) use the resulting synthesized images for everything from computer graphics rendering to human psychophysics and data augmentation in classic computer vision tasks. Today, however, generative AI has found more nefarious use cases ranging from spam to fraud and additional fuel for disinformation campaigns. Detecting manipulated or synthesized images is particularly challenging when working on large-scale networks with hundreds of millions of users. This challenge is made even more significant when the average user struggles to distinguish a real from a fake face [24]. Because we are concerned with the use of generative AI in creating fake online user accounts, we seek to develop fast and reliable techniques that can distinguish real from AI-generated faces. We next place our work in context of related techniques. 1.1. Related Work Because we will focus specifically on AI-generated faces, we will review related work also focused on, or applicable to, distinguishing real from fake faces. There are two broad categories of approaches to detecting AI-generated content [10]. In the first, hypothesis-driven approaches, specific artifacts in AI-generated faces are exploited such as inconsistencies in bilateral facial symmetry in the form of corneal reflections [13] and pupil shape [15], or inconsistencies in head pose and the spatial layout of facial features (eyes, tip of nose, corners of mouth, chin, etc.) [23, 33, 34]. The benefit of these approaches is that they learn explicit, semanticlevel anomalies. The drawback is that over time synthesis engines appear to be – either implicitly or explicitly – correcting for these artifacts. Other non-face specific artifacts include spatial frequency or noise anomalies [5,8,12,21,35], but these artifacts tend to be vulnerable to simple laundering attacks (e.g., transcoding, additive noise, image resizing). In the second, data-driven approaches, machine learning is used to learn how to distinguish between real and AIgenerated images [11, 29, 32]. These models often perform well when analyzing images consistent with their training, but then struggle with out-of-domain images and/or are vulnerable to laundering attacks because the model latches onto low-level artifacts [9]. We attempt to leverage the best of both of these approaches. By training our model on a range of synthesis engines (GAN and diffusion), we seek to avoid latching onto a specific low-level artifact that do not generalize or may be vulnerable to simple laundering attacks. By focusing on only detecting AI-generated faces (and not arbitrary synthetic images), we show that our model seems to have captured a semantic-level artifact distinct to AI-generated faces which is highly desirable for our specific application of finding potentially fraudulent user accounts. We also show that our model is resilient to detecting AI-generated faces not previously seen in training, and is resilient across a large range of image resolutions and qualities. This paper is available on arxiv under CC 4.0 license. Authors: (1) Gonzalo J. Aniano Porcile, LinkedIn; (2) Jack Gindi, LinkedIn; (3) Shivansh Mundra, LinkedIn; (4) James R. Verbus, LinkedIn; (5) Hany Farid, LinkedIn and University of California, Berkeley. Authors: Authors: (1) Gonzalo J. Aniano Porcile, LinkedIn; (2) Jack Gindi, LinkedIn; (3) Shivansh Mundra, LinkedIn; (4) James R. Verbus, LinkedIn; (5) Hany Farid, LinkedIn and University of California, Berkeley. Table of Links Abstract and Intro Data sets Model Results Discussion, Acknowledgements, and References Abstract and Intro Abstract and Intro Data sets Data sets Model Model Results Results Discussion, Acknowledgements, and References Discussion, Acknowledgements, and References Abstract AI-based image generation has continued to rapidly improve, producing increasingly more realistic images with fewer obvious visual flaws. AI-generated images are being used to create fake online profiles which in turn are being used for spam, fraud, and disinformation campaigns. As the general problem of detecting any type of manipulated or synthesized content is receiving increasing attention, here we focus on a more narrow task of distinguishing a real face from an AI-generated face. This is particularly applicable when tackling inauthentic online accounts with a fake user profile photo. We show that by focusing on only faces, a more resilient and general-purpose artifact can be detected that allows for the detection of AI-generated faces from a variety of GAN- and diffusion-based synthesis engines, and across image resolutions (as low as 128 × 128 pixels) and qualities. AI-based image generation has continued to rapidly improve, producing increasingly more realistic images with fewer obvious visual flaws. AI-generated images are being used to create fake online profiles which in turn are being used for spam, fraud, and disinformation campaigns. As the general problem of detecting any type of manipulated or synthesized content is receiving increasing attention, here we focus on a more narrow task of distinguishing a real face from an AI-generated face. This is particularly applicable when tackling inauthentic online accounts with a fake user profile photo. We show that by focusing on only faces, a more resilient and general-purpose artifact can be detected that allows for the detection of AI-generated faces from a variety of GAN- and diffusion-based synthesis engines, and across image resolutions (as low as 128 × 128 pixels) and qualities. 