<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Jubin Jose on Medium]]></title>
        <description><![CDATA[Stories by Jubin Jose on Medium]]></description>
        <link>https://medium.com/@josejubin?source=rss-6a042be9b7d9------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*p20uQBb0O61WDY4t.</url>
            <title>Stories by Jubin Jose on Medium</title>
            <link>https://medium.com/@josejubin?source=rss-6a042be9b7d9------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 00:01:19 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@josejubin/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Open Domain Question Answering Part-1 [BlenderBot 2.0]]]></title>
            <link>https://medium.com/aquila-network/open-domain-question-answering-part-1-blenderbot-2-0-bf5714efe059?source=rss-6a042be9b7d9------2</link>
            <guid isPermaLink="false">https://medium.com/p/bf5714efe059</guid>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[blenderbot]]></category>
            <category><![CDATA[parlai]]></category>
            <category><![CDATA[question-answering]]></category>
            <dc:creator><![CDATA[Jubin Jose]]></dc:creator>
            <pubDate>Fri, 31 Dec 2021 11:05:20 GMT</pubDate>
            <atom:updated>2022-01-14T07:01:52.109Z</atom:updated>
            <content:encoded><![CDATA[<h4>In this ODQA tutorial, we use BlenderBot 2.0. We discuss basic idea behind it, local setup, useful tweaks, and the pros and cons of designing an open domain QA chatbot.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KggrG7KMCvRZfHdQZV8l4A.jpeg" /><figcaption>BlenderBot 1.0 vs 2.0</figcaption></figure><p>From an engineering standpoint, in my opinion, keeping an end-to-end language model (like GPT3) up to date with the latest knowledge is very inefficient. That’s why we need a modular design based on the cost of change. For an open domain chatbot, the strategy is to separate knowledge from large models (costly to train) to external databases (cheap to update). It will also make AI affordable to anyone by reusing pre-trained, large language models for generating text with all knowledge kept in their affordable infrastructure.</p><p>BlenderBot2 roughly separates its operations into three modules — Knowledgebase (search module), long-term memory, and text generator. This design proposed by BlenderBot2 can be a general architecture to any system that generates (hallucinates) content based on real-world knowledge.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/703/1*FXIX7iFI__1qJC2Rjxn6jQ.png" /><figcaption>BlenderBot2 Architecture level 0</figcaption></figure><p><strong>KnowledgeBase</strong></p><p>A knowledgeBase can be anything that we’re using. It can be an SQL database, a search engine, knowledge graph, a bucket of documents, etc. To make our knowledgeBase compatible with other parts of our system, we need to sandwich it within a query generator and a response encoder. Query generator takes in a state/context as input and generates appropriate query. Once the knowledgeBase generates an appropriate response to the given query, an encoder will convert it into a compressed format (ex. latent vector) for the next module to use.</p><p>BlenderBot2 uses a full-text search engine (like Bing.com) as the knowledgeBase as far as we’re concerned.</p><p><strong>Long-term memory</strong></p><p>Long-term memory is fast in-memory vector storage that supports searching through conversation history for previously collected knowledge (from either the knowledgeBase or the user).</p><p>This can be done in one of two ways:</p><ul><li>Summarize the entire conversation into a single vector for a specific conversation session. Keep this updated across turns.</li><li>Summarize the entire conversation into a small paragraph for a specific conversation session. Keep this updated across turns. BlenderBot2 uses this method.</li><li>Convert each turn of a conversation into corresponding latent vectors to be k-NN searched later.</li></ul><p><strong>Generator</strong></p><p>Any general-purpose seq-to-seq NLP model can function as this module. We typically use large pre-trained end-to-end models such as GPT (WebGPT), T5, etc. These models are trained on massive datasets like CommonCrawl in an unsupervised fashion. Later on, these models get fine-tuned on conversational datasets.</p><p>In BlenderBot2, the generator takes in both text from search results and conversation summary to generate responses to the user. Optionally, there is a classifier (switcher) in between the generator and the other two modules to determine which input to use; we’re not interested in that for now.</p><p><strong>Let’s get started</strong></p><p><strong>Step 1:</strong> clone this repository: <a href="https://github.com/freakeinstein/ParlAI_SearchEngine">https://github.com/freakeinstein/ParlAI_SearchEngine</a></p><pre>git clone <a href="https://github.com/freakeinstein/ParlAI_SearchEngine.git">https://github.com/freakeinstein/ParlAI_SearchEngine.git</a></pre><p><strong>Step 2:</strong> create virtual environment &amp; activate:</p><pre>python3 -m venv env<br>source env/bin/activate</pre><p><strong>Step 3:</strong> install prerequisites:</p><pre>pip install -r requirements.txt</pre><p><strong>Step 4:</strong> run search server:</p><p>By default, Google search is used, and is relatively slow. See “<strong>Useful tweaks</strong>” section below for more search engine options.</p><pre>python search_server.py serve --host 0.0.0.0:8080</pre><p><strong>Step 5:</strong> run parlAI BlenderBot2 in a separate terminal (don’t forget to activate environment):</p><ul><li>To use 3B parameters model (your laptop wouldn’t run it):</li></ul><pre>python -m parlai interactive --model-file zoo:blenderbot2/blenderbot2_3B/model --search_server 0.0.0.0:8080</pre><ul><li>To use 400M parameters model (a high-end laptop will run it):</li></ul><pre>python -m parlai interactive --model-file zoo:blenderbot2/blenderbot2_400M/model --search_server 0.0.0.0:8080</pre><p>That’s it, you can start chatting with BlenderBot2 in your terminal.</p><p><strong>Useful tweaks</strong></p><p><strong><em>parlAI command</em></strong></p><p>Here are some great resources on command-line arguments to be used to tweak BlenderBot2’s behaviors: <a href="https://github.com/facebookresearch/ParlAI/tree/main/projects/blenderbot2/agents">https://github.com/facebookresearch/ParlAI/tree/main/projects/blenderbot2/agents</a></p><p><strong><em>Search Server</em></strong></p><ul><li>Search server uses “<strong>Google</strong>” as its default engine. It is not recommended for extended usage.</li><li>If you’re interested in reproducing results from the original paper (from FacebookResearch), use “<strong>Bing</strong>” instead.</li><li>If you’re trying to incorporate Blenderbot into your applications, use “<strong>Aquila</strong>” custom search engine. Which will restrict BlenderBot’s knowledge to custom web pages that you bookmark.</li></ul><p>For example, talk to a <a href="https://x.aquila.network/?share=CJJ9ZGQEcK1Jffs6Ji2cTpVW4oiYP6X3VCP9YH4KhC">Bitcoiner’s bookmark</a> maintained at Aquila Network:</p><pre>python search_server.py serve --host 0.0.0.0:8080  --search_engine=&quot;Aquila&quot; --subscription_key &quot;CJJ9ZGQEcK1Jffs6Ji2cTpVW4oiYP6X3VCP9YH4KhC&quot;</pre><p><strong>Analysis &amp; final thoughts</strong></p><ul><li>BlenderBot2 is still at research quality. To use it in production, you might need to consider finetuning it to your custom training data since the default responses are yet to beat our basic standards. You should also consider optimizing your reference documents (a custom search engine is recommended) to get better results.</li><li>BlenderBot2 is less <a href="https://www.unite.ai/preventing-hallucination-in-gpt-3-and-other-complex-language-models/">hallucinating</a> than its previous versions. It is also better than GPT-3 according to FacebookResearch in producing factual responses. OpenAI published WebGPT as a response to this and we haven’t tried it out by ourselves (maybe someday).</li><li>If you’re building a product that’s fact-critical, we don’t recommend using BlenderBot2 (and we recommend you to keep an eye out for Blenderbot3). In this case, we recommend you another more flexible method (link coming soon).</li></ul><p><strong>Digging down</strong></p><p>wanna go down the rabbit hole? Dig here..