Smile

smile

Authors:
(1) Clayton W. Kosonocky, Department of Molecular Biosciences, The University of Texas at Austin (clayton.kosonocky@utexas.edu);
(2) Aaron L. Feller, Department of Molecular Biosciences, The University of Texas at Austin (aaron.feller@utexas.edu);
(3) Claus O. Wilke, Department of Integrative Biology, The University of Texas at Austin and Corresponding Author (wilke@austin.utexas.edu);
(4) Andrew D. Ellington, Department of Molecular Biosciences, The University of Texas at Austin (ellingtonlab@gmail.com). Table of Links Abstract & Introduction
Methods
Results and Discussion
Determining Whether Canonicalization Impacts Search Behavior
Explanation of Search Behavior & Drawbacks, Future Improvements, and Potential for Misuse
Conclusion, Acknowledgements, Author Contributions, & more.
Supplementary Figures 3. Results and Discussion In-silico drug discovery methods have long relied on chemical similarity searches as a computational tool in pharmaceutical pipelines [22]. Recently, language models have been successful in biochemical prediction tasks for chemical properties such as drug-likeness, protein-ligand interactions, and other metrics [11, 13]. Because a multitude of characteristics can be predicted from chemical language models (CLMs), it is plausible that embeddings from these models contain a summary of molecular properties for a given molecule. So-called Simplified Molecular Line Entry System (SMILES) representations can be used to represent chemical structures as strings and can be used as inputs to CLMs, but due to the nature of chemical connectivity, there are often many valid representations for the same molecule [17]. This multiplicity of input formats causes the CLM to generate different embeddings for the same molecule, which can either be mitigated through string standardization (canonicalization) or utilized as a data augmentation technique that allows CLMs to better span molecular space [25, 26]. The latter reduces overfitting to string artifacts and improves structural comprehension; an analogy in natural language would be that training a model on all languages improves the understanding of language as a whole compared to a model trained only on English [25, 26]. However, the combination of alternative canonicalizations and data augmentation may actually prove to be advantageous for a chemical similarity search. Recent reports on emergence from the natural language processing literature demonstrated that language models understand Swahili and other languages incredibly well despite them being <0.01% of training data [27, 28]. SMILES canonicalization formats often share characters and substructures with one another, the primary difference being the specific grammatical rules used to assemble a given molecule. For CLMs trained on one canonicalization, inputs using unseen canonicalizations would be akin to underrepresented languages in natural language training sets, as the underlying string structure of the unseen canonicalization is not entirely unknown to the model, but is novel in the context of the molecule at hand. It seems likely that a vector comparison between embeddings from two different canonicalizations could ignore the differences in string and structural data and instead use whole-molecule properties approximating molecular function, serving as a novel prompt engineering strategy for the discovery of structurally distinct functional analogues. This would be equivalent to querying a predominantly English-trained model with a French phrase to obtain an embedding, and then searching amongst English phrases to find the entry with the closest semantic meaning. To our knowledge this potential behavior is largely unexplored. To test this hypothesis, we built a CLM-embedding-based chemical similarity search utilizing one canonicalization for the CLM and database, and another canonicalization for the query SMILES strings. A pipeline was developed to perform a chemical similarity search using cosine similarity on CLM embeddings obtained from ChemBERTa, an unsupervised transformer encoder-based model [29]. This pipeline was named the Chemical Semantic Search (CheSS), outlined in Figure 1. ChemBERTa was trained on SMILES strings canonicalized using the default RDKit implementation, herein referred to as “RDKit Atom 0” [29, 31]. We converted ChemBERTa’s training set into a molecular database of ∼10M RDKit Atom 0 SMILES strings, upon which CheSS searches were performed. We then explored how the CheSS search results were impacted by three different SMILES canonicalizations. The first was RDKit’s default canonicalization “RDKit Atom 0”. The second canonicalization was created by maximizing the distance of the created string embedding to RDKit Atom 0 by varying the root atom number, herein referred to as “RDKit Atom n” (Fig. S2). The third used OEChem 2.3.0, a markedly different algorithm, to canonicalize the query and is referred to as “OEChem”. All CheSS searches utilizing these three canonicalizations differed only in the representation of the query molecule, with the database and model remaining constant. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Clayton W. Kosonocky, Department of Molecular Biosciences, The University of Texas at Austin (clayton.kosonocky@utexas.edu); (2) Aaron L. Feller, Department of Molecular Biosciences, The University of Texas at Austin (aaron.feller@utexas.edu); (3) Claus O. Wilke, Department of Integrative Biology, The University of Texas at Austin and Corresponding Author (wilke@austin.utexas.edu); (4) Andrew D. Ellington, Department of Molecular Biosciences, The University of Texas at Austin (ellingtonlab@gmail.com). Authors: Authors: (1) Clayton W. Kosonocky, Department of Molecular Biosciences, The University of Texas at Austin (clayton.kosonocky@utexas.edu); (2) Aaron L. Feller, Department of Molecular Biosciences, The University of Texas at Austin (aaron.feller@utexas.edu); (3) Claus O. Wilke, Department of Integrative Biology, The University of Texas at Austin and Corresponding Author (wilke@austin.utexas.edu); (4) Andrew D. Ellington, Department of Molecular Biosciences, The University of Texas at Austin (ellingtonlab@gmail.com). Table of Links Abstract & Introduction Methods Results and Discussion Determining Whether Canonicalization Impacts Search Behavior Explanation of Search Behavior & Drawbacks, Future Improvements, and Potential for Misuse Conclusion, Acknowledgements, Author Contributions, & more. Supplementary Figures Abstract & Introduction Abstract & Introduction Methods Methods Results and Discussion Results and Discussion Determining Whether Canonicalization Impacts Search Behavior Determining Whether Canonicalization Impacts Search Behavior Explanation of Search Behavior & Drawbacks, Future Improvements, and Potential for Misuse Explanation of Search Behavior & Drawbacks, Future Improvements, and Potential for Misuse Conclusion, Acknowledgements, Author Contributions, & more. Conclusion, Acknowledgements, Author Contributions, & more. Supplementary Figures Supplementary Figures 3. Results and Discussion In-silico drug discovery methods have long relied on chemical similarity searches as a computational tool in pharmaceutical pipelines [22]. Recently, language models have been successful in biochemical prediction tasks for chemical properties such as drug-likeness, protein-ligand interactions, and other metrics [11, 13]. Because a multitude of characteristics can be predicted from chemical language models (CLMs), it is plausible that embeddings from these models contain a summary of molecular properties for a given molecule. So-called Simplified Molecular Line Entry System (SMILES) representations can be used to represent chemical structures as strings and can be used as inputs to CLMs, but due to the nature of chemical connectivity, there are often many valid representations for the same molecule [17]. This multiplicity of input formats causes the CLM to generate different embeddings for the same molecule, which can either be mitigated through string standardization (canonicalization) or utilized as a data augmentation technique that allows CLMs to better span molecular space [25, 26]. The latter reduces overfitting to string artifacts and improves structural comprehension; an analogy in natural language would be that training a model on all languages improves the understanding of language as a whole compared to a model trained only on English [25, 26]. However, the combination of alternative canonicalizations and data augmentation may actually prove to be advantageous for a chemical similarity search. Recent reports on emergence from the natural language processing literature demonstrated that language models understand Swahili and other languages incredibly well despite them being <0.01% of training data [27, 28]. SMILES canonicalization formats often share characters and substructures with one another, the primary difference being the specific grammatical rules used to assemble a given molecule. For CLMs trained on one canonicalization, inputs using unseen canonicalizations would be akin to underrepresented languages in natural language training sets, as the underlying string structure of the unseen canonicalization is not entirely unknown to the model, but is novel in the context of the molecule at hand. It seems likely that a vector comparison between embeddings from two different canonicalizations could ignore the differences in string and structural data and instead use whole-molecule properties approximating molecular function, serving as a novel prompt engineering strategy for the discovery of structurally distinct functional analogues. This would be equivalent to querying a predominantly English-trained model with a French phrase to obtain an embedding, and then searching amongst English phrases to find the entry with the closest semantic meaning. To our knowledge this potential behavior is largely unexplored. To test this hypothesis, we built a CLM-embedding-based chemical similarity search utilizing one canonicalization for the CLM and database, and another canonicalization for the query SMILES strings. A pipeline was developed to perform a chemical similarity search using cosine similarity on CLM embeddings obtained from ChemBERTa, an unsupervised transformer encoder-based model [29]. This pipeline was named the Chemical Semantic Search (CheSS), outlined in Figure 1. ChemBERTa was trained on SMILES strings canonicalized using the default RDKit implementation, herein referred to as “RDKit Atom 0” [29, 31]. We converted ChemBERTa’s training set into a molecular database of ∼10M RDKit Atom 0 SMILES strings, upon which CheSS searches were performed. We then explored how the CheSS search results were impacted by three different SMILES canonicalizations. The first was RDKit’s default canonicalization “RDKit Atom 0”. The second canonicalization was created by maximizing the distance of the created string embedding to RDKit Atom 0 by varying the root atom number, herein referred to as “RDKit Atom n” (Fig. S2). The third used OEChem 2.3.0, a markedly different algorithm, to canonicalize the query and is referred to as “OEChem”. All CheSS searches utilizing these three canonicalizations differed only in the representation of the query molecule, with the database and model remaining constant. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

