Presearch

With all of the buzz about ChatGPT and artificial intelligence (AI) impacting Big Tech search engines like Google, we’ve been asked by community members how Presearch intends to leverage AI.We’ve prepared this blog post to explore the opportunities and discuss how Presearch is using AI now and in the future.<img src="https://academy-public.coinmarketcap.com/srd-optimized-uploads/48e28c7b5ca54bf58404d02caa6a1ebc.png" /><h2 id="header-0">AI-powered search</h2>Search engines have been pioneers in the AI space, having been built on top of machine learning (ML) algorithms for years. Due to a combination of large amounts of data to search through, query input from users indicating their intent, and the potential for real-time feedback from ongoing user interactions (clicks, paging, scrolling, etc.), AI-powered search engines have an ability to constantly learn and improve from new data and interactions.AI is under the hood powering search and social media experiences for billions every day.It’s only been in recent months that public understanding of the power of AI has risen dramatically, as a result of exposure to AI responses through the ChatGPT interface launched by OpenAI. This has led to massive awareness of the impressive stage at which AI currently operates and the rapidly-evolving future possibilities with AI.The launch of chatGPT in November was an “iPhone moment” for AI technology, unveiling a compelling chat experience that feels so intelligent and human-like that public interest and rate of adoption has far exceeded all of the best technology launches of the past, reaching more than a million users in just five days.ChatGPT packages up several years of open innovation on Large Language Models across a diverse set of companies and researchers (alongside OpenAI), and is paving the way for a significant wave of innovation in artificial intelligence over the next few years.Presearch CTO, <a href="https://www.linkedin.com/in/treygrainger/" rel="noopener noreferrer nofollow">Trey Grainger</a>, is author of the book, <a href="https://aipoweredsearch.com/" rel="noopener noreferrer nofollow">AI-Powered Search</a>, which he’s been releasing chapter-by-chapter since 2019, about how to implement artificial intelligence within the context of search experiences. Trey spent years as SVP of Engineering and then Chief Algorithms Officer for an AI-powered search company, building out the search and AI capabilities powering hundreds of the top companies in the world. He’s also published numerous books, journal articles, and research publications at the intersection of Search and AI. Trey is the perfect person to lead the project into this new era of search.In this article we plan to give you a high level overview of how we’re currently using AI, and how AI will play a key role in Presearch’s decentralized search engine moving forward.Note: This article is a high-level overview — Presearch will release an updated roadmap and whitepaper going deeper into the project’s use of AI soon.<h2 id="header-1">What are LLMs and why does it matter for Presearch?</h2>Presearch is currently using OpenAI’s GPT-3 to power our AI search experience prototype that was first rolled out in closed beta in December 2022. OpenAI has been working on AI since its founding in 2015, but pivoted to focus on Large Language Models (LLMs) after the release of the famous <a href="https://arxiv.org/abs/1706.03762" rel="noopener noreferrer nofollow">“Attention is all you need”</a> article published by Google. The paper proposed a new architecture for language models called the Transformer, which uses attention mechanisms to identify patterns in language that span multiple words or sentences. Transformers are trained using deep learning on large datasets (most often text) to learn both the linguistic relationships and the conceptual relationships present in the datasets.Architecturally, Transformers usually consist of an encoder — which can map text (or other inputs like images or audio) into numerical context vectors — and a decoder, which can take those context vectors and map them back to text. You can think of the context vectors, or embeddings, as a numerical representation of the meaning of the input that is understood by the language model. So the text “she went outside to walk down the street” might be encoded to an embedding like [0.12, 9.51, 6.42, … 0.32], and that same embedding could then be decoded into similar text like “the girl walked outside and down the street”. If the text the model was trained on was multilingual, then text with the same meaning in a different language would also map to a very similar embedding because it carries the same meaning. Transformers can essentially learn the meaning of both words and phrases within different contexts, as well as the language constructs that determine how words and phrases are combined.This architecture has been able to outperform previous models on a variety of tasks, including machine translation, language modeling, and question answering.The Transformer architecture is based on the idea of self-attention, which allows the model to attend to different parts of the input simultaneously. This allows the model to capture long-term dependencies in the input, which is essential for understanding the complex relationships found in the language used across a potentially large context. The Transformer also uses a multi-head attention mechanism, which allows the model to attend to different parts of the input in parallel. This allows the model to capture more complex relationships between words.After introducing Transformers to the world, Google enhanced its search engine with its Transformer-based BERT language model starting in 2019, and many search engines have continued to improve and leverage these kinds of language models at the query interpretation and ranking layer. Since then, the Transformer architecture has enabled the development of larger language models (GPT-3, or open source models like Flan t5), which can capture more complex relationships between words and also generate compelling responses to questions and instructions by using the instructions as prompts and then predicting — word by word (or more accurately, token by token) — the correct response based upon the input context.This is the crux of what is behind ChatGPT, which enabled OpenAI to build a product that has captivated the world.