</p><ul><li>BlenderBot2 architecture:</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kfWr-aYkT5Dsax1Hu4cIHg.jpeg" /><figcaption>BlenderBot2 Architecture level 1</figcaption></figure><ul><li>project website: <a href="https://parl.ai/projects/blenderbot2/">https://parl.ai/projects/blenderbot2/</a></li><li>Search Engine augmentation: <a href="https://parl.ai/projects/sea/">https://parl.ai/projects/sea/</a></li><li>Long-term memory for multi-turn conversations: <a href="https://parl.ai/projects/msc/">https://parl.ai/projects/msc/</a></li><li>Official article: <a href="https://ai.facebook.com/blog/blender-bot-2-an-open-source-chatbot-that-builds-long-term-memory-and-searches-the-internet/">https://ai.facebook.com/blog/blender-bot-2-an-open-source-chatbot-that-builds-long-term-memory-and-searches-the-internet/</a></li><li>Source code: <a href="https://github.com/facebookresearch/ParlAI/tree/main/projects/blenderbot2">https://github.com/facebookresearch/ParlAI/tree/main/projects/blenderbot2</a></li><li>HuggingFace Transformers: <a href="https://github.com/huggingface/transformers/issues/12807">https://github.com/huggingface/transformers/issues/12807</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bf5714efe059" width="1" height="1" alt=""><hr><p><a href="https://medium.com/aquila-network/open-domain-question-answering-part-1-blenderbot-2-0-bf5714efe059">Open Domain Question Answering Part-1 [BlenderBot 2.0]</a> was originally published in <a href="https://medium.com/aquila-network">Aquila Network</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Finding the best affordable NLP model for semantic search]]></title>
            <link>https://medium.com/aquila-network/finding-the-best-affordable-nlp-model-for-semantic-search-f53652f4b621?source=rss-6a042be9b7d9------2</link>
            <guid isPermaLink="false">https://medium.com/p/f53652f4b621</guid>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[neural-search]]></category>
            <category><![CDATA[transformers]]></category>
            <category><![CDATA[benchmark]]></category>
            <category><![CDATA[semantic-search]]></category>
            <dc:creator><![CDATA[Jubin Jose]]></dc:creator>
            <pubDate>Wed, 29 Dec 2021 10:29:49 GMT</pubDate>
            <atom:updated>2021-12-29T10:29:49.970Z</atom:updated>
            <content:encoded><![CDATA[<h4>We have benchmarked popular latent encoder models for k-NN text search. Here are the results.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/812/1*rHaEtqw6NA56tagin2vA6g.png" /><figcaption>Aquila X Matthew Ball bookmarks screenshot</figcaption></figure><p>With today’s advancements in NLP architectures, now we can build consumer-quality semantic search engines that are very good at matching sentences by their meaning. At <a href="https://aquila.network/">aquila network</a>, we’ve decided to benchmark popular pre-trained models available out there for free. And here are the results:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8A5EhZMUBuzJ2-VQUz6vxA.jpeg" /><figcaption>NLP sentence encoder models for semantic search (accuracy benchmark)</figcaption></figure><h3>Interpretations of data above:</h3><ul><li>We’re measuring the retrieval accuracy of a document (Wikipedia article) for a given query. We’ve used <a href="https://rajpurkar.github.io/SQuAD-explorer/">the SQuAD</a> dataset for convenience.</li><li>Cells marked green yield better accuracy by comparison. We’re measuring top 1 | top 3 | top 5 retrieval accuracies separated by pipes.</li><li>Normalization has no effect for Fasttext when ANNOY angular/euclidean metric is used (Annoy internally normalizes vectors to then run euclidean metric when the angular metric is chosen)</li><li>ANNOY no. of trees (100) increase increases accuracy, but slow search, 500 is max. affordable by speed</li><li>Manhattan metric produces the same results as euclidean and is insignificantly slightly different (+-1%) result from angular — means, vectors are clustered around an approximately unit noisy sphere</li><li>Dot (vectors must be normalized) is slightly better than all the other metrics for Fasttext.</li><li>Cleaned text is making it worse for transformer models.</li><li>Refer our post ranking algorithm <a href="https://github.com/Aquila-Network/AquilaX-CE/blob/9aa6b076e83892ececaa23970609ae7045aa6699/src/index.py#L86">here</a>.</li></ul><h3>Summary:</h3><p>Our experiments show that <a href="https://www.sbert.net/docs/pretrained_models.html#msmarco-passage-models">MSMarco</a> models show better accuracy, utilizing affordable resources that a personal computer can offer. With this, we’re upgrading our default model from Fasttext to MSMarco in <a href="https://github.com/Aquila-Network/AquilaHub/releases/tag/v0.1.0">Aquila Hub v0.1.0</a> release. Feel free to check it out and send your feedback.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f53652f4b621" width="1" height="1" alt=""><hr><p><a href="https://medium.com/aquila-network/finding-the-best-affordable-nlp-model-for-semantic-search-f53652f4b621">Finding the best affordable NLP model for semantic search</a> was originally published in <a href="https://medium.com/aquila-network">Aquila Network</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Aquila Network 1.0.0 (Beta) Released ]]></title>
            <link>https://medium.com/aquila-network/aquila-network-1-0-0-beta-released-7ef9afb451b0?source=rss-6a042be9b7d9------2</link>
            <guid isPermaLink="false">https://medium.com/p/7ef9afb451b0</guid>
            <category><![CDATA[semantic-search]]></category>
            <category><![CDATA[bookmarking-sites]]></category>
            <category><![CDATA[bookmarks]]></category>
            <category><![CDATA[vector-search-engine]]></category>
            <category><![CDATA[fake-news]]></category>
            <dc:creator><![CDATA[Jubin Jose]]></dc:creator>
            <pubDate>Tue, 07 Dec 2021 13:31:08 GMT</pubDate>
            <atom:updated>2021-12-07T15:45:25.304Z</atom:updated>
            <content:encoded><![CDATA[<p>The first batteries included version of Aquila Network — a semantic text search engine. Available as Open Source as well as managed service.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3trQ1LT0nqLWpPqv2GRlaQ.png" /><figcaption>Aquila Network 1.0.0</figcaption></figure><p>Quick Access (v1.0.0): <a href="https://docs.aquila.network/">documentation</a> , <a href="https://x.aquila.network/releases.html">changelog</a></p><p>It’s exciting to come out of Alpha. We’ve made a lot of improvements and this is just a warm-up.</p><p>If you would like to quickly try it out, just go to <a href="https://aquila.network/">https://aquila.network</a>. To better experience all features of Aquila Network, you need to install a browser extension. In general, people are hesitant to install a browser extension, we understand that. Don’t worry, we don’t need your personal data. The good news is, you don’t have to trust us, just verify: Everything is <a href="https://github.com/Aquila-Network/awesome-aquila-network">Open Source</a>, including <a href="https://github.com/Aquila-Network/AquilaX-browser-extension">browser extensions</a>.</p><p>If you are a tech-savvy person, we recommend trying it on your device. If you’re on a <strong>Debian (Ubuntu) machine with docker and docker-compose installed already</strong>, here’s the single line command to set it up:</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/f2097b6558008c7cdea3e44f3279f8a0/href">https://medium.com/media/f2097b6558008c7cdea3e44f3279f8a0/href</a></iframe><figure><img alt="" src="https://cdn-images-1.medium.com/max/847/1*_5QQnu5eqKQJAR6X2-qtDw.png" /><figcaption>Aquila X search interface</figcaption></figure><p>When it comes to our managed service, we could offer some additional features as more and more people show up.</p><p><strong>Public sharing:</strong></p><p>Choose to share your searchable bookmarks with anybody like a content curator.</p><p><strong>Explore bookmarks:</strong></p><p>Discover hand-picked bookmarks from others.</p><p>We’re also working on a feature to allow “following” people you like.</p><p>Here’s what I haven’t told you yet. Aquila Network was originally designed to solve a bigger problem — information drainage on the Web. I will add a couple of references at the end of this article to paint the bigger picture in your mind.</p><p>Yes, it’s so cute.., but it wouldn’t fly overnight. For the bigger dreams, smaller steps need to be taken first to <a href="https://www.youtube.com/watch?v=I_izvAbhExY">stay alive</a>, especially if you have a small team of 3 people. That’s why we’re narrowing our focus to the smallest problem that we think can be solved elegantly — getting more out of your forgotten bookmarks. That’s why we need your feedback..</p><p>Have a great day!</p><p>References:</p><ul><li><a href="https://medium.com/@josejubin/the-brand-new-struggles-of-a-millennial-who-grew-up-with-the-open-web-the-story-of-aquila-network-4f85dbb8b332">The brand-new struggles of a millennial who grew up with the Open Web — the story of Aquila Network</a></li><li><a href="https://news.ycombinator.com/item?id=29392702">Ask HN: Has Google search become quantitatively worse? | Hacker News</a></li><li><a href="https://medium.com/aquila-network/we-are-building-a-web-native-search-engine-that-you-control-introducing-aquila-network-aed145156d46">We are building a web native search engine that “you” control. Introducing Aquila Network.</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7ef9afb451b0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/aquila-network/aquila-network-1-0-0-beta-released-7ef9afb451b0">Aquila Network 1.0.0 (Beta) Released 🥳</a> was originally published in <a href="https://medium.com/aquila-network">Aquila Network</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The brand-new struggles of a millennial who grew up with the Open Web — the story of Aquila Network]]></title>