Chemical Language Models Improve Similarity Search With SMILES Variations

Authors:
(1) Clayton W. Kosonocky, Department of Molecular Biosciences, The University of Texas at Austin (clayton.kosonocky@utexas.edu);
(2) Aaron L. Feller, Department of Molecular Biosciences, The University of Texas at Austin (aaron.feller@utexas.edu);
(3) Claus O. Wilke, Department of Integrative Biology, The University of Texas at Austin and Corresponding Author (wilke@austin.utexas.edu);
(4) Andrew D. Ellington, Department of Molecular Biosciences, The University of Texas at Austin (ellingtonlab@gmail.com). Table of Links Abstract & Introduction
Methods
Results and Discussion
Determining Whether Canonicalization Impacts Search Behavior
Explanation of Search Behavior & Drawbacks, Future Improvements, and Potential for Misuse
Conclusion, Acknowledgements, Author Contributions, & more.
Supplementary Figures 3.2 Explanation of Search Behavior We find that alternative canonicalizations influence the search behavior of CheSS through changes in tokenization, which causes the CLM to weight higher-order relationships more importantly when creating embeddings. There are stark differences between RDKit and OEChem canonicalizations, notably their differences in the representation of aromatic rings. OEChem prefers the Kekulé form (C1=CC=CC=C1), while RDKit prefers to use lowercase with assumed aromaticity (c1ccccc1), and these differences, among others, result in markedly different tokenization, both in the composition and the length of the tokenized vectors (Fig. 4). Since the CLM has demonstrated a bias toward embedding molecules with similar token vector lengths and (to a lesser extent) token composition to the query (Fig. 4), a query with a different canonicalization will tokenize into a radically different token vector and thereby make the CLM more likely to return molecules with SMILES representations that are highly dissimilar from the original same-canonicalized query. Despite this behavior, and potentially because of it, CheSS searches with alternative canonicalizations found diverse chemical structures with similar functional properties, as demonstrated by patent and literature searches (Fig. 5). When CLMs are forced to go beyond simple token patterns to determine similarity, more nuanced relationships may appear. Given the nature of transformers, it is indiscernible what these relationships are, but based on our analysis we find it possible that the CLM may, for example, key on the apposition of functional groups in space, in a way similar to how receptors perceive ligands. 3.3 Drawbacks, Future Improvements, and Potential for Misuse The database used by CheSS consisted of the ∼10M molecules used as a training set for ChemBERTa, and thus it is difficult to predict the behavior of queries that differ greatly from the molecules in this dataset. In addition, very small molecules may not differ in their canonicalized representations, leading to more homogeneity between queries. Nonetheless, the method itself is extensible for use with any dataset, and may invite discussions regarding what datasets and CLMs are most useful for moving between different canonicalizations to identify functional analogues. We did not at this time explore whether non-canonicalized, yet valid, variations of SMILES strings will lead to similar results. We also note that the threat of dual use for chemical machine learning models has been a topic of discussion amongst researchers [49]. While the model used in our implementation of CheSS was unsupervised and not trained for identifying toxic molecules, a successful chemical similarity search tool carries inherent risks. We therefore advise caution in considering public implementations of these tools and recommend restricting searches to avoid queries with the potential for malicious use. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Clayton W. Kosonocky, Department of Molecular Biosciences, The University of Texas at Austin (clayton.kosonocky@utexas.edu); (2) Aaron L. Feller, Department of Molecular Biosciences, The University of Texas at Austin (aaron.feller@utexas.edu); (3) Claus O. Wilke, Department of Integrative Biology, The University of Texas at Austin and Corresponding Author (wilke@austin.utexas.edu); (4) Andrew D. Ellington, Department of Molecular Biosciences, The University of Texas at Austin (ellingtonlab@gmail.com). Authors: Authors: (1) Clayton W. Kosonocky, Department of Molecular Biosciences, The University of Texas at Austin (clayton.kosonocky@utexas.edu); (2) Aaron L. Feller, Department of Molecular Biosciences, The University of Texas at Austin (aaron.feller@utexas.edu); (3) Claus O. Wilke, Department of Integrative Biology, The University of Texas at Austin and Corresponding Author (wilke@austin.utexas.edu); (4) Andrew D. Ellington, Department of Molecular Biosciences, The University of Texas at Austin (ellingtonlab@gmail.com). Table of Links Abstract & Introduction Methods Results and Discussion Determining Whether Canonicalization Impacts Search Behavior Explanation of Search Behavior & Drawbacks, Future Improvements, and Potential for Misuse Conclusion, Acknowledgements, Author Contributions, & more. Supplementary Figures Abstract & Introduction Abstract & Introduction Methods Methods Results and Discussion Results and Discussion Determining Whether Canonicalization Impacts Search Behavior Determining Whether Canonicalization Impacts Search Behavior Explanation of Search Behavior & Drawbacks, Future Improvements, and Potential for Misuse Explanation of Search Behavior & Drawbacks, Future Improvements, and Potential for Misuse Conclusion, Acknowledgements, Author Contributions, & more. Conclusion, Acknowledgements, Author Contributions, & more. Supplementary Figures Supplementary Figures 3.2 Explanation of Search Behavior We find that alternative canonicalizations influence the search behavior of CheSS through changes in tokenization, which causes the CLM to weight higher-order relationships more importantly when creating embeddings. There are stark differences between RDKit and OEChem canonicalizations, notably their differences in the representation of aromatic rings. OEChem prefers the Kekulé form (C1=CC=CC=C1), while RDKit prefers to use lowercase with assumed aromaticity (c1ccccc1), and these differences, among others, result in markedly different tokenization, both in the composition and the length of the tokenized vectors (Fig. 4). Since the CLM has demonstrated a bias toward embedding molecules with similar token vector lengths and (to a lesser extent) token composition to the query (Fig. 4), a query with a different canonicalization will tokenize into a radically different token vector and thereby make the CLM more likely to return molecules with SMILES representations that are highly dissimilar from the original same-canonicalized query. Despite this behavior, and potentially because of it, CheSS searches with alternative canonicalizations found diverse chemical structures with similar functional properties, as demonstrated by patent and literature searches (Fig. 5). When CLMs are forced to go beyond simple token patterns to determine similarity, more nuanced relationships may appear. Given the nature of transformers, it is indiscernible what these relationships are, but based on our analysis we find it possible that the CLM may, for example, key on the apposition of functional groups in space, in a way similar to how receptors perceive ligands. 3.3 Drawbacks, Future Improvements, and Potential for Misuse The database used by CheSS consisted of the ∼10M molecules used as a training set for ChemBERTa, and thus it is difficult to predict the behavior of queries that differ greatly from the molecules in this dataset. In addition, very small molecules may not differ in their canonicalized representations, leading to more homogeneity between queries. Nonetheless, the method itself is extensible for use with any dataset, and may invite discussions regarding what datasets and CLMs are most useful for moving between different canonicalizations to identify functional analogues. We did not at this time explore whether non-canonicalized, yet valid, variations of SMILES strings will lead to similar results. We also note that the threat of dual use for chemical machine learning models has been a topic of discussion amongst researchers [49]. While the model used in our implementation of CheSS was unsupervised and not trained for identifying toxic molecules, a successful chemical similarity search tool carries inherent risks. We therefore advise caution in considering public implementations of these tools and recommend restricting searches to avoid queries with the potential for malicious use. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