1. Introduction The past three decades have seen remarkable advances in the statistical modeling of natural images. The simplest power-spectral model [20] captures the 1/ω frequency magnitude fall-off typical of natural images, Figure 1(a). Because this model does not incorporate any phase information, it is unable to capture detailed structural information. By early 2000, new statistical models were able to capture the natural statistics of both magnitude and (some) phase [25], leading to breakthroughs in modeling basic texture patterns, Figure 1(b). While able to capture repeating patterns, these models are not able to capture the geometric properties of objects, faces, or complex scenes. Starting in 2017, and powered by large data sets of natural images, advances in deep learning, and powerful GPU clusters, generative models began to capture detailed properties of human faces and objects [16, 18]. Trained on a large number of images from a single category (faces, cars, cats, etc.), these generative adversarial networks (GANs) capture highly detailed properties of, for example, faces, Figure 1(c), but are constrained to only a single category. Most recently, diffusion-based models [2,26] have combined generative image models with linguistic prompts allowing for the synthesis of images from descriptive text prompts like “a beekeeper painting a self portrait”, Figure 1(d). Traditionally, the development of generative image models were driven by two primary goals: (1) understand the fundamental statistical properties of natural images; and (2) use the resulting synthesized images for everything from computer graphics rendering to human psychophysics and data augmentation in classic computer vision tasks. Today, however, generative AI has found more nefarious use cases ranging from spam to fraud and additional fuel for disinformation campaigns. Detecting manipulated or synthesized images is particularly challenging when working on large-scale networks with hundreds of millions of users. This challenge is made even more significant when the average user struggles to distinguish a real from a fake face [24]. Because we are concerned with the use of generative AI in creating fake online user accounts, we seek to develop fast and reliable techniques that can distinguish real from AI-generated faces. We next place our work in context of related techniques. 1.1. Related Work Because we will focus specifically on AI-generated faces, we will review related work also focused on, or applicable to, distinguishing real from fake faces. There are two broad categories of approaches to detecting AI-generated content [10]. In the first, hypothesis-driven approaches, specific artifacts in AI-generated faces are exploited such as inconsistencies in bilateral facial symmetry in the form of corneal reflections [13] and pupil shape [15], or inconsistencies in head pose and the spatial layout of facial features (eyes, tip of nose, corners of mouth, chin, etc.) [23, 33, 34]. The benefit of these approaches is that they learn explicit, semanticlevel anomalies. The drawback is that over time synthesis engines appear to be – either implicitly or explicitly – correcting for these artifacts. Other non-face specific artifacts include spatial frequency or noise anomalies [5,8,12,21,35], but these artifacts tend to be vulnerable to simple laundering attacks (e.g., transcoding, additive noise, image resizing). In the second, data-driven approaches, machine learning is used to learn how to distinguish between real and AIgenerated images [11, 29, 32]. These models often perform well when analyzing images consistent with their training, but then struggle with out-of-domain images and/or are vulnerable to laundering attacks because the model latches onto low-level artifacts [9]. We attempt to leverage the best of both of these approaches. By training our model on a range of synthesis engines (GAN and diffusion), we seek to avoid latching onto a specific low-level artifact that do not generalize or may be vulnerable to simple laundering attacks. By focusing on only detecting AI-generated faces (and not arbitrary synthetic images), we show that our model seems to have captured a semantic-level artifact distinct to AI-generated faces which is highly desirable for our specific application of finding potentially fraudulent user accounts. We also show that our model is resilient to detecting AI-generated faces not previously seen in training, and is resilient across a large range of image resolutions and qualities. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