            <link>https://medium.com/@josejubin/the-brand-new-struggles-of-a-millennial-who-grew-up-with-the-open-web-the-story-of-aquila-network-4f85dbb8b332?source=rss-6a042be9b7d9------2</link>
            <guid isPermaLink="false">https://medium.com/p/4f85dbb8b332</guid>
            <category><![CDATA[seo]]></category>
            <category><![CDATA[information]]></category>
            <category><![CDATA[fake-news]]></category>
            <category><![CDATA[web]]></category>
            <category><![CDATA[censorship]]></category>
            <dc:creator><![CDATA[Jubin Jose]]></dc:creator>
            <pubDate>Fri, 03 Dec 2021 16:32:26 GMT</pubDate>
            <atom:updated>2021-12-03T16:44:09.414Z</atom:updated>
            <content:encoded><![CDATA[<h3>The brand-new struggles of a millennial who grew up with the Open Web — the story of Aquila Network</h3><h4>A short story developed around recent discussions on HN <a href="https://news.ycombinator.com/item?id=29392702">https://news.ycombinator.com/item?id=29392702</a></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*da1cC9rmcbNT1gC2-wtsWQ.jpeg" /><figcaption>glitch in the matrix</figcaption></figure><h3>In the previous decades, the Web offered an unbiased source of Truth to anyone without judging who you are, what your beliefs are. It was a utopian escape hatch from the noisy physical world. Today, we’re fiercely fighting the entropy that is tearing apart our ideals once again in cyberspace.</h3><p>Ava identifies herself as a polymath. Her interests range across science, technology, philosophy, and art. A millennial girl by birth, she’s determined not to settle but to explore the realm of knowledge and wisdom. She has been a curious student of the Web for more than a decade. Anybody, without a doubt, will agree to this by looking at her bookmarks and knowledge repositories. Just like any other millennial, Ava has experienced firsthand how the Web grew over time. Even though she’s happy with the technology toolset at her reserve, she started feeling small but growing frustration in discovering the right information.</p><p>Ava is now in her thirties. She enjoys hanging out with peers of similar interest whom she met online over the years. Even during the pandemic, thanks to the Web, she meets her friends online.</p><p>“Web is full of wonders…” She would say occasionally.</p><p>One day she happened to meet El Nino from her friend circle. Nino is a web developer who occasionally writes articles reviewing different life hack tools. It didn’t take much time for them to start talking about the “eventual information drain problem” of the open Web. Nino called it “Information Bubble”, which he thinks will continue growing as days go by. Ava recognized it immediately, she had no clue that it was a real problem. Now she knows the real reason behind the growing fake information and biased opinion crisis.</p><p>“Is there a solution to this?” she asked.</p><p>“There should be multiple solutions to this at different layers. I’m interested in solving one among them” Nino replied. According to Nino, the information discovery layer of the Web should be decentralized.</p><p>“You know, simply censoring a public library’s book index could trigger a wave of damages to the society. Similarly, centralized gatekeeping of Web indexes triggers secondary effects which consciously starve and eventually kill independent content creators” Nino paused for a moment. He’s a victim of side effects caused by aggressive SEO optimizations and click-baits.</p><p>The good news is, the Web already has a native, decentralized (but limited) content discovery mechanism — a super index — DNS. It has already proven the power and freedom a decentralized discovery layer could offer. With great confidence and hope in his words, he said “I propose a decentralized, native search layer, which will help discover context-specific content. What’s more, anyone can host it anywhere. That will be awesome right?”</p><p>“Yeah, super cool! What can I do?” Ava asked with excitement.</p><p>Nino continued.</p><p>First of all, you can take a look at the <a href="https://github.com/Aquila-Network/whitepaper">technical paper</a> and <a href="https://github.com/Aquila-Network/specs">specifications</a> to make sure that the technology convinces you. Just remember, those are still under active improvement. Any suggestion you have on improving them is heartily welcome. Also, there are <a href="https://github.com/Aquila-Network/awesome-aquila-network#official-implementations">reference implementations</a> in progress. You can test them and make pull requests if you think of a necessary change. You could also <a href="https://discord.gg/5YP7zHS">keep in touch with our team</a>. Our team will let you know of the latest updates so that you can eventually join the collective efforts.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4f85dbb8b332" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[We are building a web native search engine that “you” control. Introducing Aquila Network.]]></title>
            <link>https://medium.com/aquila-network/we-are-building-a-web-native-search-engine-that-you-control-introducing-aquila-network-aed145156d46?source=rss-6a042be9b7d9------2</link>
            <guid isPermaLink="false">https://medium.com/p/aed145156d46</guid>
            <category><![CDATA[search-engines]]></category>
            <category><![CDATA[decentralization]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[search]]></category>
            <category><![CDATA[privacy]]></category>
            <dc:creator><![CDATA[Jubin Jose]]></dc:creator>
            <pubDate>Sun, 28 Mar 2021 18:42:21 GMT</pubDate>
            <atom:updated>2021-03-29T09:22:28.596Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*U54c4ZFFkKAExipM5QZO5A.jpeg" /><figcaption>Image by <a href="https://pixabay.com/users/robert_c-1017101/?utm_source=link-attribution&amp;utm_medium=referral&amp;utm_campaign=image&amp;utm_content=2045655">Robert C</a> from <a href="https://pixabay.com/?utm_source=link-attribution&amp;utm_medium=referral&amp;utm_campaign=image&amp;utm_content=2045655">Pixabay</a></figcaption></figure><h3>Search engines are the second biggest gatekeepers on the Internet after DNS. It’s a surprise that we still don’t have a native search feature built into the fabric of the Web.</h3><p><strong><em>So, we decided to do it.</em></strong></p><p>The internet is believed to be the stepping stone towards the third industrial revolution. With the Web, we have liberated information and thereby knowledge. Even from the early days, discovering the right information across the web was a pain point to be solved.</p><p>DNS was one such successful discovery mechanism (layer one) that we have adopted. It abstracts the IP address of a content server with meaningful words that are easy to remember. As the web grew exponentially, search engines added another layer of abstractions over DNS to discover relevant sections of the content being served.</p><p>Today, we are highly reliant on these search engines to navigate through the ever exploding complex web of knowledge. The big players in the space know this fact and some of them are abusing this power to a greater extent beyond their logical boundaries, that almost broke the free expression of the Web itself.</p><h3>The problem</h3><p>We know that almost everything on the Web is sustained by direct or indirect advertisement. Each click you make nudges you closer to a product that you would buy in the near future. We haven’t seen an alternative economic model that’s capable of coming in par with this yet.</p><p><em>After 2009, some visionaries (including us) claimed the dawn of a new model driven by digital tokenomics — which anyway should prove its antifragility in the coming future to be useful.