Superlinked

superlinked

Hey Hackers! Welcome back to another edition of Company of the Week! We like to share an awesome tech company from our tech company database every week, making its evergreen mark on the internet. Company of the Week tech company database This week, we’re excited to showcase Superlinked - A Python framework & cloud infrastructure for AI engineers. Superlinked Being an AI engineer can be challenging, but Superlinked is here to alleviate some of that struggle. It can help you find behavioral clusters of your customers, build an e-commerce product feed, and combine semantic relevance in your search system. Superlinked can help you and your company every step of the way. Want to be featured on HackerNoon’s Company of the Week? Request Your Tech Company Page on HackerNoon! Want to be featured on HackerNoon’s Company of the Week? Request Your Tech Company Page on HackerNoon! Want to be featured on HackerNoon’s Company of the Week? Request Your Tech Company Page on HackerNoon Want to be featured on HackerNoon’s Company of the Week? Request Your Tech Company Page on HackerNoon Meet Superlinked: Fun Facts! Superlinked has worked with a who’s who of tech companies that includes MongoDB, BrandAlley, and Redis. With their powers combined, Superlinked and MongoDB have made building and deploying GenAI-powered applications easier than ever. Superlinked and MongoDB Greg Maxson, Global Lead, AI GTM at MongoDB, said this about the two teams’ partnership: MongoDB’s partnership with Superlinked aims to make it easier for customers to create and maintain entity-level and sub-entity-level vector embeddings for enterprise retrieval augmented generation and other use cases, including analytics or more standard semantic search and recommendation systems. MongoDB’s partnership with Superlinked aims to make it easier for customers to create and maintain entity-level and sub-entity-level vector embeddings for enterprise retrieval augmented generation and other use cases, including analytics or more standard semantic search and recommendation systems. MongoDB’s partnership with Superlinked aims to make it easier for customers to create and maintain entity-level and sub-entity-level vector embeddings for enterprise retrieval augmented generation and other use cases, including analytics or more standard semantic search and recommendation systems. Superlinked also worked with BrandAlley to improve their conversion rate, bumping it by 77%. Ricky Thomas, CTO of BrandAlley, had this to say about Superlinked: say about Superlinked From pilot to production launch of real-time personalization across our shopper experiences, the Superlinked team made sure their platform gave our engineers & product high quality results, lifting our CTR by 77% & AOV by 68%. From pilot to production launch of real-time personalization across our shopper experiences, the Superlinked team made sure their platform gave our engineers & product high quality results, lifting our CTR by 77% & AOV by 68%. From pilot to production launch of real-time personalization across our shopper experiences, the Superlinked team made sure their platform gave our engineers & product high quality results, lifting our CTR by 77% & AOV by 68%. But Superlinked wasn’t done yet because they also worked with Redis to help teams develop and launch GenAI applications. develop and launch GenAI applications Speaking about their partnership, Ash Vijayakanthan, SVP of Cloud Sales and Partnerships at Redis, said this: said this Redis is partnering with Superlinked to make vector search easier to apply to complex data for use cases like RAG with LLMs and e-commerce recommendation systems. Redis is partnering with Superlinked to make vector search easier to apply to complex data for use cases like RAG with LLMs and e-commerce recommendation systems. Redis is partnering with Superlinked to make vector search easier to apply to complex data for use cases like RAG with LLMs and e-commerce recommendation systems. Share Your Company's Story Via HackerNoon Share Your Company's Story Via HackerNoon Share Your Company's Story Via HackerNoon Share Your Company's Story Via HackerNoon Share Your Company's Story Via HackerNoon Superlinked 🤝HackerNoon Business Blogging Superlinked is an open source framework for building search, recommendation, and analytics solutions on top of complex structured+unstructured data. They have partnered with HackerNoon through our Business Blogging Program, and contributed insightful articles on multi-attribute search (featuring Dungeons and Dragons character creation as a use case) and movie recommender systems. Superlinked Business Blogging Program Business Blogging Program multi-attribute search movie recommender systems By improving the way users search for items and making recommendations smarter, e-commerce companies can boost their customer retention and overall conversion rates. What's Business Blogging? What's Business Blogging? HackerNoon’s Business Blogging Program is one of the many ways we help brands grow their reach and connect with the right audience. This program lets businesses publish content directly on HackerNoon to boost brand awareness and build SEO authority by tapping into ours. HackerNoon’s Business Blogging Program many ways Here's what you get: Here's what you get: Backlinks to your website (yes, including CTAs)
A personalized Tech Company News Page featuring your logo, intro, call-to-action, and socials
Full editorial support to make your story shine
Multiple permanent placements on HackerNoon and social media promotions
Stories converted into audio format and distributed via audio RSS feeds
Automatic translation into 12 languages for global reach
Your brand also gains domain authority and SEO via canonical links and the story is distributed across 8 different relevant keyword/tagged pages for better organic discoverability. Backlinks to your website (yes, including CTAs) Backlinks A personalized Tech Company News Page featuring your logo, intro, call-to-action, and socials Tech Company News Page Tech Company News Page Full editorial support to make your story shine Full editorial support Multiple permanent placements on HackerNoon and social media promotions Multiple permanent placements Stories converted into audio format and distributed via audio RSS feeds audio format audio format Automatic translation into 12 languages for global reach Automatic translation into 12 languages 12 languages Your brand also gains domain authority and SEO via canonical links and the story is distributed across 8 different relevant keyword/tagged pages for better organic discoverability. Your brand also gains domain authority and SEO Learn How to Publish with HackerNoon Learn How to Publish with HackerNoon Learn How to Publish with HackerNoon Learn How to Publish with HackerNoon That’s it for this week. See you in the next one! The HackerNoon Team

Meet Superlinked: HackerNoon Company of the Week

ከSuperlinked: HackerNoon የሳምንቱ ኩባንያ ጋር ይተዋወቁ

تعرف على Superlinked: HackerNoon الشركة الأسبوعية

Seznamte se s Superlinked: HackerNoon Company of the Week

Conozca Superlinked: la empresa de la semana de HackerNoon

トップページ > HackerNoon > HackerNoon Company of the Week

Superlinked-пен танысыңыз: аптаның HackerNoon компаниясы

Susipažinkite su „Superlinked“: „HackerNoon“ savaitės kompanija

Hihaona amin'ny Superlinked: HackerNoon Company of the Week

ERROR: invalid target_lang (nso)

Upoznajte Superlinked: HackerNoon kompanija nedelje

Maghanap ng Superlinked: HackerNoon Company ng Semana

Superlinked bilan tanishing: Haftaning HackerNoon kompaniyasi

Build an AI System to Recommend You the Jazziest Pants (Or Any Other Apparel) on the Planet 🩳

Bouw 'n AI-stelsel om jou die Jazziest Pants (Of enige ander klere) op die planeet te aanbeveel

بناء نظام الذكاء الاصطناعي لتشجيعك على أكثر جزر الجاز (أو أي ملابس أخرى) على الكوكب

Xatırladaq ki, dünyanın ən yaxşı futbolçularından biri və ya ən yaxşı futbolçudur.