Finding AI-Generated Faces in the Wild: Abstract and Intro

AI-জেনারেটেড ফেস ইন দ্য ওয়াইল্ড খোঁজা: অ্যাবস্ট্রাক্ট এবং ইন্ট্রো

KI-generierte Gesichter in freier Wildbahn finden: Zusammenfassung und Einführung

Encontrar rostros generados por IA en la naturaleza: resumen e introducción

Trouver des visages générés par l'IA dans la nature : résumé et introduction

जंगल में एआई-जनरेटेड चेहरे खोजना: सार और परिचय

AI が生成した顔の発見: 概要と導入

야생에서 AI가 생성한 얼굴 찾기: 초록 및 소개

Encontrando rostos gerados por IA na natureza: resumo e introdução

Поиск лиц, созданных искусственным интеллектом, в дикой природе: краткое содержание и введение

Vahşi Doğada Yapay Zeka Tarafından Üretilen Yüzleri Bulma: Soyut ve Giriş

Tìm kiếm khuôn mặt do AI tạo ra trong thế giới hoang dã: Tóm tắt và giới thiệu

在野外寻找人工智能生成的面孔：摘要和简介

Authors:
(1) Gonzalo J. Aniano Porcile, LinkedIn;
(2) Jack Gindi, LinkedIn;
(3) Shivansh Mundra, LinkedIn;
(4) James R. Verbus, LinkedIn;
(5) Hany Farid, LinkedIn and University of California, Berkeley. Table of Links Abstract and Intro
Data sets
Model
Results
Discussion, Acknowledgements, and References 4. Results Our baseline training and evaluation performance is shown in Table 2. The evaluation is broken down based on whether the evaluation images contain a face or not (training images contained only faces) and whether the images were generated with the same (in-engine) or different (out-of-engine) synthesis engines as those used in training (see Section 2.6). In order to provide a direct comparison of the true positive rate[8] (TPR) for the training and evaluation, we adjust the final classification threshold to yield a false positive rate[9] (FPR) of 0.5%. With a fixed FPR of 0.5%, AI-generated faces are correctly classified in training and evaluation at a rate of 98%. Across different synthesis engines (StyleGAN 1,2,3, Stable Diffusion 1,2, and DALL-E 2) used for training, TPR was varied somewhat from a low of 93.3% for Stable Diffusion 1 to a high of 99.5% for StyleGAN 2, and 98.9% for StyleGAN1, 99.9% for StyleGAN3, 94.9% for Stable Diffusion 2, and 99.2% for DALL-E 2. For faces generated by synthesis engines not used in training (out-of-engine), TPR drops to 84.5% at the same FPR, showing good but not perfect out-of-domain generalization. Across the different synthesis engines not used in training, TPR varied widely with a low of 19.4% for Midjourney to a high of 99.5% for EG3D, and 95.4% for generated.photos. Our classifier generalizes well in some cases, and fails in others. This limitation, however, can likely be mitigated by incorporating these out-of-engine images into the initial training. In a particularly striking result, non-faces – generated by the same synthesis engines as used in training – are all incorrectly classified. This is most likely because some of our real images contain non-faces (see Section 2.1) while all of the AI-generated images contain faces. Since we are only interested in detecting fake faces used to create an account, we don’t see this as a major limitation. This result also suggests that our classifier has latched onto a specific property of an AI-generated face and not some low-level artifact from the underlying synthesis (e.g., a noise fingerprint [8]). In Section 4.1, we provide additional evidence to support this hypothesis. The above baseline results are based on training and evaluating images at a resolution of 512×512 pixels. Shown in Figure 3(a) (solid blue) is the TPR when the training images are down scaled to a lower resolution (256, 128, 64, and 32) and then up scaled back up to 512 for classification. With the same FPR of 0.5%, TPR for classifying an AI-generated face drops fairly quickly from a baseline of 98.0%. The true positive rate, however, improves significantly when the model is trained on images at a resolution of N × N (N = 32, 64, 128, or 256) and then evaluated against the same TPR seen in training, Figure 3(a) (dashed red). As before, the false positive rate is fixed at 0.5%. Here we see that TPR at a resolution of 128 × 128 remains relatively high (91.9%) and only degrades at the lowest resolution of 32×32 (44.1%). The ability to detect AI-generated faces at even relatively low resolutions suggests that our model has not latched onto a low-level artifact that would not survive this level of down-sampling. Shown in Figure 3(b) is the TPR of the classifier, trained on uncompressed PNG and JPEG images of varying quality, evaluated against images across a range of JPEG qualities (ranging from the highest quality of 100 to the lowest quality of 20). Here we see that TPR for identifying an AIgenerated face (FPR is 0.5%) degrades slowly with a TPR of 94.3% at quality 80 and a TPR of 88.0% at a quality of 60. Again, the ability to detect AI-generated faces in the presence of JPEG compression artifacts suggests that our model has not latched onto a low-level artifact. 4.1. Explainability As shown in Section 4, our classifier is highly capable of distinguishing AI faces generated from a range of different synthesis engines. This classifier, however, is limited to only faces, Table 2. That is, when presented with non-face images from the same synthesis engines as used in training, the classifier completely fails to classify them as AI-generated. We posit that our classifier may have learned a semanticlevel artifact. This claim is partly supported by the fact that our classifier remains highly accurate even at resolutions as low as 128×128 pixels, Figure 3(a), and remains reasonably accurate even in the face of fairly aggressive JPEG compression, Figure 3(b). Here we provide further evidence to support this claim that we have learned a structural- or semantic-level artifact. It is well established that while general-purpose object recognition in the human visual system is highly robust to object orientation, pose, and perspective distortion, face recognition and processing are less robust to even a simple inversion [27]. This effect is delightfully illustrated in the classic Margaret Thatcher illusion [31]. The faces in the top row of Figure 4 are inverted versions of those in the bottom row. In the version on the right, the eyes and mouth are inverted relative to the face. This grotesque feature cocktail is obvious in the upright face but not in the