</em></p><p>So as we know of today, the Web needed the advertisement industry to sustain and provide remarkable services. However, the top commercial search engines (which rely on original content creators) took this model and built monopoly blackholes that are so powerful to attract most of the advertisement revenues to themselves. This allowed them to carefully control the revenue pipelines in such a way to forcefully shape the content ecosystem by starving the original content creators to abide by the rules they define (SEO).</p><p><strong>The actual problem</strong></p><p>It doesn’t matter whatever business models (big or small) that anyone adopts as long as it doesn’t do much harm to the ecosystem. The actual problem here is that the ground is not level. The fruitful public content on the web is disappearing, misinformation, fake news and clickbaits are growing exponentially, many of the genuine websites are adopting paywalls and becoming only fans; a remarkable percentage of content is being served app only.</p><p>So, what’s the actual problem? Yes, the open Web is shrinking, it’s going backwards, becoming less discoverable. And who is responsible? The search engines and their control that extends beyond the Web’s (inc. W3) protocols itself.</p><h3>The links are disappearing, Page-Rank is dead and the open Web is draining. Today’s centralized search engines are cutting the branch they’re sitting on.</h3><h3>The solution</h3><p>The solution is obviously simple. Sadly, it’s being ignored through conscious efforts.</p><ul><li>Level the technology gap in building search engines for everybody</li><li>Define a protocol to seamlessly replicate search indexes across search engines</li><li>Implement an incentive structure that reward both the content creators and technology providers equally</li></ul><p>Search engine technology is not rocket science. With today’s technology, building a search engine is relatively easy. The existing complexity in scalability is merely a side effect of maintaining a stubborn central system that does want to keep everything to itself. Multiple centralized search engines doing the same practice of keeping redundant, non compatible search indexes at the expense of energy and space, only to crutch the Web is just insane when we think about it.</p><p>A few years back, two of the most popular mobile operating systems introduced different features to extend their search capabilities into applications. It seems the majority of developers are reluctant to fully open up their search indexes. Why? Because, as long as the blackhole business model stays the same, the app traffic would decay by only making it up to the host’s virtual assistants. Most app developers rely on app traffic and cross selling as an income source.</p><h3>Progress is when we have information discoverability including closed platforms. The incentives should be structured in a way to encourage the creators to do so.</h3><h3>Technology landscape</h3><p>We now have the right technology advancements to achieve all three proposals we have made in the solution section.</p><ul><li>Fixing the gap through universal data structures: <strong>Deep learning</strong> models are the best information compressors that we have so far. We’re now able to map any type of information (text, images, video, audio, graph) into tensors (vectors) and perform search over it.</li><li>Every search node is compatible with one another and indexes are transferable: We have designed a replication protocol (inspired by couchDB and IPFS) that every node should follow. This protocol ensures <strong>eventual consistency</strong> across search nodes and eliminates unwanted redundancy through <strong>content addressing</strong>.</li><li>The protocol is open ended to allow experimentation of any economic models by the users: We have made the protocol simple and open to experimentation. This we hope will allow the users to try out both old and new economic models on top of it as a separate layer. To encourage experimentation on decentralized tokenomics models, we’ve added native support for state of the art public key cryptography. It is architectured in such a way that the end users have full control over the keys. Each content a user (usually a wallet) put out in the network is signed (ownership and authentication) and optionally can be encrypted (privacy) between authorized nodes.</li></ul><h3>About us</h3><p>We’re a tiny team of millennial engineers who grew with the Internet. We have experienced firsthand how the dynamics of the Web changed over the years. Now frustrated by the information drainage problem the Web is facing, we wanted to do something. We have been brainstorming for the past 2+ years, trying out different architectures to come up with a solution to this problem. Finally, inspired by the Simplicity of Bitcoin whitepaper itself, we’ve realized that many big problems can be solved through “<strong>locally centralized, globally decentralized</strong>” architecture. That’s how Aquila Network is designed.</p><p>We’re solving this problem for ourselves and others like us. If you want to support this effort, contribute a tiny fraction of your brain power. Everything is open sourced, try it out and let us know your feedback. Check out <a href="https://aquila.network/">https://aquila.network</a> for more information. If you are a non technical person, we could manage everything for you at an affordable price.</p><p>Cheers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*RR4tzltSguAk8c3FO6PX8A.gif" /></figure><blockquote>Aquila Network matured in <a href="https://github.com/a-mma">a-mma</a>’s technology incubation as AquilaDB</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=aed145156d46" width="1" height="1" alt=""><hr><p><a href="https://medium.com/aquila-network/we-are-building-a-web-native-search-engine-that-you-control-introducing-aquila-network-aed145156d46">We are building a web native search engine that “you” control. Introducing Aquila Network.</a> was originally published in <a href="https://medium.com/aquila-network">Aquila Network</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[This is how Google 2.0 might look like; Anyone can be part of  it, without competition.]]></title>
            <link>https://medium.com/a-mma/this-is-how-google-2-0-might-look-like-anyone-can-be-part-of-it-without-competition-dda41099f049?source=rss-6a042be9b7d9------2</link>
            <guid isPermaLink="false">https://medium.com/p/dda41099f049</guid>
            <category><![CDATA[token-economy]]></category>
            <category><![CDATA[decentralization]]></category>
            <category><![CDATA[ipfs]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-marketplace]]></category>
            <dc:creator><![CDATA[Jubin Jose]]></dc:creator>
            <pubDate>Sat, 27 Jun 2020 08:42:32 GMT</pubDate>
            <atom:updated>2021-04-11T08:06:54.543Z</atom:updated>
            <content:encoded><![CDATA[<h3>This is how Google 2.0 might look like; Anyone can be a part of it, without competition.</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ABUhV1opyAf2alYwyNYkTA.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@rajeshwerbatchu7?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Rajeshwar Bachu</a> on <a href="https://medium.com/s/photos/google?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><blockquote>“Some people say, “Give the customers what they want.” But that’s not my approach. Our job is to figure out what they’re going to want before they do. I think Henry Ford once said, “If I’d asked customers what they wanted, they would have told me, ‘A faster horse!’” People don’t know what they want until you show it to them. That’s why I never rely on market research. Our task is to read things that are not yet on the page.” — Steve Jobs</blockquote><p>One of the giants in digital advertising — Google (now Alphabet) is well known for its search service. They have started with a mission to organize the world’s information the best way possible. The web search engine was one of the initial offerings they have made and is still the successful one. For the past 20 years, they have played a major role in shaping the Internet. One such change I admire is continuous, sometimes monopoly efforts to transform unstructured web to some level of semi-structured form. The adoption into ‘Knowledge Graph’ and the introduction of schema.org was one of the smartest moves in this. These allowed them for near real-time, hassle-free information extraction from dynamic websites like e-commerce or live events. In addition to this, indirect forcing towards well-defined SEO rules, making influences in web development tools (Angular, Flutter) and practices, defining order through client-side monopoly (Chrome, Android) are some other genius steps they took. Even though these measures brought some level of structure into the Internet, something went wrong. Individuals and organizations started asking the questions — “Who owns my data? What do I get in return if it’s not me?”