আপনাকে পৃথিবীর সবচেয়ে জাজি পোশাক (অথবা অন্য কোনও পোশাক) সুপারিশ করার জন্য একটি এআই সিস্টেম তৈরি করুন

Construeix un sistema d'IA per recomanar-te els pantalons Jazziest (o qualsevol altre vestit) del planeta

Construye un sistema de IA para recomendarle los pantalones Jazziest (o cualquier otra ropa) en el planeta

Construír un sistema de IA para recomendarche os pantalóns de jazz (ou calquera outro traxe) no planeta

AIシステムを構築して、あなたに世界で最もジャズのパンツ(または他の服装)を推薦します。

Construiți un sistem AI pentru a vă recomanda cele mai jazz pantalonii (sau orice alt costum) de pe planetă

Постройте систему ИИ, чтобы рекомендовать вам самые джазовые брюки (или любой другой наряд) на планете

Изградите систем АИ да вам препоручи најјаззије панталоне (или било који други накит) на планети

Sizlere gezegendeki en iyi jazzi pantolonları (veya başka herhangi bir elbise) tavsiye etmek için bir AI sistemi oluşturun

Stop Scrolling, Start Building: Create Your Own AI Movie Recommender

Deja de desplazarte y empieza a crear: crea tu propio recomendador de películas con inteligencia artificial

Lopeta vieritys, aloita rakentaminen: Luo oma AI-elokuvasuosittelija

スクロールをやめて構築を始めましょう: 独自の AI 映画レコメンダーを作成しましょう

შეაჩერე გადახვევა, დაიწყე მშენებლობა: შექმენი შენი საკუთარი AI ფილმების რეკომენდაცია

Tika kosala défilement, Bandá kotonga: Salá Recommandé na yo moko ya film AI

Tlogela go Scrolling, Thoma go Aga: Hlama Mokgothaledi wa Gago wa Filimi ya AI

Hagarika Kuzunguruka, Tangira Kubaka: Kora ibyawe bwite bya firime ya AI

Sluta rulla, börja bygga: Skapa din egen AI-filmrekommenderare

หยุดเลื่อนหน้าจอและเริ่มสร้างสรรค์: สร้างระบบแนะนำภาพยนตร์ AI ของคุณเอง

ምጥዋይ ኣቋርጽ፣ ምህናጽ ጀምር: ናይ ገዛእ ርእስኻ AI Movie Recommender ፍጠር

اسکرولنگ بند کریں، بلڈنگ شروع کریں: اپنی خود کی AI مووی تجویز کنندہ بنائیں

O‘tkazishni to‘xtating, qurishni boshlang: O‘z sun’iy intellekt bo‘yicha film tavsiyachisini yarating

Build a Monster-Finding Tool For Your Next D&D Session That Picks the Right Encounter For You

Революционизиращ лов на чудовища: D&D срещи с AI

Revolucionirajući lov na čudovišta: D&D susreti poboljšani umjetnom inteligencijom

Revolutionerende monsterjagter: AI-forbedrede D&D-møder

Revolucionando la caza de monstruos: encuentros de D&D mejorados por IA

شکار هیولاهای انقلابی: برخوردهای D&D با هوش مصنوعی

モンスター狩りに革命を起こす: AI 強化 D&D エンカウンター

क्रान्तिकारी मोन्स्टर हन्टहरू: एआई-एन्हान्स्ड डी एन्काउन्टरहरू

Een revolutie in monsterjachten: AI-verbeterde D&D-ontmoetingen

Kacaanka Ugaadhsiga Bahalnimada: AI-la xoojiyey D&D kulanada

Uwindaji wa Monster wa Mapinduzi: Mikutano ya D&D Iliyoimarishwa kwa AI

புரட்சிகரமான மான்ஸ்டர் ஹன்ட்ஸ்: AI-மேம்படுத்தப்பட்ட D&D என்கவுண்டர்கள்

انقلابی مونسٹر ہنٹس: AI سے بہتر ڈی اینڈ ڈی مقابلے

If you’ve ever used an app or software which lets you collaborate or connect with others, you’ve experienced the power of social+. The term ‘social+’ was coined by former a16z partner, D’Arcy Coolican. In his words: “The best version of every consumer product is the one that is intrinsically social”. Simply put, social+ is . The goals vary and could range from entertainment to fitness, education, and even investment in the stock market (i.e., social+investment). Successful examples include Duolingo, Figma, and Notion. Below, we outline the major benefits of the Social+ approach: any software product where people interact with each other, as means to achieve their goals FEEDING GROWTH LOOPS Social+ nurtures and supports healthy growth loops for both community and product revenue. Inviting collaborators actually makes the product better. A non-social approach is transactional and motivated by referral fees. It has no context, unlike social+ which creates and builds experience in a customer-orientated world. USER-TO-USER ENGAGEMENT The standalone non-social model offers very little engagement beyond the individual transaction, in contrast with social+ which is centred around giving individual experience a place within a meaningful community.  In many cases, the result is achieved by interacting with others. NURTURES RETENTION Social+ nurtures user retention through ongoing interactions that keep bringing people back because of the value provided by other users, not just the product, instead of the greater effort required to achieve this with a non-social model. COMMUNITY AS A COMPETITIVE MOAT The non-social model is vulnerable because it is based on just one level: transaction. The community that forms around a social+ product makes the product better, and harder to copy for competition. There’s little doubt that social+ has the ability to transform each user’s experience from a transactional one-off to becoming part of a meaningful community that keeps them coming back to your product. But there are some serious obstacles in the way of making this happen. So let’s start with what you’re up against. ‍ The TikTok effect The attention economy and the fight for eyeballs is an undeniable force today. In this landscape, personalisation - delivering a highly relevant experience across all touch points - is vital for driving everything from user acquisition to retention to monetisation. One app is successfully delivering that all-important relevant engagement and setting up and meeting user expectations. As a