inverted face. We wondered if our classifier would struggle to classify vertically inverted faces. The same 10,000 validation images (Section 2.6) were inverted and re-classified. With the same fixed FPR of 0.5%, TPR dropped by 20 percentage points from 98.0% to 77.7%. By comparison, flipping the validation images about just the vertical axis (i.e., left-right flip) yields no change in the TPR of 98.0% with the same 0.5% FPR. This pair of results, combined with the robustness to resolution and compression quality, suggest that our model has not latched onto a low-level artifact, and may have instead discovered a structural or semantic property that distinguishes AI-generated faces from real faces. We further explore the nature of our classifier using the method of integrated gradients [28]. This method attributes the predictions made by a deep network to its input features. Because this method can be applied without any changes to the trained model, it allows us to compute the relevance of each input image pixel with respect to the model’s decision. Shown in Figure 5(a) is the unsigned magnitude of the normalized (into the range [0, 1]) integrated gradients averaged over 100 StyleGAN 2 images (because the StyleGANgenerated faces are all aligned, the averaged gradient is consistent with facial features across all images). Shown in Figure 5(b)-(e) are representative images and their normalized integrated gradients for an image generated by DALL-2, Midjourney, Stable Diffusion 1, and Stable Diffusion 2. In all cases, we see that the most relevant pixels, corresponding to larger gradients, are primarily focused around the facial region and other areas of skin. 4.2. Comparison Because it focused specifically on detecting GANgenerated faces, the work of [23] is most directly related to ours. In this work, the authors show that a low-dimensional linear model captures the common facial alignment of StyleGAN-generated faces. Evaluated against 3,000 StyleGAN faces, their model correctly classifies 99.5% of the GAN faces with 1% of real faces incorrectly classified as AI. By comparison, we achieve a similar TPR, but with a lower 0.5% FPR. Unlike our approach, however, which generalizes to other GAN faces like generated.photos, TPR for this earlier work drops to 86.0% (with the same 1% FPR). Furthermore, this earlier work fails to detect diffusion-based faces because these faces simply do not contain the same alignment artifact as StyleGAN faces. By comparison, our technique generalizes across GAN- and diffusion-generated faces and to synthesis engines not seen in training. We also evaluated a recent state-of-the-art model that exploits the presence of Fourier artifacts in AI-generated images [8]. On our evaluation dataset of real and in-engine AIgenerated faces this model correctly classifies only 23.8% of the AI-generated faces at the same FPR of 0.5%. This TPR is considerably lower than our model’s TPR of 98.0% and also lower than 90% TPR reported in [8]. We hypothesize that this discrepancy is due to the more diverse and challenging in-the-wild real images of our dataset. This paper is available on arxiv under CC 4.0 license. [8] True positive rate (TPR) is the fraction of AI-generated photos that are correctly classified. [9] False positive rate (FPR) is the fraction of real photos that are incorrectly classified. Authors: (1) Gonzalo J. Aniano Porcile, LinkedIn; (2) Jack Gindi, LinkedIn; (3) Shivansh Mundra, LinkedIn; (4) James R. Verbus, LinkedIn; (5) Hany Farid, LinkedIn and University of California, Berkeley. Authors: Authors: (1) Gonzalo J. Aniano Porcile, LinkedIn; (2) Jack Gindi, LinkedIn; (3) Shivansh Mundra, LinkedIn; (4) James R. Verbus, LinkedIn; (5) Hany Farid, LinkedIn and University of California, Berkeley. Table of Links Abstract and Intro Data sets Model Results Discussion, Acknowledgements, and References Abstract and Intro Abstract and Intro Data sets Data sets Model Model Results Results Discussion, Acknowledgements, and References Discussion, Acknowledgements, and References 4. Results Our baseline training and evaluation performance is shown in Table 2. The evaluation is broken down based on whether the evaluation images contain a face or not (training images contained only faces) and whether the images were generated with the same (in-engine) or different (out-of-engine) synthesis engines as those used in training (see Section 2.6). In order to provide a direct comparison of the true positive rate[8] (TPR) for the training and evaluation, we adjust the final classification threshold to yield a false positive rate[9] (FPR) of 0.5%. With a fixed FPR of 0.5%, AI-generated faces are correctly classified in training and evaluation at a rate of 98%. Across different synthesis engines (StyleGAN 1,2,3, Stable Diffusion 1,2, and DALL-E 2) used for training, TPR was varied somewhat from a low of 93.3% for Stable Diffusion 1 to a high of 99.5% for StyleGAN 2, and 98.9% for StyleGAN1, 99.9% for StyleGAN3, 94.9% for Stable Diffusion 2, and 99.2% for DALL-E 2. For faces generated by synthesis engines not used in training (out-of-engine), TPR drops to 84.5% at the same FPR, showing good but not perfect out-of-domain generalization. Across the different synthesis engines not used in training, TPR varied widely with a low of 19.4% for Midjourney to a high of 99.5% for EG3D, and 95.4% for generated.photos. Our classifier generalizes well in some cases, and fails in others. This limitation, however, can likely be mitigated by incorporating these out-of-engine images into the initial training. In a particularly striking result, non-faces – generated by the same synthesis engines as used in training – are all incorrectly classified. This is most likely because some of our real images contain non-faces (see Section 2.1) while all of the AI-generated images contain faces. Since we are only interested in detecting fake faces used to create an account, we don’t see this as a major limitation. This result also suggests that our classifier has latched onto a specific property of an AI-generated face and not some low-level artifact from the underlying synthesis (e.g., a noise fingerprint [8]). In Section 4.1, we provide additional evidence to support this hypothesis. The above baseline results are based on training