SMILES Variations Influence Chemical Search Behavior and Functional Discovery

NEW YORK, United States, December 3rd, 2024/Chainwire/--Smile Shop, the premier Asian e-commerce platform, has partnered with  Conflux Network, China's only regulatory-compliant public blockchain, to launch the BitUnion prepaid card. This partnership signifies a major expansion into global digital asset payment with cards welcome in 183 countries and regions. The BitUnion prepaid card operates on the recently launched UnionPay International USD prepaid card framework. UnionPay International, the world’s second-largest card payment processor, ensures seamless financial transactions. Users can load funds onto the card to make online purchases, use it at point-of-sale machines, or withdraw cash from UnionPay ATMs across 183 countries. The card can also be linked with popular third-party payment platforms like Alipay and WeChat Pay. Whether offline or online, transactions are settled at real-time exchange rates between local currencies and USD. The account approval process is quick, and management fees are waived during the initial launch period. As a key product in Conflux's PayFi ecosystem, the prepaid card incorporates experienced security authentication systems from traditional finance. User data is managed by professional institutions, ensuring security and compliance. Fiat assets are held within the UnionPay account system, guaranteeing absolute security and reliability. The BitUnion prepaid card has obtained the highest-level financial security certifications, including 3DS and PCI-DSS, comprehensively protecting cardholders' payment security. The prepaid card will support digital asset transactions and transfers within the Conflux Network, leveraging the blockchain-based PayFi system to overcome limitations in traditional payment infrastructure. Introducing traditional financial models (from credit cards to invoice financing and reverse factoring) into blockchain creates a more integrated value network. Conflux’s PayFi (Pay Finance) addresses inefficiencies in traditional payment systems while keeping financial operations aligned with real-time data, creating a large-scale model for blockchain consumer application ecosystems. As a high-performance Layer1 blockchain, Conflux has been at the forefront of technological advancements, particularly in the development of Stablecoins. They are now expanding their focus to encompass a comprehensive Payments infrastructure and cultivate the PayFi ecosystem. Aiming to become the blockchain of choice for consumer-grade Payments in the future, Conflux Foundation has committed 500 million CFX from the ecosystem fund to fuel the growth of PayFi stack components. About Smile Shop Smile Shop is a super e-commerce platform under Smile Shop Holdings Pte (Singapore), targeting Southeast Asian markets with the vision of becoming Southeast Asia's most trusted fintech super e-commerce platform. About Conflux Network Conflux Network is a permissionless Layer 1 blockchain that connects decentralized economies worldwide. It utilizes a hybrid PoW/PoS consensus mechanism, ensuring a fast, secure, and scalable blockchain environment. Conflux operates without congestion, maintains low fees, and prioritizes network security. Being the leading regulatory-compliant public blockchain in China, Conflux offers advantages for projects entering the Asian market. In its partnerships, Conflux collaborates with global brands and government entities including, Shanghai, China Telecom, Little Red Book (China’s Instagram), McDonald’s China, and Oreo. These noteworthy collaborations serve as a testament to Conflux’s unwavering dedication to driving blockchain and metaverse initiatives. Contact Melissa Tirey melissa@shift6studios.com This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here NEW YORK, United States, December 3rd, 2024/Chainwire/--Smile Shop, the premier Asian e-commerce platform, has partnered with  Conflux Network, China's only regulatory-compliant public blockchain, to launch the BitUnion prepaid card. This partnership signifies a major expansion into global digital asset payment with cards welcome in 183 countries and regions. The BitUnion prepaid card operates on the recently launched UnionPay International USD prepaid card framework . UnionPay International, the world’s second-largest card payment processor, ensures seamless financial transactions. UnionPay International USD prepaid card framework UnionPay International USD prepaid card framework Users can load funds onto the card to make online purchases, use it at point-of-sale machines, or withdraw cash from UnionPay ATMs across 183 countries. The card can also be linked with popular third-party payment platforms like Alipay and WeChat Pay. Whether offline or online, transactions are settled at real-time exchange rates between local currencies and USD. The account approval process is quick, and management fees are waived during the initial launch period. As a key product in Conflux's PayFi ecosystem, the prepaid card incorporates experienced security authentication systems from traditional finance. User data is managed by professional institutions, ensuring security and compliance. Fiat assets are held within the UnionPay account system, guaranteeing absolute security and reliability. The BitUnion prepaid card has obtained the highest-level financial security certifications, including 3DS and PCI-DSS, comprehensively protecting cardholders' payment security. The prepaid card will support digital asset transactions and transfers within the Conflux Network, leveraging the blockchain-based PayFi system to overcome limitations in traditional payment infrastructure. Introducing traditional financial models (from credit cards to invoice financing and reverse factoring) into blockchain creates a more integrated value network. Conflux’s PayFi (Pay Finance) addresses inefficiencies in traditional payment systems while keeping financial operations aligned with real-time data, creating a large-scale model for blockchain consumer application ecosystems. As a high-performance Layer1 blockchain, Conflux has been at the forefront of technological advancements, particularly in the development of Stablecoins. They are now expanding their focus to encompass a comprehensive Payments infrastructure and cultivate the PayFi ecosystem. Aiming to become the blockchain of choice for consumer-grade Payments in the future, Conflux Foundation has committed 500 million CFX from the ecosystem fund to fuel the growth of PayFi stack components. About Smile Shop Smile Shop is a super e-commerce platform under Smile Shop Holdings Pte (Singapore), targeting Southeast Asian markets with the vision of becoming Southeast Asia's most trusted fintech super e-commerce platform. Smile Shop Smile Shop About Conflux Network Conflux Network is a permissionless Layer 1 blockchain that connects decentralized economies worldwide. It utilizes a hybrid PoW/PoS consensus mechanism, ensuring a fast, secure, and scalable blockchain environment. Conflux Network Conflux Network Conflux operates without congestion, maintains low fees, and prioritizes network security. Being the leading regulatory-compliant public blockchain in China, Conflux offers advantages for projects entering the Asian market. In its partnerships, Conflux collaborates with global brands and government entities including, Shanghai, China Telecom, Little Red Book (China’s Instagram), McDonald’s China, and Oreo. These noteworthy collaborations serve as a testament to Conflux’s unwavering dedication to driving blockchain and metaverse initiatives. Contact Melissa Tirey melissa@shift6studios.com This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here This story was distributed as a release by Chainwire under HackerNoon’s Business Blogging Program. Learn more about the program here here here