result, it’s eating everyone’s lunch. You’ve probably heard of it… The stratospheric success of TikTok is bad news for other social+ platforms. Why? Because users only have a finite amount of time, so the more they spend on TikTok, the less they have to spend on your product. Winning the attention economy battle is certainly paying off for TikTok. Valued at $50 billion in 2020 and having generated an estimated in 2021, it is now the most valuable startup of all time. $4.6 billion in revenue Wiping the floor Even the tech giant Google isn’t immune to the TikTok effect. The company’s senior vice president Prabhakar Raghavan recently shared that Google’s research revealed that almost looking for a place to go for lunch don’t turn to Google Maps or Search. They go to TikTok or Instagram instead. 40% of young people It’s not just Google Search. As co-founder Daniel Svonava recently commented in , “TikTok is wiping the floor with YouTube.” So, how does TikTok outperform Google and so many others in the attention economy? As Daniel goes on to discuss in the podcast, it achieves this by anticipating the exact content a user wants to see. Before they know they want to see it. He highlights the big difference with TikTok - content discoverability makes the user experience highly relevant and specific to that individual. Superlinked’s his podcast The 40 minutes test Here’s a question: How well would you understand a user after they spend 40 minutes on your app? The answer? After just one 40-minute session and on the basis of just a few interactions, TikTok knows a user more intimately than any social app, and probably even better than the closest friends of that user. Even though it only collects name and email, it’s able to analyse user interactions within the platform in real-time to provide a truly personalised experience. Not as well as TikTok. From the first user session, the platform figures out what the user wants to see and delivers exactly that. It is because of this power TikTok is estimated to generate revenues of by the end of 2022. $12 billion The personalisation challenge So, how can you learn from TikTok’s highly successful approach and serve your users with content that fits perfectly with their interests and goals? After all, users who experienced AI-powered recommendations expect you to provide them with highly personalised content and relevant users to engage with, right from the very first interaction. To achieve this, you need scalable user models which not only ingest user data and their interactions with your product, but also update in real-time. It is possible to build them. That’s what TikTok and Facebook did. But it’s not easy, for four key reasons: You need a lot of data You need a cutting-edge technology stack You need to hire the right to develop, test and iterate the models talent And last but definitely not least, there’s the You could fail to deliver a model that actually works. risk factor: How Social+ can improve metrics across your product However, the significant challenge is well worth the effort. Successful integration of Social+ strategies into a product can supercharge its success across critical measures. If you can get users to engage with your content, you’ll see on relevant articles. 40-60% higher clickthrough rates If you encourage user-to-user interactions, you can enjoy . 2x more users actually creating connections If you have a feed, you can expect to , thanks to much more relevant content. an increase in the time users spend in the app If you let users post their content, you’ll benefit from by delivering their content to relevant users who drive the engagement and the feedback the creators crave. higher creator retention You don't need to build it yourself to win with Social+ Instead of building your own user models and taking on the many associated risks and costs, you could use API and user models to personalise every single touch point in your product with minimal effort. Superlinked’s Disclosure: Jan is a strategic advisor at . Superlinked Benefits of integrating with include: Superlinked : enables users to bring in their data from other social platforms to deliver relevance to your users from the first interaction. Relevance right from the first session Superlinked : client-specific models learn about your users and how they interact with your product - e.g., click, share, like content, or follow other users. Learn about your users with less effort Superlinked’s : APIs are easy to integrate into your product within days. Easily integrated Superlinked’s The takeaway - don’t sleep on Social+ Social+ is big and only getting bigger, but the attention economy is highly competitive. That’s before you even consider the TikTok effect, which is impacting even the biggest players like Google. You can build your own user models to create a truly engaging social+ product if you have the resources for it. It is also possible to make your platform highly relevant at every touch point by using their party APIs, such as social infrastructure. Social+ leads to measurable results such as double the scroll depth, much higher clickthrough rates and user connections, and triple creator retention. Learn more about how Superlinked works . Superlinked’s here