and evaluating images at a resolution of 512×512 pixels. Shown in Figure 3(a) (solid blue) is the TPR when the training images are down scaled to a lower resolution (256, 128, 64, and 32) and then up scaled back up to 512 for classification. With the same FPR of 0.5%, TPR for classifying an AI-generated face drops fairly quickly from a baseline of 98.0%. The true positive rate, however, improves significantly when the model is trained on images at a resolution of N × N (N = 32, 64, 128, or 256) and then evaluated against the same TPR seen in training, Figure 3(a) (dashed red). As before, the false positive rate is fixed at 0.5%. Here we see that TPR at a resolution of 128 × 128 remains relatively high (91.9%) and only degrades at the lowest resolution of 32×32 (44.1%). The ability to detect AI-generated faces at even relatively low resolutions suggests that our model has not latched onto a low-level artifact that would not survive this level of down-sampling. Shown in Figure 3(b) is the TPR of the classifier, trained on uncompressed PNG and JPEG images of varying quality, evaluated against images across a range of JPEG qualities (ranging from the highest quality of 100 to the lowest quality of 20). Here we see that TPR for identifying an AIgenerated face (FPR is 0.5%) degrades slowly with a TPR of 94.3% at quality 80 and a TPR of 88.0% at a quality of 60. Again, the ability to detect AI-generated faces in the presence of JPEG compression artifacts suggests that our model has not latched onto a low-level artifact. 4.1. Explainability As shown in Section 4, our classifier is highly capable of distinguishing AI faces generated from a range of different synthesis engines. This classifier, however, is limited to only faces, Table 2. That is, when presented with non-face images from the same synthesis engines as used in training, the classifier completely fails to classify them as AI-generated. We posit that our classifier may have learned a semanticlevel artifact. This claim is partly supported by the fact that our classifier remains highly accurate even at resolutions as low as 128×128 pixels, Figure 3(a), and remains reasonably accurate even in the face of fairly aggressive JPEG compression, Figure 3(b). Here we provide further evidence to support this claim that we have learned a structural- or semantic-level artifact. It is well established that while general-purpose object recognition in the human visual system is highly robust to object orientation, pose, and perspective distortion, face recognition and processing are less robust to even a simple inversion [27]. This effect is delightfully illustrated in the classic Margaret Thatcher illusion [31]. The faces in the top row of Figure 4 are inverted versions of those in the bottom row. In the version on the right, the eyes and mouth are inverted relative to the face. This grotesque feature cocktail is obvious in the upright face but not in the inverted face. We wondered if our classifier would struggle to classify vertically inverted faces. The same 10,000 validation images (Section 2.6) were inverted and re-classified. With the same fixed FPR of 0.5%, TPR dropped by 20 percentage points from 98.0% to 77.7%. By comparison, flipping the validation images about just the vertical axis (i.e., left-right flip) yields no change in the TPR of 98.0% with the same 0.5% FPR. This pair of results, combined with the robustness to resolution and compression quality, suggest that our model has not latched onto a low-level artifact, and may have instead discovered a structural or semantic property that distinguishes AI-generated faces from real faces. We further explore the nature of our classifier using the method of integrated gradients [28]. This method attributes the predictions made by a deep network to its input features. Because this method can be applied without any changes to the trained model, it allows us to compute the relevance of each input image pixel with respect to the model’s decision. Shown in Figure 5(a) is the unsigned magnitude of the normalized (into the range [0, 1]) integrated gradients averaged over 100 StyleGAN 2 images (because the StyleGANgenerated faces are all aligned, the averaged gradient is consistent with facial features across all images). Shown in Figure 5(b)-(e) are representative images and their normalized integrated gradients for an image generated by DALL-2, Midjourney, Stable Diffusion 1, and Stable Diffusion 2. In all cases, we see that the most relevant pixels, corresponding to larger gradients, are primarily focused around the facial region and other areas of skin. 4.2. Comparison Because it focused specifically on detecting GANgenerated faces, the work of [23] is most directly related to ours. In this work, the authors show that a low-dimensional linear model captures the common facial alignment of StyleGAN-generated faces. Evaluated against 3,000 StyleGAN faces, their model correctly classifies 99.5% of the GAN faces with 1% of real faces incorrectly classified as AI. By comparison, we achieve a similar TPR, but with a lower 0.5% FPR. Unlike our approach, however, which generalizes to other GAN faces like generated.photos, TPR for this earlier work drops to 86.0% (with the same 1% FPR). Furthermore, this earlier work fails to detect diffusion-based faces because these faces simply do not contain the same alignment artifact as StyleGAN faces. By comparison, our technique generalizes across GAN- and diffusion-generated faces and to synthesis engines not seen in training. We also evaluated a recent state-of-the-art model that exploits the presence of Fourier artifacts in AI-generated images [8]. On our evaluation dataset of real and in-engine AIgenerated faces this model correctly classifies only 23.8% of the AI-generated faces at the same FPR of 0.5%. This TPR is considerably lower than our model’s TPR of 98.0% and also lower than 90% TPR reported in [8]. We hypothesize that this discrepancy is due to the more diverse and challenging in-the-wild real images of our dataset. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv [8] True positive rate (TPR) is the fraction of AI-generated photos that are correctly classified. [9] False positive rate (FPR) is the fraction of real photos that are incorrectly classified.