Smile Shop Joins Conflux PayFi Ecosystem With BitUnion Prepaid Card

የፈገግታ ሱቅ Conflux PayFi ስነ-ምህዳርን ከ BitUnion ቅድመ ክፍያ ካርድ ጋር ተቀላቅሏል።

Smile Shop, BitUnion Əvvəlcədən Ödənişli Kartla Conflux PayFi Ekosisteminə qoşulur

Το Smile Shop συμμετέχει στο οικοσύστημα Conflux PayFi με προπληρωμένη κάρτα BitUnion

Smile Shop se suma al ecosistema Conflux PayFi con la tarjeta prepaga BitUnion

A Smile Shop csatlakozik a Conflux PayFi ökorendszerhez a BitUnion kártyás kártyával

スマイルショップがBitUnionプリペイドカードでConflux PayFiエコシステムに参加

Smile Shop ចូលរួមជាមួយប្រព័ន្ធអេកូ Conflux PayFi ជាមួយនឹងកាតបង់ប្រាក់ជាមុន BitUnion

Smile Shop د BitUnion پری پیډ کارت سره Conflux PayFi اکوسیستم سره یوځای کیږي

Smile Shop เข้าร่วมระบบนิเวศ Conflux PayFi ด้วยบัตรเติมเงิน BitUnion

Ileylgyryş dükany BitUnion öňünden tölenen kartoçkasy bilen “Conflux PayFi” ekosistemasyna goşulýar

Smile Shop приєднується до екосистеми Conflux PayFi за допомогою передплаченої картки BitUnion

Smile Shop BitUnion پری پیڈ کارڈ کے ساتھ Conflux PayFi ایکو سسٹم میں شامل ہو گئی۔