How to Build Engaging Products With Flourishing Communities: Social+

Legacy Social Media is a Sh*tshow In the current social media landscape, a handful of large platforms own most of our data and attention. It has become clear that they take advantage of this power in ways that actively harm their users, other companies, and society at large. These problems have recently reached a level of public awareness that may finally cause things to change. Initiatives like , whose Netflix feature film helped clarify the nefarious effects of social media when left to authoritarian control by a corporation that seeks to maximize shareholder profits instead of value for users. Humane Tech “The Social Dilemma” The monopolies of current social media conglomerates, from Facebook to Tencent, may seem “too big to fail”, in terms of network effects and lock-ins. However, because the user experience keeps deteriorating and there is simultaneous pressure from new entrants, government, NGOs, and now the public, there may be a tipping point approaching where we’ll see an exodus into the “New Social”. As Facebook etc. faces pressures to keep growing or becoming more profitable (in an increasingly saturated market), the “deal” for the users will keep getting worse: More ads, less control, more insidious psychological influence to hook people to their screens. At the same time, we are seeing a range of new companies who seek to rebuild the social media architecture from scratch, having realized how deliberate we need to be when it comes to the public sphere in the digital and our brand new “information highway”. The Emerging New Social Platforms The New Social, at this point, is already underway and its early iterations and building blocks are starting to emerge. We can already describe some of these different building blocks and patterns, with the caution that we don’t yet know how it all fits together, and on what timeline. We are briefly discussing 3 areas in which we are seeing these building blocks emerge: Online communities, social platforms and products, and self-sovereign identity and data management. Communities From where we are now, it sure looks like online communities are the future of social media. Not only has group engagement within legacy social increased dramatically (e.g. Facebook also puts ), but we are seeing online communities across a range of different platforms, from messaging apps (e.g. Telegram, Discord) to new social platforms like Mighty Networks. offers additional insights about this megatrend: increasing focus on groups Data platform GWI Users are joining online communities: already in 2019, a whopping 76% of global internet users have engaged with at least one online community. Brands are increasingly using online communities: 75% of large organizations worldwide had some sort of an online community in 2019. Social Platforms and Products Many of the online communities we discussed above choose an independent platform, which offers them more freedom than legacy social and more elaborate features than messaging apps. Examples are or which focus on hosting online communities, and and which are centered around the online event. Hivebrite Mighty Networks Hopin Bizzaboo Social products are similar in that they provide the infrastructure for communities to meet, but they are centered around a product. Examples include , an anonymous professional social network, and , a social network for entrepreneurs. These products allow different communities to connect, interact, and create value for each other. On Blind, thousands of users share professional information in an anonymous context, allowing them to share things they could not previously (using their official name). CoFoundersLab is a great example of a community formed around a purpose (building companies and finding co-founders) and enabled by a product (profiles, matching, etc.). Blind CoFoundersLab What these different approaches of social platforms and products have in common is that they augment social interaction with technologies and products. They often focus on enabling creators and communities to connect more deeply and collaborate better. The economics of New Social platforms revolve around value creation, and only then taking a piece of the now much larger pie. That shift towards human-centricity and the co-creation of value is at the fundament of New Social. The New Social tech Stack From the developments outlined above, we can start describing what layers of technology are required for the New Social as envisioned above. The main entities we consider are and (for now excluding autonomous software agents). They interact through different protocols or algorithms, which are composable and interoperable. Importantly, users have control over their data and selectively decide what they share with other users, communities, and algorithms. The intersection of what they share is what we call “ ”. Think of it as an avatar that can move on the spectrum from anonymity to pseudonymity all the way to full transparency, based on the context. users communities Persona Protocols and algorithms necessary for the New Social can be organized into the following main layers: The user layer (wallets, credentials, user data) The UX layer (platforms, interaction, experience) The computation layer (ZK-systems, AIs, reputation models, etc) The political layer (decision-making and rules) The monetary/legal layer (value distribution and exchange) The User Layer: Self-sovereign User Data Closest to the user in the New Social tech stack is the user layer. In the new paradigm, user data is owned at the edges, not within the applications or platforms. One could imagine a “personal data locker” that contains all relevant user information. It’s astonishing how large our data footprint has become, consider the following kinds of data that could all be included: Credentials (identifiers, badges, proofs, etc.) Behavior and personal data (everything that falls under GDPR) Created data (everything that falls under DRM) In the New Social tech stack, all of that data is owned by the user. The cryptographically secured data locker selectively shares data with approved users, communities, or algorithms, creating a unique Persona for each interaction. Versions of this interaction are also described as “self-sovereign identity”, meaning users are sovereign over their online identity (their data footprint). In the future, the user layer could be powered by zero-knowledge technology and secure multi-party computation (SMC), which would enable proofs about data and computation on shared data without losing privacy. The UX Layer: Communication and Collaboration The UX layer of the New Social tech stack is where all the products and platforms that we described above sit. It’s where users communicate and collaborate with each other, across different media. Communities usually manifest on one or several Social Products or Platforms, providing a political (and monetary frame) for user interaction. In that sense, the UX layer could be described as the bridge which connects all of the layers into a concrete user experience. Most users will only be aware of the UX they interact with on a given product, not all the other layers in the background that make the experience possible. The Computation Layer: Access and Reputation The computation layer is how personas connect to one another. Users connect their data locker, especially their social graph, to the UX layer. This enables them to interact as a persona with others in the context of a specific community. These interactions will tend to happen in the context of a specific community or at least within the context of a specific New Social platform or product. The computation layer regulates which communities, and which spaces within these communities any given Persona has access to. Similarly, Personas may have a different reputation within different communities or platforms. Both access and reputation have a strong influence on the political and monetary layers. In this way, the computation layer also extends the persona into the political and monetary layers in the context of a specific community/platform/product. Superlinked connects the user layer to the computation layer, powering the UX of the New Social. is a great example of a company operating across the New Social tech stack: Users are in control of their social graph (and other data) across platforms and products, in line with a user layer in the self-sovereign identity paradigm. Users can connect their social graph with any given platform but always own and are in control of their data. Superlinked Superlinked mostly lives on the computation layer, where it provides APIs to enhance social products and platforms, allowing them to offer better features on the UX layer. With the increased computational capacity provided by Superlinked, these social platforms and products can improve the quality of engagement, and how much value is created as a result. Superlinked allows the platforms of the New Social to create more value for their users by enriching interactions with social intelligence. Web 3 Adds Two New Layers to the New Social Tech Stack We have also seen the emergence of tokenized communities in crypto, or “Social tokens”, as explained . The fundamental shift is that blockchain-based tokens provide a new way for online communities to monetize and distribute ownership. This new capacity could eventually also power new social platforms, allowing users to become co-owners of the platform, and communities to monetize through their own token. We are calling this additional layer . here the monetary/legal layer Once every stakeholder holds tokens, yet another layer opens up: . Decisions on how the monetary/legal layer evolves, along with any other type of decision from product to staffing, can now be made collectively by all relevant stakeholders. We are seeing a whole range of different mechanisms, from one-person-one-vote to liquid democracy — being used to govern digital platforms and communities. Smart contracts and the increasing prevalence of ready-to-use frameworks have enabled this innovation. The political layer connects to the monetary layer through the governance mechanism, which is often powered by tokens or NFTs. The political layer Even though previous technologies have also been embedded in monetary/legal and political structures, these structures were provided by the respective jurisdiction and not up for debate. In Web 3, they become part of the tech stack. In Web 3, there are a growing number of products that have their own communities that co-own the product through tokens. The components they use to communicate, make decisions, and monetize are all open-source, composable, and interoperable. All the while, user information, and funds are self-custodied through users’ Web 3 wallets, and not on the platform. The monetary/legal layer As alluded to above, the monetary/legal layer concerns the assets, cash flows, and contractual agreements of the different communities and personas. There has been a clear trend of social media to more closely interconnect with financial flows already, from ad revenues to influencers and the growing Creator economy. We expect that in the New Social tech stack, these monetary flows will become more transparent and made accessible to users. For example, users could selectively monetize their data, as well as get paid for actions like curation and participation. At the same time, they would own part of the communities and platforms they participate in. The political layer (decision-making and rules) Online communities and even New Social platforms will be powered by different decision-making mechanisms than simple top-down command-and-control. Even Facebook has set up an ethical council to at least try to make their decision-making consider their users as a stakeholder. In the New Social, most communities and platforms will be user-owned and governed. Different users will have different decision-making power in different communities and on different platforms, depending on their reputation within that context and on the given governance mechanism. New Social: An attempt towards healthier social media The modular, interoperable tech stack described in this piece could turn legacy media upside down and enable human-centric, empowering online interactions. In contrast to the current paradigm where the platforms own all the data and users have no control, all personal data lives on the user layer, controlled by every user. The required computation necessary to enable social interactions at scale, from the curation of content to the social intelligence provided by , happens with user consent and while preserving privacy. Furthermore, algorithms may become increasingly transparent and modular too, allowing for more transparency and optionality for both platforms and users. Superlinked These developments could break open both the user data and computation monopolies of legacy social media giants, allowing emerging social platforms and products to operate on the same terms and compete effectively. Under these conditions, it seems obvious that users will eventually make the choice for an online experience that helps them create value and connect with others, rather than being lured into binge-watching and “doom-scrolling” for the sake of advertising revenues.