Finding AI-Generated Faces in the Wild: Results

বন্য এ এআই-উত্পন্ন মুখগুলি সন্ধান করা: ফলাফল

KI-generierte Gesichter in freier Wildbahn finden: Ergebnisse

Encontrar rostros generados por IA en la naturaleza: resultados

Trouver des visages générés par l'IA dans la nature : résultats

जंगल में एआई-जनरेटेड चेहरे खोजना: परिणाम

AI が生成した顔の実例の発見: 結果

야생에서 AI가 생성한 얼굴 찾기: 결과

Encontrando rostos gerados por IA na natureza: resultados

Поиск лиц, созданных искусственным интеллектом, в дикой природе: результаты

Vahşi Doğada Yapay Zekanın Oluşturduğu Yüzleri Bulma: Sonuçlar

Tìm kiếm khuôn mặt do AI tạo ra trong tự nhiên: Kết quả

在野外寻找人工智能生成的面孔：结果

Okay, be honest – you've seen the phrase plastered all over the internet. It's like this weird battle cry for those of us who spend way too much time online. But did you ever stop to think about where it actually came from? Get ready for a nostalgic trip because I'm about to take you down memory lane! "Do a Barrel Roll!" My love affair with Star Fox 64 Back in 1997, I was obsessed with . I don't mean your run-of-the-mill kid obsession; I mean full-blown "this is all I think about" kind of mania. Those space battles, the crazy characters, and, of course, the voice commands! Peppy Hare, Fox McCloud's slightly overprotective wingman, had this way of shouting advice you couldn't quite ignore. Even in the heat of battle, his one-liners made me grin. Star Fox 64 But none of those were line. You know what I'm talking about. When things got dicey when enemies swarmed, Peppy would bellow in glorious desperation: Honestly, at the time, it was just a quick way to dodge out of trouble. But oh boy, did that phrase stick with me. the "Do a Barrel Roll!" A meme is born (and I have no idea how) Years later, I'm not even sure when the internet exploded with I was shocked, amazed, and weirdly proud. Turns out, I wasn't the only one who remembered Peppy's panicked yells. Message boards lit up, people shared images, and, of course, . It felt like this secret inside joke from my childhood had suddenly burst into the mainstream. "Do a Barrel Roll!" memes were absolutely everywhere Google knows a good meme when it sees one. Even Google decided to have a bit of fun! For a while now, if you typed " " into the search bar, the whole dang page would actually spin around. How cool is that? It's like this little bit of nostalgia hidden in the code, just waiting for fans to get a kick out of it. Do a Barrel Roll The thing about memes... It's weird how "Do a Barrel Roll" went from this throwaway line to a full-on internet phenomenon. But that's just how memes work, right? They latch onto our sense of humor, our weird shared experiences, and the simple things we love. video game Barrel Roll Mania: Beyond just memes The influence of this meme didn't stop at message boards and image macros. "Do a Barrel Roll!" has a way of sneaking into all corners of pop culture. Who can resist a good ol' barrel roll, right? And the folks behind Star Fox didn't forget their roots, either. In Star Fox Zero, way back in 2016, good ol' Peppy still hollers "Barrel Roll!" every time he gets those maneuvers going. It's heartwarming to see that they acknowledge the impact the game had. The "Do a Barrel Roll!" spirit pops up in the most unexpected places, too. Remember ? At one point, you're straight-up ordered to "do a barrel roll" in the middle of the chaos. Plus, in the 2022 reboot of Saints Row, they have an achievement called which means, you guessed it – more barrel rolls! It might not be a direct shout-out to Star Fox, but you can't convince me it wasn't at least a little wink and a nod at the classic meme. Saints Row IV "The Spins," And, sometimes, a flying bunny yelling a space pilot command is all it takes to create something awesome and enduring. I mean, at least that's what my younger self would have thought!

"Do a Barrel Roll!" – How My Favourite Video Game Catchphrase Took Over the Internet

Have you ever thought about why databases are so complex internally and why we can’t use a text file to store the data? This blog post aims to go from the most basic database that writes data to text files to a more complex setting where we use LSM-tree and B-tree. We will understand why we need these data structures and more. World’s Simplest Key-Value Store What does the world’s simplest key-value store look like? 2 bash functions that write data to a text file and get data from the text file. db_set(){
	echo "$1, $2" >> database
}