. People and organizations have started building walls. Sadly, the Internet now is getting fragmented, as it gets structured.</p><h3>Competition versus Monopoly</h3><p>Competition is good for consumers as it keeps a balance as we call a fair price for products and services. However, when it comes to standardization of new and sophisticated technology, monopoly plays a crucial role. Standardization brings innovation. It spreads knowledge, brings in a level field in engineering, and a fast-growing ecosystem around it. Also, when it comes to quality, a monopoly can deliver off the roof results.</p><p>When it comes to Google, everyone agrees that they are a monopoly in the search industry, even though the company is not positioned in a way to avoid it. They are delivering the best search service in the industry for free*. They are one of the entities that push AI innovation in aggression. They disrupted the telecommunication industry, powering the people in the lowest sectors in society. With YouTube, the whole media and publication industry got massively diversified. So, “Google is <strong>not</strong> evil” in contrary to many voices stay out there. But I would say, it is time to retire/rethink some, painfully one of the most profitable practices for the greater good of the Internet and technology. The sky is not a limit for Google, so why not?</p><h3>Public, Closed and Private data</h3><p>Classification of data into Public, Closed, and Private — although this is not a genetic approach, I have introduced this to convey a couple of points conveniently. Web as a network of linked documents is a public library to keep human knowledge and communications. We human race, all agree that our progress has always driven by knowledge sharing. This knowledge gets aggregated generations after generations. Nobody has to reinvent the wheel every time. Finding the right information with minimal effort boosts this pace of development in large multiples. As we know, most of this information is publicly accessible to anyone on the internet. It’s at your fingertips if you know exactly where to look for it. This public data resides in an unstructured format, which makes it difficult to — first of all, find the right document and then the right data, recorded in these documents.</p><p>Google’s intelligent search mechanism powered by a well-organized knowledge graph (<a href="https://en.wikipedia.org/wiki/Freebase_(database)">by closing once open but unsustainable system</a>) solved this pain point and been serving quality service for the past decade. This unlocked a wide, much rewarding ecosystem under the growing monopoly of Google. Even though this system intentionally caused no harm, as an aftereffect of centralized and tight holding of data and related technologies, a large amount of useful but non entertained (under Google ecosystem) information doesn’t make its way into it. Let’s call it (partially / fully) closed data.</p><p>The generators of closed data are mainly organizations. We nowadays call this data — the big data staying frozen inside organizations. If you are from the IT industry, every product/service has an offering nowadays to address this closed behavior or organization data. Even though this is a common practice and an opportunity for product developers from a business perspective, the organizations mostly don’t get to taste the best technologies being offered to them (in most cases, it’s at the price of data privacy). It’s worth noting that, companies including Google are investing in <a href="https://en.wikipedia.org/wiki/Homomorphic_encryption">homomorphic encryption</a> to address this problem to provide a unified service across all customers. Still, the opportunity limited by the centralized behavior of service and gate-keeping is a big problem when it comes to the accessibility of the data across multiple vendors. Even with these encrypted computations, vendor locking will persist by moving to higher levels and a boundary will get redrawn around the organizations. Many organizations will possibly move out data with much-relaxed restrictions for the public good if it were free out of these boundaries.</p><p>Private data should stay private. The user (person/organization) should have control over it whatever the conditions are. This works best when the computation is moved to the user. As of now, it’s not fully practical and stays as an open problem. One thing for sure, this requires methods to establish the controlled flow of data and computation beyond central boundaries.</p><h3>Open-source software and Non-profit organizations</h3><p>When we look back to history, Open source software has disrupted the technology industry like a storm. At the very beginning, the idea of open-sourcing software was a hacker thing. Now it’s standard. Many companies ranging from startups to corporate have started acknowledging the power of the community for reasons of their own. Most startups see this as their customer acquisition technique while many of the corporate see this as a competition killer or their own standards propagation. Talent/product hunt, public relations are possibly other motivations for this. Whatever the intentions, the results are impactful that we all know. The software industry as a whole is benefiting from this. It doesn’t matter who authored a piece of code as long as it is valuable to someone and it will be. This is a co-operative non-zero-sum game. Continuous forking and improvements across developers who are indie, from an organization or an institution, are boosting both technology education and social evolution. To me, the beauty lies in the openness of source code brings with it trust and co-operation without revealing the identity of engaging peers — thus enabling fairness in innovation.</p><p>This same thing can happen to data as well if we build a sustainable system. As we saw in open-source software, the data producers and consumers can come down build a self-sustainable marketplace. Open-source software gave birth to cloud computing services. And open data could give birth to storage and analytics services that anyone can offer. Everyone adds more value to the system; data generators, data keepers, and data analysts — data can be forked by generating derivatives (insights) to bring more value.</p><p>There are non-profit organizations like Wikipedia already working on this. And as we have seen the derivative — Wiki-data has added more value to the overall system. Also, developers and companies are using Wiki-data to generate more information and value is keeping it closed. In most cases, the reason is that there’s no mechanism to open it up without any incentives. Because the value gets locked up in the pipeline, non-profit organizations like Wikipedia are always on the verge of existential threat. And it is a known fact that the companies that generate more value (Google, Amazon Alexa) give very less in return to Wikipedia which is mainly funded by the generous contributions from the general public. This model is neither sustainable nor progressive. What we need is a self-sustainable data market.</p><h3>Google 2.0</h3><p>Google 2.0 will not be a single person or company. It will be an open market for data and management of it. New and existing entrepreneurs or companies have their own space in this. Of cause Google 1.0 can be a part of this. This market can eventually open up more of closed data in the long run, as entities start recognizing the value for any data they have. Individuals will generate and sell data and services as they wish.</p><h4>Data as a foundation layer</h4><p>To make this a reality, we need to design the foundation layer. Which asks the question of how the data represented at this layer. It should be,</p><ul><li>uniform across networks (<a href="https://en.wikipedia.org/wiki/Ontology_(information_science)">ontology</a> should be the same)</li><li>structured</li><li>both machine and human-readable</li><li>distributed (edge storage and processing)</li></ul><p>Numerous efforts are happening in this domain already. <a href="https://ipfs.io/">IPFS</a> is already setting up a decentralized data storage with incentives to the participants. <a href="https://www.underlay.org/">Underlay</a> is working on a distributed knowledge graph over the IPFS network, <a href="https://www.bigchaindb.com/">BigchainDB</a> is working on the BFT NoSQL database over IPFS, <a href="https://aquila.network/">Aquila Network</a> is working on semantic search indexing to be integrated with IPFS and so on.