Web 3 Adds Two Layers to the New Social Tech Stack

How to get unstuck when you’re feeling down You work hard. Every day you are prospecting clients, pumping out valuable content, and perfecting your craft. But do you ever feel like, no matter how much work you put in, maybe you’re just not the type of person who does ? this I’m not the type of person who can start their own business… I’m not the kind of person who could ever write a book… How can I call myself a “consultant”? When you do get results, maybe you brush them off as just luck. Or you feel like you have to be “tricking” people somehow into thinking you’re smarter than you are. Well, I have some great news for you. It’s not just you! by the “It is estimated that 70% of people will experience at LEAST one episode of this Impostor Phenomenon in their lives” According to a 2011 report International Journal of Behavioral Science Everybody struggles with these kinds of thoughts, especially if you are trying to achieve a new goal. Something that you’ve never done before. I’ve been struggling with this over the last 2 weeks myself. Last night I was laying awake at 3 in the morning thinking to myself “What the hell am I even DOING with my time?” and stressing myself out! It can be so easy to feel bad about yourself. But you know what? After working my way through it, I realized I was actually productive this week. way I went to all of my classes, was able to do all of my homework assignments on time, went to an Innovation Summit in another city, I finished quite a few lessons in a marketing course I’m taking, and I finished not just one — but client projects while I was at it. On top of that, I did some errand running for household stuff and was able to check off my 3 priorities for the week. TWO So what gives? How can I be doing all of this stuff, and still not feel good about myself? For me, the worst thing about Impostor Syndrome is how it can mask the of the tasks that I’m spending my time on during the day. value I wasn’t thinking about all of the stuff that I did this week while I was stressed. Do you know what I was thinking about? I too busy beating myself up and making myself feel guilty! I was making myself feel guilty about a concert I went to on Tuesday (I was able to do a lot of stuff for school, but not that many things for my clients that day), and for chilling out on Thursday and… Well… Basically not doing anything. I was beating myself up about not walking my dog 30 minutes every single day, and that I haven’t practiced German lately, and I haven’t been reading every day (even though I’ve already read 5 books this month!). Anyway, all of this stuff was going through my mind last night — and I wanted to talk a little bit about the strategies I use to try to overcome the feeling. Because the negative anxiety feelings can definitely paralyze you! It can make you feel like nothing you do moves the needle in your business or makes a difference, why should you bother trying something else, what would you even try? There are 2 strategies that I use that I find helpful. super But before I get into that, here is the first thing you need to accept: If you are practicing your “craft” (whatever you think you “can’t” do, or “aren’t the type of person” who does “that”), then the kind of person who does that! you are If you have a novel you want to get out there that you think about all the time, then who’s a writer! you are the kind of person If you have a cool new product that’s always on your mind, then who’s an entrepreneur. you are the kind of person Whatever types of ideas, or dreams, or goals that gravitate to you the most naturally? They come to you easily, because YOU’RE “that” type of person ;) Now let’s get down to the nitty gritty. Here are my 2 simple strategies for overcoming Impostor Syndrome: #1 Commit to Pushing Through — NO MATTER WHAT! Steven Pressfield wrote an entire book called “The War of Art” to hammer this one point home: If you want to be a writer, or an entrepreneur, or a polyglot, or whatever it is you want then… You HAVE to show up every day! The only way to get good at something is to chip away at it, consistently, until you reach that tipping point. Remember, the only way to lose is if you don’t show up at all. What if you’re not sure what to do next? What should you work on? Gary Vaynerchuk has a philosophy called . “The Clouds & The Dirt” To sum it up: The principles behind your greatest product, the message you want to express to the masses through your art — all of that feel good “higher plane” crap? That is the clouds. The dirt is the grime on your face after you’ve spent a day sharpening your craft. It’s reading HUGE quantities of your favorite literature to cram all those good ideas in there. It’s reading your industry journals to keep up to date. It’s studying with mentors, or in high quality courses to in an extremely focused way improve your skills. And those are the only things that matter. Gary says that most people waste their time living in the middle. If most of your day is spent putting out fires in your company, or debating on which payment processor to use for online check out then you’re not in the clouds or the dirt. How can you use this as a strategy to fight back against Impostor Syndrome? Make a list of “agreed upon” activities in advance! and get super clear on what it is that you actually want. Use a Tony Robbins style of goal planning, Ask yourself “What NEEDS to happen for me to feel successful?”. Be honest with yourself. Do you have a revenue goal? A # of email subscribers? To be “successful” as an author do you need to publish 1 book? 3 books? 12 books? Should those books sell a certain number of copies? How long of a conversation do you need to be able to have in your target language? Those are just some suggestions, but it’s okay if your answer is different! The key is to understand what needs to happen for you to feel that you’ve successfully accomplished your goal. Next, brainstorm a MASSIVE list of every single thing you could possibly do to get yourself to this success. You will probably be guessing — and that’s okay! Tony calls this a Massive Action Plan (MAP), you should have a paper overflowing with ideas on next steps you can take to reach your goals. there is a lot of helpful advice in there that explains how to take a long term goal (like “To feel successful as an Entrepreneur I need to have 3 businesses with a revenue of over $700,000”) and into smaller sub goals. If you have time to watch the video I linked to Tony’s goal planning workshop, chunk it Don’t skip it! Alright. So, now you have a few things here that will guide you in creating you “agreed on behaviors” list: Gary’s “Clouds & Dirt” philosophy A clear idea on what you need to do to that you’ve successfully accomplished your goal feel A big brainstorm of everything you think you could try to get yourself there Using these 3 things to guide you, start reflecting on what you need to “close the gap”! How will you get from here to there? Some things to start you off on what you could include on your “A-Okay!” list would be: reading books, taking courses, working on your company culture, of course sitting down to do your craft (so a writer for example, sitting down to write or a freelancer working on a client project OR creating a sample project) would all be things that could help you get to where you need to go. Here are some of the things that are on my “Good to Go” list (you can call the list whatever you want lol): Working on a high quality marketing course, reading a book, reviewing relevant academic research, working on writing something for a client, writing a high value article, doing lead generating activities. You could even list specific websites you want to stick to, listing sites that have consistent high quality information and are low on distractions. But if you can think critically and agree in advance with what is okay for you to be doing, then there’s no reason to feel bad about it after the fact. You have already sat down and decided that the type of person who is a successful writer does X, Y, Z and taken the decision making out of the process. #2 — Count your wins! Look, when you’re trying to make positive progress obviously a lot of your attention is focused on “where you’ve got to go”. But you shouldn’t let your wins vanish into the night either! If you use a paper journal, an online “gamification” tool like , or use a Google Keep note (if you turn on “checkboxes” — Google will show you items 1–3 that need to be checked “And 27 checked items!” or whatever at the bottom) then you will be able to quantify your progress as evidence of your completed tasks pile up. Habitica Seeing your accomplishments stockpile like this can help your quantify what’s actually coming out of your spent time. Reviewing this list of wins can also be a great source of strength when you’re feeling stuck. The easiest way to crawl up from the lowest lows of Impostor Syndrome is to change your focus. When you’re making performance driven decisions, of course your mind is going to be on your endgame a lot. An easy way to shift your state is to instead look back, and simply appreciate how far you’ve already come :) — — — — — Sign up here! Psst… Did you find this article helpful? ^ Sign up for my email list for actionable business and marketing tips — once a week.