db_get(){
	grep "^$1," database | sed -e "s/^$1,//" | tail -n 1
} Let me explain both the functions. The db_set() function appends to the end of the file without any logic. That means if you update the value of a key, it will not overwrite the value but create a new key value at the end of the file. The db_get() function gets the last occurrence of the key. What’s the problem with this implementation? Doesn’t deal with concurrency control, reclaiming disk space, handling errors, and partially written records. db_get() function has terrible performance as it takes O(n) to lookup for a record. To efficiently find the value of a key, we need a different data structure: . The general idea behind indexes is to keep some additional metadata on the side, which acts as a signpost and helps you locate the data you want. But this comes with a tradeoff as it slows down writes but offers fast reads. an index Hash Indexes I am not planning to dive deep into the workings of hash indexes in this particular blog. I plan on covering it in a future blog, but till then, you can refer to to understand more about it. this blog Let’s again go back to our key-value store. The strategy to improve the lookup speed is to keep an in-memory hash map where every key is mapped to a byte offset in the file. Whenever a new key is appended to the file, the hash map is also updated to reflect the offset. How do we avoid running out of disk space as we are only appending to the file? We break the files into smaller segments and perform on each segment. Compaction is throwing away the duplicate keys in the file and keeping only the most recent update for each key. We can perform compaction over multiple segments together at the same time. Segments are never modified, so the merged segment is written to a new file. It is done in the background while we serve read and write requests using the old segment files. But there are some things to take care of with the implementation and limitations of the hash table index: compaction The hash table must fit in memory because, on disk, it is hard to get the same performance We can’t do range queries efficiently If the database is restarted, the in-memory hash maps are lost. We rebuild the index by reading the entire segment file from beginning to end. It will take a lot of time. We can have a snapshot of each segment’s hash map on disk to solve this issue The database may crash halfway through appending a record that can be resolved by including checksums Writes are appended to the file in a strictly sequential order. A common implementation is to have only 1 write thread and multiple read threads When merging segments, the tells the merging process to discard any previous values for the deleted key tombstone SSTables A simple change to our segment files can help us get rid of maintaining the hash index with all keys. We store the key-value pairs in a sorted-by-key order. We call this format SSTable What advantages do we gain over the hash index? Merging the segment files gets simple and efficient even if the files are larger than the available memory. The merge sort algorithm is used to create new merged segment files, also sorted by key. We now don’t need to maintain an index of all keys in memory. We build a sparse index where we get the keys between which our key lies. It saves disk space and I/O bandwidth. How is this SSTable constructed and maintained? When writes come in we add them to the in-memory tree . When it gets bigger than a threshold, we write all the data from the memtable to the SSTable file. The new SSTable file becomes the most recent segment of the database. While we write to SSTable, writes continue to the new memtable instance. When read requests come in we first search for the key in memtable then in the most recent on-disk segment, then in the next older segment, and so on. And from time to time we run compaction in the background. We can also keep a separate log on disk to which every write is immediately appended for durability purposes. memtable LSM Tree Log-structure merge-tree or LSM tree is a clever algorithm based on the above principle of merging and compacting sorted files. LSM is append-only as we only append all the modifications to the SSTable. The LSM tree algorithm can be slow when looking up keys that do not exist in the database. We first check the memtable and then all the segments. Storage engines use to optimize the algorithm. Bloom Filters A bloom filter is a memory-efficient data structure for approximating the contents of a set. It can tell if a key exists in the database and thus save many unnecessary disk reads for keys that don’t exist. B Tree B Tree keeps key-value pairs sorted by keys like SSTable, which allows efficient key lookups and range queries. It breaks down the database into fixed-size blocks or pages, traditionally 4Kb in size, and reads or writes one page at a time. This design corresponds closely to the underlying hardware, as disks are also arranged in fixed-size blocks. An append-only file where we capture all B tree modifications before applying them to the pages of the tree itself. It is a (WAL or redo log) that helps make the database crash-resilient. We use latches (lightweight locks) to protect our B tree when multiple threads try to access the B tree at the same time. write-ahead log Which one is better? LSM Tree vs B Tree B Tree writes every piece of data at least twice: once to the write-ahead log and once to the tree page itself (and sometimes again when the pages split). The log-structure index rewrites data multiple times due to the repeated compaction and merging of SSTable. It results in one write to the database and multitudinous writes to the disk throughout the database’s lifetime - known as . It is a concern on SSDs as a limited number of overwrites can perform before wearing out. write amplification LSM trees can be compressed better, resulting in smaller files on disk than B trees. B tree storage engine leaves some disk space unused due to fragmentation when the page is split or the row cannot fit in an existing table. LSM trees aren’t page-oriented and periodically rewrite SSTables to remove fragmentation. The compaction process can sometimes interfere with the performance of ongoing reads and writes. At high throughput, the disk’s finite bandwidth is shared between the initial write and the compaction threads running in the background. The bigger the database gets, the more disk bandwidth is required for compaction. In B trees, each key only exists in exactly one place in the index, whereas a log-structured storage engine may have multiple copies of the same key in different segments. It makes B trees attractive in databases that want to offer strong transactional semantics. LSM trees are typically able to sustain a higher write throughput than B trees. But B trees are thought to offer faster read throughput. Parting Words There are more data structures that power databases like skiplist, inverted index, R tree, suffix tree, etc. This blog merely introduces the more famous ones and helps you understand the way they work. If you liked the blog don’t forget to share it on , LinkedIn, Peerlist, and other social platforms and tag me. You may also write to me at to share your thoughts about the blog. Twitter guptaamanthan01@gmail.com References Designing Data-Intensive Applications: The Big Ideas Behind Reliable, Scalable, and Maintainable Systems by Martin Kleppmann Originally published . here