</p><h4>Data marketplace</h4><p>A data marketplace’s core activity is to bring together a buyer and a seller. This gives a boost to an ecosystem of related services around this including finding the right data, subscribing to events, prediction markets, data restructuring, analytics services, isolated computation, etc. The ecosystem will be driven by clusters of decentralized organizations and carefully designed token economics.</p><p>Data ownership is another important feature of data markets. When data got introduced to the market, the owner who created it will get credited during its lifetime. Data protection makes sure only the authorized parties can access to it even though the storage and networking are made possible by trustless nodes.</p><p>As we have discussed previously, innovations in homomorphic hashing will multiply the possibilities of privacy-first computation and distribution of data without trusting anyone.</p><h4>Decentralized collective intelligence</h4><p>We at ‘a-mma’ believe that in the long run Artificial General Intelligence (AGI) needs to be solved through the <a href="https://medium.com/a-mma/can-we-achieve-safe-ai-through-incentivized-emergence-b2cb7adf6248">emergence out of Human-Machine interactions</a>. As we approach the problem, the first and foremost problem to be solved now is a self-sustaining data layer which is accessible to both humans and machines. Decentralized data markets present a potential opportunity for us.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*pWxi-wsyKqZX5toLJi_7bw.png" /></figure><p><strong><em>a-mma (a_മ്മ)</em></strong><em> is a non profit organization with focus on long term development of swarm intelligence and related technologies. a-mma gives incubation &amp; community support to commercial/non commercial projects in this field of interest and doesn’t own them.</em></p><p><strong><em>Jubin Jose</em></strong><em> is one of the early members of a-mma, still helping it to reach a sustainable point of independent decentralized operation.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dda41099f049" width="1" height="1" alt=""><hr><p><a href="https://medium.com/a-mma/this-is-how-google-2-0-might-look-like-anyone-can-be-part-of-it-without-competition-dda41099f049">This is how Google 2.0 might look like; Anyone can be part of  it, without competition.</a> was originally published in <a href="https://medium.com/a-mma">a-mma</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Can we achieve safe AI through incentivized emergence]]></title>
            <link>https://medium.com/a-mma/can-we-achieve-safe-ai-through-incentivized-emergence-b2cb7adf6248?source=rss-6a042be9b7d9------2</link>
            <guid isPermaLink="false">https://medium.com/p/b2cb7adf6248</guid>
            <category><![CDATA[game-theory]]></category>
            <category><![CDATA[swarm-intelligence]]></category>
            <category><![CDATA[token-economy]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai-safety]]></category>
            <dc:creator><![CDATA[Jubin Jose]]></dc:creator>
            <pubDate>Sun, 21 Jun 2020 18:58:18 GMT</pubDate>
            <atom:updated>2020-06-27T07:13:01.381Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DPqKU_WWSTmIyUuIURdrUA.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@jorgezapatag?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Jorge Zapata</a> on <a href="https://medium.com/s/photos/robot-toys?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p>Despite the debates on how Artificial General Intelligence (AGI) will look like, everyone agrees on one point — human-machine harmony in the future. It is not clear the way down through which we would reach this. Different people possess different opinions. Of course, the future is not predictable, all we can do is speculate based on what we know now. However, in this article, we are going to focus on one among them, that a-mma is contributing to. Let us see how far we can see through with the technology we currently have.</p><h3>End-to-end vs Emergence</h3><p>Recently, a lot of research and engineering efforts are focused on pushing the boundary of end-to-end systems. These systems are typically <a href="https://en.wikipedia.org/wiki/Black_box">black-box</a> mathematical models capable of observing the environment they reside in to plan and generate a probability distribution over the actions they are supposed to take at that state. Also, these systems are supposed to learn and achieve machine consciousness as they behave within a contained or open environment. Although we were able to make great progress in perception capabilities, we still do not know whether these systems achieve some form of primitive representation of machine consciousness due to its limited explainability. In contrast to mathematical models, other approaches like the simulation or replication of the animal brain are also being experimented. However, these simulation-based systems are criticized for the lack of motivation that best generated by an animal body experience during survival. There are intermediary works such as Neuralink which propose to enhance the human brain through support AI components.</p><p>Our interest is in a high-level intelligence, that <a href="https://en.wikipedia.org/wiki/Emergence">emerges</a> from the synergy between stupid independent entities. We sometimes call this a <a href="https://en.wikipedia.org/wiki/Swarm_intelligence">swarm intelligence</a>. We internally (at a-mma) see this as a sophisticated version of <a href="https://en.wikipedia.org/wiki/Distributed_artificial_intelligence">Decentralized AI</a>. Decentralized AI is looking at swarm intelligence through engineer’s eyes at micro-level with macro-level influence from philosophy and natural inspiration. Yeah! (Sorry, if I provoked your instincts over there.) 😀</p><p>This idea of emergent systems are highly inspired by nature itself. Also, we as humans do have experience in engineering complex systems. We could see that the individual components that do not behave anything like the final system, could work with other individual components to produce a highly complex, yet entirely new system that somehow possesses basic intelligence in its environment. Some examples include ant colonies, a mechanical car, a business organization, or human society itself. Not to mention, the internet is a very good example of the emergent system, which we will look more into in the following sections.</p><h3>Closed centralized development of AI</h3><p>Influencers like Elon Musk expressed their concerns on the potential of AI to become dangerous in the future, if not regulated. The possibility of this might escalate if the developments are happening behind the curtain. Currently, most end-to-end systems are being developed and tested by centralized, closed people or organizations. The pile of money and resources required for developing the end-to-end systems, followed by the ROI that should bring — might be the main reasons for being it closed. These probably will give birth to highly biased AI, which might not be much favorable to the public consumers but the organizations itself, and worst, if it goes out of control.</p><h3>In contrast, Open, Decentralized, Swarm Intelligence could reduce the risk levels further down through open collaboration and community benefits.</h3><h3>Designing Emergent AI</h3><p>As we have mentioned earlier, the internet is our friend. We are going to take advantage of whatever currently the internet can provide. We will propose a roadmap based on what we can see now. We will solve problems as a community, and embrace what new technology can offer, to further build the road.