2 Ways I Fight Impostor Syndrome

Authors:
(1) Mengshuo Jia, Department of Information Technology and Electrical Engineering, ETH Zürich, Physikstrasse 3, 8092, Zürich, Switzerland;
(2) Gabriela Hug, Department of Information Technology and Electrical Engineering, ETH Zürich, Physikstrasse 3, 8092, Zürich, Switzerland;
(3) Ning Zhang, Department of Electrical Engineering, Tsinghua University, Shuangqing Rd 30, 100084, Beijing, China;
(4) Zhaojian Wang, Department of Automation, Shanghai Jiao Tong University, Dongchuan Rd 800, 200240, Shanghai, China;
(5) Yi Wang, Department of Electrical and Electronic Engineering, The University of Hong Kong, Pok Fu Lam, Hong Kong, China;
(6) Chongqing Kang, Department of Electrical Engineering, Tsinghua University, Shuangqing Rd 30, 100084, Beijing, China. Table of Links Abstract and 1. Introduction 2. Evaluated Methods 3. Review of Existing Experiments 4. Generalizability and Applicability Evaluations and 4.1. Predictor and Response Generalizability 4.2. Applicability to Cases with Multicollinearity and 4.3. Zero Predictor Applicability 4.4. Constant Predictor Applicability and 4.5. Normalization Applicability 5. Numerical Evaluations and 5.1. Experiment Settings 5.2. Evaluation Overview 5.3. Failure Evaluation 5.4. Accuracy Evaluation 5.5. Efficiency Evaluation 6. Open Questions 7. Conclusion Appendix A and References 5.5. Efficiency Evaluation The evaluation of computational efficiency in this section includes a comparison of DPFL and PPFL methods, a general review of DPFL methods, and individual examinations of specific groups of DPFL approaches. 5.5.1. DPFL Performance vs. PPFL Performance As illustrated in Fig. 8, the computational efficiency of PPFL methods notably outperforms that of DPFL approaches. This advantage largely stems from the inherent nature of PPFL approaches, i.e., avoiding the need for a training process, in contrast to DPFL methods. Fundamentally, the absence of training in PPFL methods is attributed to their reliance on predefined physical models of the system, assuming that all necessary physical parameters are known and accurate. However, this assumption might not always be valid, particularly in distribution grids. 5.5.2. DPFL Performance: General View The computational costs of the majority of DPFL methods exhibit superlinear growth with increasing system size, approaching or even exceeding quadratic growth rates, even for methods with closed-form solutions. Much of the computational effort is devoted to model fitting, particularly for DPFL methods that require tuning of hyperparameters. For instance, clustering-based methods have to tune the number of clusters by cross-validation, while ridge-regression-related methods need cross-validation to adjust the regularization factor. This tuning process can be computationally intensive, as it involves systematically searching for better hyperparameter values, which often require multiple iterations of model training and validation. It is important to highlight that, aside from the DPFL methods which encountered OOT failures, several DPFL approaches, particularly in the 1354-bus-S case, showed computational times exceeding 300 seconds. Considering that the real-time dispatch in some countries operates on a five-minute cycle, these slower DPFL methods fall short of meeting the real-world computational speed requirements. Given the considerable computational burden many DPFL methods face in large systems, exploring ways to accelerate their training becomes a practical concern. However, it is crucial to clarify that not all DPFL methods suffer from excessive computational efforts. Recursive DPFL methods, in particular, demonstrate more promising speed, which will be further explained in the subsequent discussion. 5.5.3. DPFL Performance: Individual View In the following, we present performance analyses of particular groups of methods. High-burden Methods High-burden methods refer to the methods that result in OOT failures in large-scale test cases. Such a high burden is primarily due to the reliance on optimization programming techniques. Examples include LS_HBLD, LS_HBLE, LS_WEI, SVR, SVR_CCP, SVR_RR, DRC_XYD, DRC_XYM, DRC_XM, LCP_BOX, LCP_BOXN, LCP_COU, LCP_COUN, LCP_JGD, and LCP_JGDN. Even for unconstrained quadratic programming, e.g., LCP_JGDN, the computational demands can quickly become overwhelming and surpass memory capabilities. For a situation where the optimization models have more complex constraints like distributionally robust or chance constraints, the computational demand can be even higher. On the other hand, LS_LIFX and LS_LIFXi, which can also be considered as, fail in large-scale tests as they significantly increase the dataset size by dimension lifting, exceeding MATLAB’s maximum array size limitations. Additionally, SVR_POL and LS_GEN, the last two highburden methods, experience significant slowdowns at a system size of 1354 buses, a consequence of their specialized design of 3rd-order polynomial-kernel fitting and iterative training, respectively, as explained before. Clustering-based Methods Clustering-based methods like RR_KPC and PSL_CLS often outperform other DPFL approaches in accuracy but are notably more time-intensive. For instance, processing a 1354-bus-S case requires over 1000 seconds with these methods. The primary burden lies in the clustering phase, which is particularly demanding for large-scale datasets. This is because both RR_KPC’s K-plane clustering and PSL_CLS’s K-means clustering involve iterative searches for better numbers of clusters, an inherently resource-intensive process. Furthermore, these approaches require the tuning of hyperparameters through cross-validation, adding more layers of iterative computation. E.g., RR_KPC needs adjustments in both the number of clusters and regularization parameters; such a combinational search plus the clustering process significantly increases the computational demand. Recursive Methods Methods PLS_REC and LS_REC are distinguished by their recursive nature, setting them apart from techniques that rely on the entire training dataset for model fitting. Instead, PLS_REC and LS_REC incrementally refine the model through hundreds or thousands of updates. For instance, in the 1343-bus-S case, PLS_REC begins by using 40% of the training dataset, which equates to the old dataset with 1200 data samples, to establish an initial linear model via the ordinary partial least squares approach. Subsequently, PLS_REC imitates an online procedure by incorporating each new data point as it arrives, thereby updating the linear model recursively. Ultimately, PLS_REC performs 1800 updates to derive the final model, which amounts to the entire training set comprising 3000 data points. Note that the entire recursive process is remarkably efficient, i.e., 50.85 seconds in total or approximately 0.028 seconds for each point-wise update for this test case. Considering a measuring time resolution of one point every half second, where five minutes yield 600 data points, these data can be recursively integrated by PLS_REC in roughly 16.8 seconds using a consumer-grade laptop. This efficiency aligns well with the requirements of five-minute real-time dispatch scenarios. Well-rounded Methods Methods that balance high accuracy, efficiency, and practicality (where “practical” means not significantly increasing the complexity of the application) include LS_COD, DLPF_C, PLS_NIP, and PLS_REC. While the transformationbased RR_VCS method has advantages in both accuracy and computational time, its model is challenging to apply in optimization and similar applications. This paper is available on arxiv under CC BY-NC-ND 4.0 Deed (Attribution-Noncommercial-Noderivs 4.0 International) license. Authors: (1) Mengshuo Jia, Department of Information Technology and Electrical Engineering, ETH Zürich, Physikstrasse 3, 8092, Zürich, Switzerland; (2) Gabriela Hug, Department of Information Technology and Electrical Engineering, ETH Zürich, Physikstrasse 3, 8092, Zürich, Switzerland; (3) Ning Zhang, Department of Electrical Engineering, Tsinghua University, Shuangqing Rd 30, 100084, Beijing, China; (4) Zhaojian Wang, Department of Automation, Shanghai Jiao Tong University, Dongchuan Rd 800, 200240, Shanghai, China; (5) Yi Wang, Department of Electrical and Electronic Engineering, The University of Hong Kong, Pok Fu Lam, Hong Kong, China; (6) Chongqing Kang, Department of Electrical Engineering, Tsinghua University, Shuangqing Rd 30, 100084, Beijing, China. Authors: Authors: (1) Mengshuo Jia, Department of Information Technology and Electrical Engineering, ETH Zürich, Physikstrasse 3, 8092, Zürich, Switzerland; (2) Gabriela Hug, Department of Information Technology and Electrical Engineering, ETH Zürich, Physikstrasse 3, 8092, Zürich, Switzerland; (3) Ning Zhang, Department of Electrical Engineering, Tsinghua University, Shuangqing Rd 30, 100084, Beijing, China; (4) Zhaojian Wang, Department of Automation, Shanghai Jiao Tong University, Dongchuan Rd 800, 200240, Shanghai, China; (5) Yi Wang, Department of Electrical and Electronic Engineering, The University of Hong Kong, Pok Fu Lam, Hong Kong, China; (6) Chongqing Kang, Department of Electrical Engineering, Tsinghua University, Shuangqing Rd 30, 100084, Beijing, China. Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Evaluated Methods 3. Review of Existing Experiments 3. Review of Existing Experiments 4. Generalizability and Applicability Evaluations and 4.1. Predictor and Response Generalizability 4. Generalizability and Applicability Evaluations and 4.1. Predictor and Response Generalizability 4.2. Applicability to Cases with Multicollinearity and 4.3. Zero Predictor Applicability 4.2. Applicability to Cases with Multicollinearity and 4.3. Zero Predictor Applicability 4.4. Constant Predictor Applicability and 4.5. Normalization Applicability 4.4. Constant Predictor Applicability and 4.5. Normalization Applicability 5. Numerical Evaluations and 5.1. Experiment Settings 5. Numerical Evaluations and 5.1. Experiment Settings 5.2. Evaluation Overview 5.2. Evaluation Overview 5.3. Failure Evaluation 5.3. Failure Evaluation 5.4. Accuracy Evaluation 5.4. Accuracy Evaluation 5.5. Efficiency Evaluation 5.5. Efficiency Evaluation 6. Open Questions 6. Open Questions 7. Conclusion 7. Conclusion Appendix A and References Appendix A and References 5.5. Efficiency Evaluation The evaluation of computational efficiency in this section includes a comparison of DPFL and PPFL methods, a general review of DPFL methods, and individual examinations of specific groups of DPFL approaches. 5.5.1. DPFL Performance vs. PPFL Performance 5.5.1. DPFL Performance vs. PPFL Performance 5.5.1. DPFL Performance vs. PPFL Performance As illustrated in Fig. 8, the computational efficiency of PPFL methods notably outperforms that of DPFL approaches. This advantage largely stems from the inherent nature of PPFL approaches, i.e., avoiding the need for a training process, in contrast to DPFL methods. Fundamentally, the absence of training in PPFL methods is attributed to their reliance on predefined physical models of the system, assuming that all necessary physical parameters are known and accurate. However, this assumption might not always be valid, particularly in distribution grids. 5.5.2. DPFL Performance: General View 5.5.2. DPFL Performance: General View 5.5.2. DPFL Performance: General View The computational costs of the majority of DPFL methods exhibit superlinear growth with increasing system size, approaching or even exceeding quadratic growth rates, even for methods with closed-form solutions. Much of the computational effort is devoted to model fitting, particularly for DPFL methods that require tuning of hyperparameters. For instance, clustering-based methods have to tune the number of clusters by cross-validation, while ridge-regression-related methods need cross-validation to adjust the regularization factor. This tuning process can be computationally intensive, as it involves systematically searching for better hyperparameter values, which often require multiple iterations of model training and validation. It is important to highlight that, aside from the DPFL methods which encountered OOT failures, several DPFL approaches, particularly in the 1354-bus-S case, showed computational times exceeding 300 seconds. Considering that the real-time dispatch in some countries operates on a five-minute cycle, these slower DPFL methods fall short of meeting the real-world computational speed requirements. Given the considerable computational burden many DPFL methods face in large systems, exploring ways to accelerate their training becomes a practical concern. However, it is crucial to clarify that not all DPFL methods suffer from excessive computational efforts. Recursive DPFL methods, in particular, demonstrate more promising speed, which will be further explained in the subsequent discussion. 5.5.3. DPFL Performance: Individual View 5.5.3. DPFL Performance: Individual View 5.5.3. DPFL Performance: Individual View In the following, we present performance analyses of particular groups of methods. High-burden Methods High-burden Methods High-burden methods refer to the methods that result in OOT failures in large-scale test cases. Such a high burden is primarily due to the reliance on optimization programming techniques. Examples include LS_HBLD, LS_HBLE, LS_WEI, SVR, SVR_CCP, SVR_RR, DRC_XYD, DRC_XYM, DRC_XM, LCP_BOX, LCP_BOXN, LCP_COU, LCP_COUN, LCP_JGD, and LCP_JGDN. Even for unconstrained quadratic programming, e.g., LCP_JGDN, the computational demands can quickly become overwhelming and surpass memory capabilities. For a situation where the optimization models have more complex constraints like distributionally robust or chance constraints, the computational demand can be even higher. On the other hand, LS_LIFX and LS_LIFXi, which can also be considered as, fail in large-scale tests as they significantly increase the dataset size by dimension lifting, exceeding MATLAB’s maximum array size limitations. Additionally, SVR_POL and LS_GEN, the last two highburden methods, experience significant slowdowns at a system size of 1354 buses, a consequence of their specialized design of 3rd-order polynomial-kernel fitting and iterative training, respectively, as explained before. Clustering-based Methods Clustering-based Methods Clustering-based methods like RR_KPC and PSL_CLS often outperform other DPFL approaches in accuracy but are notably more time-intensive. For instance, processing a 1354-bus-S case requires over 1000 seconds with these methods. The primary burden lies in the clustering phase, which is particularly demanding for large-scale datasets. This is because both RR_KPC’s K-plane clustering and PSL_CLS’s K-means clustering involve iterative searches for better numbers of clusters, an inherently resource-intensive process. Furthermore, these approaches require the tuning of hyperparameters through cross-validation, adding more layers of iterative computation. E.g., RR_KPC needs adjustments in both the number of clusters and regularization parameters; such a combinational search plus the clustering process significantly increases the computational demand. Recursive Methods Recursive Methods Methods PLS_REC and LS_REC are distinguished by their recursive nature, setting them apart from techniques that rely on the entire training dataset for model fitting. Instead, PLS_REC and LS_REC incrementally refine the model through hundreds or thousands of updates. For instance, in the 1343-bus-S case, PLS_REC begins by using 40% of the training dataset, which equates to the old dataset with 1200 data samples, to establish an initial linear model via the ordinary partial least squares approach. Subsequently, PLS_REC imitates an online procedure by incorporating each new data point as it arrives, thereby updating the linear model recursively. Ultimately, PLS_REC performs 1800 updates to derive the final model, which amounts to the entire training set comprising 3000 data points. Note that the entire recursive process is remarkably efficient, i.e., 50.85 seconds in total or approximately 0.028 seconds for each point-wise update for this test case. Considering a measuring time resolution of one point every half second, where five minutes yield 600 data points, these data can be recursively integrated by PLS_REC in roughly 16.8 seconds using a consumer-grade laptop. This efficiency aligns well with the requirements of five-minute real-time dispatch scenarios. Well-rounded Methods Well-rounded Methods Methods that balance high accuracy, efficiency, and practicality (where “practical” means not significantly increasing the complexity of the application) include LS_COD, DLPF_C, PLS_NIP, and PLS_REC. While the transformationbased RR_VCS method has advantages in both accuracy and computational time, its model is challenging to apply in optimization and similar applications. This paper is available on arxiv under CC BY-NC-ND 4.0 Deed (Attribution-Noncommercial-Noderivs 4.0 International) license. This paper is available on arxiv under CC BY-NC-ND 4.0 Deed (Attribution-Noncommercial-Noderivs 4.0 International) license. available on arxiv