Decoding Database Complexity: A Journey from Text Files to LSM Trees and B-Trees

I recently spoke with Marcel Fohrmann, the Co-Founder and CFO of Subsquid, a decentralized data lake and query engine in Web3, aiming to make canonical blockchain data more accessible and affordable for developers. Our conversation delved into the intricacies of the Web3 landscape, the pressing issues facing the industry, and how decentralized solutions can provide better access to blockchain data. Welcome Marcel. Thanks for joining me. Let’s kick this off with your views on how 2023 went and 2024 will go. Thanks for having me. 2023 was better than most people think. We started on a somber note. Series of upsets, negative events, and regulatory onslaught. Markets bled often, and it felt like winter overall. But some of Web3’s most groundbreaking innovations happened in 2023. Projects with long-term vision were hard at work while the noise was low. Take Layer-2 protocols for example. Optimism, Metis, Base, etc., took giant strides towards faster, cheaper, and more secure blockchain transactions. They latched on to developments in Multi-Party Computation (MPC), Zero-Knowledge Proofs, and other such privacy/security-enhancing solutions. Even L1s like Solana achieved significant growth, with over 2500 developers building on the network and DEX trading volumes surpassing Ethereum. This improved Web3’s diversification and fostered healthy competition. Likewise, we launched Subsquid to make blockchain data more affordable and efficient for developers. The point is 2023 was about improving and expanding the core infrastructure. Scalability, interoperability, security, and a better user experience were the main focus areas. While also improving profitability and sustainability for businesses and for-profit projects. It’s a win-win situation in the making. This will continue and grow in 2024. I’m super bullish about that. We have positive figures and data to show the progress from 2023. Now, it’s about honing in on what’s working and doing it better going forward. We’re at it already. To set realistic expectations, what are the biggest challenges for Web3 right now? In general and, of course, in your sector. Security is one of the main challenges. Hackers are becoming smarter and more driven by the day. With so much money at stake, that’s pretty obvious. Web3 needs stronger security mechanisms that align with the principles of decentralization, autonomy, and privacy. Building them is a tough task, but we’re getting there fast. Decentralized, real-time threat monitoring and mitigation systems are already functional and effective. It’s about raising collective security awareness and making best practices the norm. Cost and scalability are the other main obstacles. They have been for many years now. And though most people don’t see it this way, these are essentially data availability and retrieval problems. Let me explain. Any blockchain-based app or platform must retrieve on-chain data. Be it for updating wallet balances or the latest level of in-game avatars. How fast and cheap you do this determines your solution’s scalability and cost-effectiveness. But Web3 developers have to rely heavily on RPC endpoints with severe rate limitations. This skyrockets data retrieval costs. It also hurts the performance and slows networks down. Especially when there’s high traffic—i.e., low scalability and bad UX. Building a decentralized data retrieval system for fast, responsive, and secure blockchain-based actions became a necessity. We filled that gap. Tell us a bit more about how you’re solving these problems and what that could mean for Web3’s future. Sure. We enable Web3 developers to access canonical blockchain data at 90% less costs compared to RPCs. In Web2, centralized data lakes like BigQuery, Snowflake, and others store and provide access to archival application data at scale. Subsquid is a decentralized, community-driven version of this critical system. It aggregates and filters Web3 app data, making it easily extractable at scale. Without forcing developers to go through complex query processes and whatnot. But while this decentralized data lake takes inspiration from its centralized counterparts, it’s not a replica. It supports infinite horizontal scalability, permissionless data access, credible neutrality, trust-minimized queries, and low maintenance costs. I won’t go into the details here. Anyone who’s interested can read our white paper at . subsquid.io Coming to the impact on Web3’s future. Even with simple, localized queries, Subsquid’s model provides a novel primitive for on-chain data availability. It eliminates the need to maintain expensive archival nodes for historical data access. Unrestricted access to archival data enables developers to build data-intensive blockchain applications at scale. This opens the path for highly complex solutions with deep functionalities that are feasible for providers to maintain and seamless for end-users. That’s the balance we need for Web3’s adoption. Maximalism is all the rage in Web3. But you’ve been vocal about multi-chain systems. Please tell us why they’re important. Bitcoin is among the best tech inventions in mankind’s history. It introduced the stable, functional form of blockchain, which has since enabled such cutting-edge innovations. But with Ethereum revolutionizing general-purpose computing, we’ve come a long way from Bitcoin’s single-chain, single-utility paradigm. The multi-chain landscape is the best way to utilize blockchain’s full potential. It’s not merely about the store of value or payment anymore. DeFi, RWAs, DeSoc, DePIN, GameFi—so many possibilities have emerged, and ignoring them for maximalism is shooting yourself in the foot. From a data access perspective, multi-chain systems help break silos and solve data fragmentation. They allow developers to access the entire data pool easily. Better interoperability, scalability, and innovation, and you can build more robust dApps this way. Thanks for the insight. So, you’ve been doing great work with Subsquid. What’s next on this journey? It’s more important than great. Strengthening Web3’s core infrastructure is a must, and that’s what we’re working to achieve. The next big thing for us is launching support for Solana. Ethereum maxis won’t agree, but Solana is one of the most resilient and high-performance Alt-L1s right now. It also proved that L2s aren’t the only viable scalability solution available. With a Nakamoto Coefficient of 22, Solana has a solid validator distribution while optimizing for latency and throughput. Growth is another reason why we’re excited about Solana. We’ve seen how it bounced back from the ashes after the Alameda crisis last year to become one of the top five chains by market cap. Above all, though, we’re rooting for Solana’s tech stack. Its innovative token standard supports fast, inexpensive, and confidential transactions—great for balancing privacy and UX. Compressed NFTs are exciting as well, and the potential launch of new clients will make the ecosystem more decentralized. Finally, as a progressive project, Subsquid is always at the cutting-edge of innovation. Solana is set to become the hub for futuristic developments, from DePIN to DeSoc. We’ll provide the infrastructure to make that happen—it’s good for all. Before the wrap, what’s your advice to newcomers who might be entering Web3 now that markets are rising again? Learn to filter the noise. Focus on signals. The sooner you achieve this, the smoother your Web3 journey will be. There’s so much to do, it’s overwhelming. Set your priorities, and they’ll guide you better than anything else. I’d also ask developers and project owners to build core solutions, optimized for experience and utility. Hype won’t get you far unless you deliver real, long-term value. No one’s a fool anymore, so don’t take things for granted. The future is yours to build.

The Future of Web3 Data: An Interview with Subsquid CFO & Co-Founder Marcel Fohrmann

Stories, News & Mentions about LATCH

Latch’s Stories on HackerNoon