</p><p>Consider anything connected to the internet as an <a href="https://en.wikipedia.org/wiki/Rational_agent">agent</a>, whether it is a server, web client (human), IoT device, on the grid software or virus, and whatnot. Normally, an agent possesses an identity (a single agent can have multiple identities on multiple occasions). It makes rational decisions influenced by the dependency with other agents and incentives they might receive as a reward. The rational decision each agent take will be a reflection of their limited intelligence. The agent is free to alter their planning, maybe, themselves through <a href="https://en.wikipedia.org/wiki/Replicator_equation">evolution</a> (software agents can do this faster — makes sense mostly for them). The agents thus form cooperative clusters of organizations that are governed by each of them. This can be achieved through cluster native <a href="https://en.wikipedia.org/wiki/Token_economy">token economics</a> and <a href="https://en.wikipedia.org/wiki/Evolutionary_game_theory">non-zero-sum game</a> mechanics. Don’t worry, we have intentionally compressed and skimmed through multiple ideas in this paragraph — which we will discuss in detail in the articles to come. Also, we are in the initial stages of our exploration as well. However, the key takeaway is, the internet already puts forward solutions to this giant problem either in mature or pre-mature form. Which includes, decentralization, one-one communication protocols, commercial value to be extended with native tokens, etc. All we need to do is, embrace and nurture a native ecosystem of human-machine agents in synergy for mutual benefit.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*4UjxAQgxsvgGGCXbItXGMQ.png" /></figure><blockquote><strong>a-mma (a_മ്മ)</strong> is a non profit organization with focus on long term development of swarm intelligence and related technologies. a-mma gives incubation &amp; community support to commercial/non commercial projects in this field of interest and doesn’t own them.</blockquote><blockquote><strong>Jubin Jose</strong> is one of the early members of a-mma, still helping it to reach a sustainable point of independent decentralized operation.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b2cb7adf6248" width="1" height="1" alt=""><hr><p><a href="https://medium.com/a-mma/can-we-achieve-safe-ai-through-incentivized-emergence-b2cb7adf6248">Can we achieve safe AI through incentivized emergence</a> was originally published in <a href="https://medium.com/a-mma">a-mma</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Tu app necesita un motor de búsqueda de IA]]></title>
            <link>https://medium.com/@josejubin/tu-app-necesita-motor-busqueda-ia-40c3495ad63e?source=rss-6a042be9b7d9------2</link>
            <guid isPermaLink="false">https://medium.com/p/40c3495ad63e</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[search]]></category>
            <category><![CDATA[decentralization]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Jubin Jose]]></dc:creator>
            <pubDate>Mon, 09 Mar 2020 09:01:01 GMT</pubDate>
            <atom:updated>2020-03-09T09:01:01.279Z</atom:updated>
            <content:encoded><![CDATA[<figure><a href="https://chatbotchocolate.com/?utm_source=pc-post&amp;utm_medium=B-top&amp;utm_campaign=general&amp;utm_term=id0050"><img alt="" src="https://cdn-images-1.medium.com/max/680/1*lB6KDoqrcbd_XSVLxIxIpw.png" /></a></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZAwDbQhMPtBusZX1jLC8nQ.jpeg" /><figcaption>Imagen de Pexels de Pixabay</figcaption></figure><p>Mira a tu alrededor … La barra de búsqueda está en todas partes. Está en la lista de contactos de tu teléfono, es tu puerta de entrada a internet, es parte de su biblioteca de fotos. ¿Hablando con Siri o Alexa? Está impulsado por la búsqueda en bases de conocimiento. ¿Reservar una habitación de hotel? ¿Pidiendo comida? ¿Explorando tu red de amigos? ¡Oh espera! Estoy seguro de que también obtienes recomendaciones convincentes. Tienes razón. La búsqueda es inevitable. Honestamente, la palabra ‘buscar’ es bastante vaga. La recuperación de información (RI) es la palabra correcta. Con la ayuda de Recuperación de información, hemos estado intentando durante décadas organizar y generar información perspicaz a partir de la disposición cada vez mayor de datos. No importa de dónde provenga: puede ser desarrollo de software, toma de decisiones, entrega de contenido o académico, la recuperación de información puede agregar valor a tu vida diaria.</p><figure><a href="https://chatbotchocolate.com/?utm_source=Pc-post&amp;utm_medium=B-medio&amp;utm_campaign=general&amp;utm_term=id0051"><img alt="" src="https://cdn-images-1.medium.com/max/680/1*lB6KDoqrcbd_XSVLxIxIpw.png" /></a></figure><p>Hoy, la Inteligencia Artificial (IA) está afectando a todas las industrias. Y la recuperación de información no es una excepción. Un logro notable es que, con la ayuda de los modelos de deep learning, ahora puedes codificar cualquier información en una forma compacta: representación vectorial latente. No importa qué datos estés codificando: Puede ser imágenes, texto, audio, video, gráficos de conocimiento, ADN, compuestos químicos y otros.</p><p>Está bien, está bien … lo entiendo. La idea es codificar cualquier dato en una representación común para que cualquier sistema construido sobre eso no tenga que lidiar con los datos brutos y en su lugar use la codificación en sí, que contiene la esencia de los datos originales.</p><p>Pero, ¿qué significa esta codificación en primer lugar? ¿Y porqué es importante?</p><p>Técnicamente, cuando entrenas un modelo de deep learning, generalizarás ese modelo para aprender patrones comunes de los datos a los que lo has expuesto. Estos patrones aprendidos en los datos se almacenan dentro del propio modelo como capas ocultas. Entonces podemos alcanzar este modelo entrenado para recopilar incrustaciones de alto nivel para cualquier dato en su pase directo. Estas incrustaciones recopiladas estarán más cerca en el espacio de incrustación si los datos de entrada son semánticamente similares y estarán muy separados si las entradas son diferentes. Este ranking de similitud juega un papel importante en la recuperación de información, al agregar mucha más información contextual. Esto nos permite buscar contenido por su significado y es un cambio de juego en la práctica. De ahora en adelante, lo llamaremos un sistema de recuperación de información neuronal porque está respaldado por redes neuronales.</p><p>Incluso si no eres una persona técnica, ahora puedes hacer uso de modelos públicos previamente capacitados disponibles por Google, Microsoft y otras instituciones. Si deseas personalizar aún más esos modelos para tus casos de uso específicos, es muy fácil aplicar el aprendizaje de transferencia a esos modelos previamente entrenados. Una vez que tengas el modelo deseado para tu aplicación específica, ya está.</p><p>¡Terminaste! y estás listo para implementar un sistema IR con un solo paso más.</p><h3>Introduciendo AquilaDB</h3><p><a href="http://aquiladb.xyz/">AquilaDB</a> es una solución <a href="http://aquiladb.xyz/">directa</a> para la recuperación de información neuronal descentralizada. Comienza a ejecutar AquilaDB en un solo comando. El objetivo de AquilaDB es hacer que sea muy fácil para un data science, un ingeniero de ML o un desarrollador de aplicaciones móviles / front-end para iniciar la recuperación de información neuronal en minutos.</p><p>¿Has notado esa cobertura adicional — ‘Descentralizada’? No estas equivocado. AquilaDB está totalmente descentralizado y apunta a la próxima era de la web. Sin punto central de fallo, fuera de línea / borde primero y eventualmente consistente. Impresionante, ¿verdad?</p><p>AquilaDB tiene licencia de código abierto bajo Apache 2.0. <a href="https://aquiladb.xyz/docs/introduction">Comienza con eso</a>. AquilaDB todavía está pasando por un desarrollo activo. Tenemos grandes adiciones de funciones en la cola. Para eso, es necesaria la expansión del desarrollo y las pruebas. Nuestras puertas están completamente abiertas a todo tipo de contribuciones de la comunidad.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Ff%2F68bc97%3Fas_embed%3Dtrue&amp;dntp=1&amp;display_name=Upscribe&amp;url=https%3A%2F%2Fupscri.be%2F68bc97%2F&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/d8021540cbf50434828deb5439e1cf04/href">https://medium.com/media/d8021540cbf50434828deb5439e1cf04/href</a></iframe><figure><a href="https://chatbotchocolate.com/?utm_source=Pc-post&amp;utm_medium=B-pie&amp;utm_campaign=general&amp;utm_term=id0052"><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Axmr4bMScpjBU_WcOm263A.png" /></a></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/788/1*u6pz8xzjblHCXLMQUQGF3Q.gif" /></figure><figure><a href="https://twitter.com/PlanetaChatbot"><img alt="" src="https://cdn-images-1.medium.com/max/788/1*qQxzn3vV9Xu4lYMpuc5OOw.png" /></a></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=40c3495ad63e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your app needs an AI search engine today. Here’s your two-minute recipe.]]></title>
            <link>https://medium.com/a-mma/your-app-needs-an-ai-search-engine-today-heres-your-two-minute-recipe-c5640b2d967e?source=rss-6a042be9b7d9------2</link>