Beyond the Smile: Decoding Emoji Lengths in JavaScript

Fortune Smiles On Mitya

From 1958 to 1963, Americans had many reasons to be nervous about the future of technology. The Cuban Missile Crisis viscerally brought home that the crowning achievement of the atomic age, the nuclear bomb, had introduced the risk that a minor conflict could destroy the world by accident, all while Sputnik showed that the USSR had surpassed American engineering. Yet, one of the most popular comic strips during that period was , which ran in newspapers across America. Even in the depths of existential despair, we managed to maintain a sense of optimism surrounding the future that could flow from technology. It’s not just Radeaugh. In times long past, the popular press focused on our exciting future. Hugo Gernsbacks, of the Hugo Award, had a series called “Amazing Stories.” The most prominent pop culture from the time includes Star Trek, a utopian exploration series, and Star Wars, about a successful rebellion in space. Arthur Radeaugh’s Closer Than We Think Today’s builders are met with a parade of haters, the kind of people who would of the history aficionado who is Wikipedia’s biggest contributor. One of the , historically a booster for technology, was about the risks; was about the need of a national biology policy. , though optimistic about self-driving cars, fears gravely for the future of a general artificial intelligence. has even wondered whether the Luddites were on to something. And Mark Zuckerberg’s goal of curing disease was a “sickeningly bad idea” (seriously, lay off the Malthus!). This comes on the heels of a , spurred on in part by the lingering effects of the financial crisis and in part by the depressing state of democracy in the West. make fun of the looks first articles about CRISPR out of Wired another Elon Musk The Smithsonian one writer in The Telegraph general depression with the direction of the world When it comes to technology, the world feels like it’s in a special funk. There are legitimate concerns, and they should be discussed, but they aren’t the driving trend behind recent coverage. The dominant factor is just that , even if they miss the big picture. After all, “if it bleeds, it leads.” But the irony is that society’s amazing progress comes from the clear-eyed optimism of entrepreneurs, not the pessimism of nitpicking naysayers. Founder optimism isn’t the belief that there’s nothing wrong (then why start a business?) but that we can make things better with some cleverness and hard work. And technology is an essential part of that story. negative stories tend to sell better than positive ones The way you make things better isn’t by debate but by taking action. People won’t always understand why or how they agree but they’ll vote with their wallets and their feet. Once society has actually gotten better and approached stability people will start to realize the good of what you’ve done, with the nice side effect of improving the lives of potential millions or billions of people. The chattering class critiques that change because they fear losing their privileged position. Even reading was once criticized as harmful to youth. Not by the medieval Church attacking printing press as a threat to its informational hegemony, but by 1930s parenting guides. The reality is that technology is making the world better. It’s to save lives. It’s to fend off hunger. It’s brought car service to while . While the media loves to bash large tech companies, they’re also some of the and . While the rest of , . They’re building the future of , , and . Global inequality is way down, life expectancy is way up, and you can call someone halfway across the world for free. That’s pretty great stuff. In fact, Steven Pinker wrote a whole book about it called that you should read. letting us bring blood to remote locations improving crop yields underserved areas cutting down on drunk driving most loved brands admired companies the world is cutting down on R&D tech companies aren’t medical devices self-driving cars same-day delivery Enlightenment Now! The world isn’t perfect. The so-called sometimes seems to be mostly composed of loss-making gig and subscription startups, some fraud (*cough* Theranos), and Juicero-scale money-pits instead of life-altering technologies. Caricatures aside, I’m concerned that technologists aren’t talking enough to subject matter experts and are therefore wasting energy in bad problem spaces. But that concern is tempered by the amazing founders who are rethinking not only our global economy but life and the human experience itself. And it pales in comparison to my fear that vocal pessimists are discouraging some people from becoming founders and changing the world, resulting in a huge opportunity cost. , but when you’re changing the world that’s just part of the game. I’ll be on the side of the people who are at least trying to make things better, thanks. VC economy There will be risk and failure And when it comes to the press, right now it feels like much of the coverage is driven by vengeance, not neutral factfinding. It’s just victim blaming to print a calling Twitter “complicit” in helping Saudi Arabia silence dissidents when it was spied on — and then promptly fired the employee when they discovered he was a mole. Journalists have often criticized Amazon, calling its HQ2 search a “ ” — but I only know of contemporaneous , and they both show overwhelming support for the project, probably because 25,000 people want the chance at a higher paying job and the attendant local development. This is part of a pattern. If only a journalist would say the quiet part loud. headline con, not a contest two polls Oh. As a founder, it’s easy to get discouraged by the headlines. Just keep your head down and build the pessimists just won’t matter. Their complaints will stay the same. Your company won’t. Remember the timeless words of Teddy Roosevelt: It is not the critic who counts; not the man who points out how the strong man stumbles, or where the doer of deeds could have done them better. The credit belongs to the man who is actually in the arena, whose face is marred by dust and sweat and blood; who strives valiantly; who errs, who comes short again and again, because there is no effort without error and shortcoming; but who does actually strive to do the deeds; who knows great enthusiasms, the great devotions; who spends himself in a worthy cause; who at the best knows in the end the triumph of high achievement, and who at the worst, if he fails, at least fails while daring greatly, so that his place shall never be with those cold and timid souls who neither know victory nor defeat. Even the relatively dark “Metropolis,” considered by some to be the first Sci-Fi film, had an optimistic message at the end: “The Mediator Between the Head and the Hands Must Be the Heart.” So keep at it. The world needs you, even if it won’t always be grateful.