Why Some AI Power Flow Models Are Faster Than Others

It’s no secret we've stumbled upon a formidable obstacle to our AI-powered future: the staggering energy consumption of our current models. Industry leaders are scrambling to come up with short-term answers to ensure they don’t miss the wave, with ambitious efforts like Microsoft’s reopening of nuclear reactors on Three Mile Island and Google working on “first-of-its-kind” geothermal projects. And while that plays out at the big kid’s table, a flock of new startups are building on the progress made in recent years, re-thinking the fundamentals to see if there are solutions that could serve as the long-term solution. One that doesn’t require hundreds of millions of dollars in infrastructure investment. AI’s Power Demands Are a Ticking Time Bomb Operating large language models in their current iteration is an energy-intensive process that's rapidly approaching unsustainable levels. Training a single AI model can emit as much carbon as five cars over their entire lifetimes. It's not just an environmental concern; it's a scalability nightmare threatening to derail the AI revolution before it fully takes flight. Consider these sobering facts: GPT-3, with its 175 billion parameters, reportedly required 1,287 MWh for a single training run.
The carbon footprint of training a large NLP model is estimated to be around 626,000 pounds of CO2 equivalent.
As models grow, so does their energy consumption – often at a superlinear rate. And as the industry pushes for more advanced AI capabilities, this energy consumption is set to skyrocket. This is not only a problem on an operational level but also in the greater picture, as industry leaders like Google have pledged to achieve net-zero carbon emissions through the purchase of carbon credits from firms doing things like plugging orphaned oil and gas wells - a market where demand is already far out-pacing supply. Efficiency & Minimal Parameter Activation: The New North Star of AI Innovation The solution may be rather simple: smarter, smaller, more efficient models built for a set of specific purposes. Narrowing the scope, if you will. One such example is the open-source Aria model built by Rhymes, which employs minimal parameter activation. While the Aria model boasts a total of 25.3 billion parameters, it only activates a mere 3.9 billion for any given task. Traditional, mainstream models like GPT-3 activate all of their parameters for every task, regardless of complexity, while Aria's approach is like a surgeon using only the necessary instruments for a specific procedure. Many surgeons would tell you they don’t need to deploy the entire operating room's worth of equipment for every operation. Rhyme has deployed this in practice with BeaGo, which they call “a smarter, faster AI search.” Based on my tests, BeaGo’s results were indistinguishable from competing products from Perplexity and other, most energy and time-intensive products. But it’s more than simply narrowing the scope: the startup has built a multimodal open-source mixture-of-experts model that sorts and intelligently manages large, long-context data of all types, including text, video, and images. Rhymes’ solution may be lighting the pathway for AI in 2025 and beyond, all without the hundreds of millions of dollars in infrastructure spending. The Reminder: Innovation Never Stops In the end, the work of companies like Rhymes serves as a reminder that just because we found something that works, doesn’t mean the task of innovating is over. While Microsoft and Google run with our existing large language models, working to productize and bring AI to the mass-market, others cannot stop working on building something even better. I’m heartened by the startup-driven approach that I see here at the end of 2024 – combining multimodal capabilities, selective parameter activation, and open-source collaboration – offers a blueprint for how we can achieve a vision of an AI that both works and works for the planet. It’s no secret we've stumbled upon a formidable obstacle to our AI-powered future: the staggering energy consumption of our current models. Industry leaders are scrambling to come up with short-term answers to ensure they don’t miss the wave, with ambitious efforts like Microsoft’s reopening of nuclear reactors on Three Mile Island and Google working on “first-of-its-kind” geothermal projects . Microsoft’s reopening of nuclear reactors on Three Mile Island Microsoft’s reopening of nuclear reactors on Three Mile Island Google working on “first-of-its-kind” geothermal projects Google working on “first-of-its-kind” geothermal projects And while that plays out at the big kid’s table, a flock of new startups are building on the progress made in recent years, re-thinking the fundamentals to see if there are solutions that could serve as the long-term solution. One that doesn’t require hundreds of millions of dollars in infrastructure investment. doesn’t AI’s Power Demands Are a Ticking Time Bomb Operating large language models in their current iteration is an energy-intensive process that's rapidly approaching unsustainable levels. Training a single AI model can emit as much carbon as five cars over their entire lifetimes. It's not just an environmental concern; it's a scalability nightmare threatening to derail the AI revolution before it fully takes flight. Consider these sobering facts: GPT-3, with its 175 billion parameters, reportedly required 1,287 MWh for a single training run. The carbon footprint of training a large NLP model is estimated to be around 626,000 pounds of CO2 equivalent. As models grow, so does their energy consumption – often at a superlinear rate. GPT-3, with its 175 billion parameters, reportedly required 1,287 MWh for a single training run. The carbon footprint of training a large NLP model is estimated to be around 626,000 pounds of CO2 equivalent. As models grow, so does their energy consumption – often at a superlinear rate. And as the industry pushes for more advanced AI capabilities, this energy consumption is set to skyrocket. This is not only a problem on an operational level but also in the greater picture, as industry leaders like Google have pledged to achieve net-zero carbon emissions through the purchase of carbon credits from firms doing things like plugging orphaned oil and gas wells - a market where demand is already far out-pacing supply. plugging orphaned oil and gas wells Efficiency & Minimal Parameter Activation: The New North Star of AI Innovation The solution may be rather simple: smarter, smaller, more efficient models built for a set of specific purposes. Narrowing the scope, if you will. One such example is the open-source Aria model built by Rhymes , which employs minimal parameter activation. While the Aria model boasts a total of 25.3 billion parameters, it only activates a mere 3.9 billion for any given task. Traditional, mainstream models like GPT-3 activate all of their parameters for every task, regardless of complexity, while Aria's approach is like a surgeon using only the necessary instruments for a specific procedure. Many surgeons would tell you they don’t need to deploy the entire operating room's worth of equipment for every operation. Aria model built by Rhymes Rhyme has deployed this in practice with BeaGo, which they call “a smarter, faster AI search.” Based on my tests, BeaGo’s results were indistinguishable from competing products from Perplexity and other, most energy and time-intensive products. But it’s more than simply narrowing the scope: the startup has built a multimodal open-source mixture-of-experts model that sorts and intelligently manages large, long-context data of all types, including text, video, and images. Rhymes’ solution may be lighting the pathway for AI in 2025 and beyond, all without the hundreds of millions of dollars in infrastructure spending. The Reminder: Innovation Never Stops In the end, the work of companies like Rhymes serves as a reminder that just because we found something that works , doesn’t mean the task of innovating is over. While Microsoft and Google run with our existing large language models, working to productize and bring AI to the mass-market, others cannot stop working on building something even better. works I’m heartened by the startup-driven approach that I see here at the end of 2024 – combining multimodal capabilities, selective parameter activation, and open-source collaboration – offers a blueprint for how we can achieve a vision of an AI that both works and works for the planet.

The AI Energy Crisis & A Newfound Push for Efficiency

Die KI-Energiekrise und ein neuer Drang nach Effizienz

La crise énergétique de l'IA et une nouvelle dynamique en faveur de l'efficacité

משבר האנרגיה של AI ודחיפה חדשה ליעילות

La crisi energetica dell’intelligenza artificiale e una nuova spinta verso l’efficienza

De AI-energiecrisis en een nieuwe drang naar efficiëntie

A crise energética da IA e um novo impulso para a eficiência

Astounding Stories of Super-Science May 1931: Dark Moon - Chapter II

Stories, News & Mentions about SUPERLINKED

Categories

Trending Topics

Stories, News & Mentions about SUPERLINKED

Superlinked’s Stories on HackerNoon

Categories

Trending Topics