<h2 id="header-2">Leveraging AI through a completely decentralized index</h2>It isn’t a secret that Presearch’s long term goal is to build a completely decentralized search engine. A core component of this is a decentralized index of the world’s information. While it’s not here yet, we are actively architecting it and believe this revolutionary technology will play a key role in the next iteration of the internet.AI technology will not only sit atop the index as the consumer experience, but will be used actively in partitioning and indexing the data, storing and replicating it appropriately, and routing queries to the appropriate index partitions. While Presearch is currently using OpenAI to power its AI experience, this will ultimately be replaced with open models that the best engineers and data scientists in the world can help build, train, and integrate as part of the Presearch community.OpenAI’s models have been a great value-add for Presearch, enabling us to get to market quickly with a very robust AI experience, but the Presearch network will ultimately be powered by our own AI technology, leveraging open source models integrated deeply into the decentralized index. Presearch is taking a fundamentally different approach to building AI technology, one that is decentralized and open, leveraging the power of distributed training to power its AI technology.Distributed training is a method of training AI models on multiple machines simultaneously. This will allow for faster training times across a decentralized architecture, leading to more accurate models that can better understand the context around user queries and ultimately provide more relevant search results.Presearch is also looking to leverage a globally-decentralized index, with embeddings from these trained AI models deeply integrated. This will allow for faster and more accurate searches by providing a deeper understanding of user queries and then limiting the number of nodes needed to perform any given search. By leveraging a globally-decentralized index partitioned based upon AI-based embeddings, Presearch can optimize the speed of queries and maximize network resources, while simultaneously providing the most relevant results.The combination of distributed training and a globally-decentralized index will enable Presearch to deliver a truly decentralized search engine powered by the best open-source and community-built AI models.Large Language Models, while very popular right now, are far from the only AI capabilities common in search engines. Presearch will leverage the distributed training for machine-learned ranking (also known as learning to rank), signals boosting models, knowledge graph generation, and many other techniques to learn and generate improved ranking factors. Large Language Models will play a critical role, however, in both interpreting and summarizing results in a thoughtful and useful way.Presearch is taking a fundamentally different approach than OpenAI and other search engines by focusing on open collaboration and decentralization. What Bitcoin did for digital currency, Presearch will do for access to the world’s information. If you’d like to get involved with the project, run a decentralized node, help us build, or just learn more, we’d love to welcome you as part of our community!PresearchAI TwitterWe’ve created a dedicated account to share our progress and insights on artificial intelligence. Follow along here — <a href="https://twitter.com/PresearchAI" rel="noopener noreferrer nofollow">https://twitter.com/PresearchAI</a><h3 id="if-you-have-questions-please-join-one-of-our-community-links-below">If you have questions, please join one of our Community links below</h3>Run a node:<a href="https://nodes.presearch.com/" rel="noopener noreferrer nofollow"> https://nodes.presearch.com</a> Keyword Stake: <a href="https://keywords.presearch.com/" rel="noopener noreferrer nofollow">https://keywords.presearch.com</a> Info site: <a href="https://presearch.io/" rel="noopener noreferrer nofollow">https://presearch.io</a> Search with Presearch: <a href="https://presearch.com/" rel="noopener noreferrer nofollow">https://presearch.com</a><h3 id="community-channels">Community channels:</h3>Telegram: t.me/Presearch Twitter: <a href="https://twitter.com/presearchnews" rel="noopener noreferrer nofollow">https://twitter.com/presearchnews</a> PresearchAI Twitter: <a href="https://twitter.com/PresearchAI" rel="noopener noreferrer nofollow">https://twitter.com/PresearchAI</a> YouTube: <a href="https://www.youtube.com/c/Presearch/videos" rel="noopener noreferrer nofollow">https://www.youtube.com/c/Presearch/videos</a> Rumble: <a href="https://rumble.presearch.io/" rel="noopener noreferrer nofollow">https://rumble.presearch.io</a> Discord: <a href="https://discord.presearch.io/" rel="noopener noreferrer nofollow">https://discord.presearch.io</a> Minds: <a href="https://www.minds.com/presearch/" rel="noopener noreferrer nofollow">https://www.minds.com/presearch</a> Reddit: <a href="https://www.reddit.com/r/Presearch/" rel="noopener noreferrer nofollow">https://www.reddit.com/r/Presearch/</a> Twitter (Community): <a href="https://twitter.com/TeamPresearch" rel="noopener noreferrer nofollow">https://twitter.com/TeamPresearch</a> Telegram (News): t.me/PresearchNews Facebook: <a href="https://www.facebook.com/presearch.io" rel="noopener noreferrer nofollow">https://www.facebook.com/presearch.io</a> Email: support&#64;presearch.ioInterested in partnering? Send us an email at — promote&#64;presearch.io

Presearch AI

Table of Contents

Presearch

Other articles published on Feb 22, 2023

Saudi Aramco Signs Agreement With droppGroup to Build Web3 Tech

Lasagna-Another Way for Users to Create and Earn in SecondLive Metaverse

Hong Kong To Allow Retail Trading

Ankr and Microsoft Partner To Offer Enterprise Node Services

The problem with mixing currency and data

WEMIX.Fi Lend & Borrow New Asset Adding Event

Join our free newsletter for daily crypto updates!