            <guid isPermaLink="false">https://medium.com/p/c5640b2d967e</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[search]]></category>
            <category><![CDATA[decentralization]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Jubin Jose]]></dc:creator>
            <pubDate>Tue, 18 Feb 2020 18:18:16 GMT</pubDate>
            <atom:updated>2021-04-11T08:04:48.157Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZAwDbQhMPtBusZX1jLC8nQ.jpeg" /><figcaption>Image by Pexels from Pixabay</figcaption></figure><p>Look around.. Search is everywhere. It’s in your phone’s contact list. It’s your gateway to the internet. Search is part of your photo library. Talking to Siri or Alexa? It’s powered by search over knowledge bases. Booking a hotel room? Ordering food? Exploring your friend network? Oh, wait! I’m sure that you’re getting compelling recommendations as well. You’re right. Search is inevitable. Honestly, the word ‘search’ is quite vague. Information Retrieval (IR) is the right word to use. With the help of Information Retrieval, we have been trying for decades to organize and generate insightful information from the ever-growing disposal of data. It doesn’t matter whatever background you’re from — it can be software development, decision making, content delivery or academic, Information Retrieval could add value to your daily life.</p><p>Today, Artificial Intelligence (AI) is disrupting every industry. And Information Retrieval is no exception. One notable achievement is, with the help of Deep learning models, you can now encode any information into a compact form — latent vector representation. It doesn’t matter what data you are encoding. It can be images, text, audio, video, knowledge graphs, DNAs, chemical compounds and what not.</p><p>Okay, okay… I get it. Idea is to encode any data into a common representation so that any system built on top of that doesn’t have to deal with the raw data and instead use the encoding itself, which contains the essence of original data.</p><p>But what does this encoding mean in the first place? And why is it important?</p><p>Technically, when you train a deep learning model, you will generalize that model to learn common patterns from the data you have exposed it to. These learned patterns in data are stored within the model itself as hidden layers. We can then peak into this trained model to collect high-level embeddings for any data on its forward pass. These collected embeddings will be closer in embedding space if the input data is semantically similar and will be far apart if the inputs are dissimilar. This similarity ranking plays a big role in Information Retrieval — by adding much more contextual information. This allows us to search for content by its meaning and is a game-changer in practice. From now on we will call this a Neural Information Retrieval System because it is backed by Neural Networks.</p><p>Even if you are a nontechnical person, you can now make use of public pre-trained models made available by Google, Microsoft, and other institutions. If you want to further customize those models to your specific use cases, it’s very easy to apply transfer learning to those pre-trained models. Once you have the desired model to use for your specific application, that’s it. You’re done! and you’re ready to deploy an IR system with one more step.</p><h3>Introducing Aquila Network</h3><p><a href="https://aquila.network/">Aquila Network</a> is a drop-in solution for Decentralized Neural Information Retrieval. Start running Aquila Network in a single command. The goal of Aquila Network is to make it dead easy for a Data Scientist, ML engineer or a front-end / mobile application developer to start Neural Information Retrieval in minutes.</p><p>Have you noticed that extra topping — ‘Decentralized’? You are not wrong. Aquila Network is fully decentralized and is aiming for the next era of the web. No central point of failure, offline / edge first and eventually consistent. Awesome, right?</p><p>Aquila Network is fully open-source licensed under Apache 2.0. <a href="https://aquila.network/">Get started with it now</a>. Aquila Network is still going through active development. We have big feature additions in the queue. For that, expansion of development and testing is necessary. Our doors are fully open to all kinds of community contributions. If you’re interested to work on the futuristic stack or wanted to use it internally and give back to the community, it’s our pleasure to provide a comfy on-boarding.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c5640b2d967e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/a-mma/your-app-needs-an-ai-search-engine-today-heres-your-two-minute-recipe-c5640b2d967e">Your app needs an AI search engine today. Here’s your two-minute recipe.</a> was originally published in <a href="https://medium.com/a-mma">a-mma</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What impact ‘a-mma’ can make in 2020 or simply, why ‘a-mma’?]]></title>
            <link>https://medium.com/a-mma/what-impact-a-mma-can-make-in-2020-or-simply-why-a-mma-6e20d6cf9ab0?source=rss-6a042be9b7d9------2</link>
            <guid isPermaLink="false">https://medium.com/p/6e20d6cf9ab0</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[future]]></category>
            <category><![CDATA[disruption]]></category>
            <category><![CDATA[big-data]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Jubin Jose]]></dc:creator>
            <pubDate>Wed, 22 Jan 2020 11:32:36 GMT</pubDate>
            <atom:updated>2020-01-24T17:34:08.177Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*fHZVIiZ6uqRPz81Ax8shJA.png" /><figcaption>child-learning-in-womb (Artwork: Jubin Jose)</figcaption></figure><p>“Over <strong>2.5 quintillion bytes</strong> of data are created every <strong>single day</strong>” and it’s only going to grow from there. With the advent of affordable mobile and connected devices, cheap cloud services and increasing internet population this data volume will show exponential growth to the future. State of the art advances in Artificial Intelligence is another big boom that happened in this decade as an after effect of this. So, this is very critical for both small and large businesses to figure out and adapt to the technology infrastructure that cope with this data explosion to stay ahead of the game. Identifying the problem in hand, finding the best solution and early laying out the founding infrastructure is the one way out of this.</p><p>We at <strong>a-mma</strong> identified and started working on one such problem that would disrupt the industry in the next decade. To us, Information Retrieval powered by Machine Learning would dramatically change the way a human and a machine interact today. Machines will understand your intuitions faster, with much reduced input from your side. It will be able to process more inputs like speech, visuals, knowledge representations, text and sensor data to build a combined context out of them, compare it with ground truth and make decisions on your behalf. We at <strong>a-mma</strong> are well aware of the power of web and respect human privacy. This force steers our roadmaps since the day we’ve started out. That’s why we’ve put community first, to inspire and motivate us to move forward. And guess what? <strong>a-mma</strong> is an open source community.</p><p><a href="https://aquiladb.xyz/"><strong>AquilaDB</strong></a> is our first effort in all this. It is a Document + Vector database to perform efficient <a href="https://www.google.com/search?q=Neural+Information+Retrieval"><strong>Neural Information Retrieval</strong></a> on very large data of any kind.</p><p>You can check our Github community <a href="https://github.com/a-mma"><strong>a-mma</strong></a> for more interesting Open Source projects like this. As of now, we’re very limited in resources. Because of this, our current focus is on <a href="https://github.com/a-mma/AquilaDB"><strong>AquilaDB</strong></a> only. If you are interested in contributing in any manner, please check <a href="https://github.com/a-mma/AquilaDB"><strong>AquilaDB repository</strong></a>, star it and try using it, report bugs, submit pull requests, help fellow community members, spread the word and contact us at <a href="http://humans@aquiladb.xyz/"><strong>humans@aquiladb.xyz</strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6e20d6cf9ab0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/a-mma/what-impact-a-mma-can-make-in-2020-or-simply-why-a-mma-6e20d6cf9ab0">What impact ‘a-mma’ can make in 2020 or simply, why ‘a-mma’?</a> was originally published in <a href="https://medium.com/a-mma">a-mma</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>