Smile and Keep Building

A very good friend of mine has heard me tell many a tale of trips in which I have been given free drinks, gratis hotel and car rental upgrades and various freebies. However, without experiencing my stories first hand, he always suspected there was a degree of embellishment involved to make my stories a bit more entertaining. travel We eventually had the opportunity to together and I was able to demonstrate the power of a little respect and an authentic smile. travel It’s Kool To Be Kind We began our trip sitting in the dreary, uncomfortable airport terminal, with ample time before our flight. Bored, I explored the terminal and located the airline’s club lounge. Despite not being a member, I entered and immediately knew where I was going to spend the next couple of hours. Sitting behind the receptionist desk, in a chair that was two sizes too small for him was, Don, a large Hawaiian with a hearty smile. He was clearly uninterested in the paperback book in his oversized hands and he seemed open to chatting. After some brief pleasantries, I asked him what it cost to join the club and if they offered a one-day pass. He laughed and apologetically said, “It’s something like $500 a year. We don’t have day rates, but I can let you slide this time if you just want to check it out.” I thanked him and hurried back to grab my buddy. After a leisurely time in the airline club, we boarded our flight and met Kelly, a young surfer dude in his early twenties who was our attendant for the next five hours. Once everyone was settled in their seats, I asked him where he lived in Hawaii, where he surfed, what sort of board he used, etc. As soon as we were airborne, I asked Kelly if he could hook me up with a Mai Tai. He laughed and asked if I wanted one of his “special creations.” My reply, without hesitation, “Why of course.” He returned and waited for me to sample the drink. I took a large swallow and sincerely told him, “You are an artist. You should call this ‘Kelly’s Creation.” I had several more of Kelly’s Creations before we landed, all at no charge. We bid Kelly farewell, hoping we’d see him again on our return flight and headed toward the car rental agency. It was very hot and the woman at the counter did not look happy. I approached her with a broad smile and asked how her day was going. She replied, “It’s about to get worse. All the flights from the mainland are landing.” She clearly wasn’t happy. After a bit of joking around, she begrudgingly grinned and eventually was laughing at my corny humor. She waived the fee for my friend to be an authorized driver and she gave us a gratis upgrade to a slightly nicer vehicle. So What Entrepreneurs are likewise well served to seek out and cultivate transitory relationships with service employees whom self-important professionals generally overlook, such as: waitresses, car rental agents, flight attendants and call-center agents. Such folks encounter a horde of unhappy, disgruntled and downright surly travelers each and every day. Thus, it is easy to differentiate yourself by simply smiling and showing these hard-working people the courtesy they deserve. Greeting someone by genuinely asking, “Wow. How can you look so cool under pressure when you are so busy?” will often be rewarded with a smile and a special deal. Making A Connection Make an authentic connection with a friendly, sincere smile. Then engage the person you are approaching as a peer, a courtesy many service workers seldom experience from most of their harried, self-absorbed customers. The interaction should always be based on a genuine expression of interest. For instance, after a long day of travel, I approached the night clerk at a hotel at about 1:00 in the morning. He was clearly studying a textbook and I genuinely admired that he had a night job while attending college. Rather than ignoring his book or asking him a superficial question like, “What’s your major” or “Where do you go to school?” I asked him, “What are your dreams once you graduate?” He hesitated, clearly debating if he should give me a thoughtful, honest answer or if he should remain in transaction mode and give me a cursory, glib reply. He opted for honesty and told me that he wanted to someday open a veterinary clinic because he loved animals. We had a pleasant conversation, I was given a free upgrade without even asking and we parted with me sincerely wishing him the best. I didn’t even realized he had upgraded me until I arrived in my multi-room suite. I wasn’t thrilled with the suite, which I only spent a couple waking hours in. What made my evening was the personal connection we briefly shared. The Real Reward If you are rolling your eyes , I will let you in on a — the real reward is not saving a few dollars. secret The foundation upon which such personal connections are built is the levity and kindness infused into otherwise tedious situations. Traveling becomes less mundane and a lot more joyful when you create a genuine rapport with the folks you meet along the way. The small, tangible freebies that sometimes result from such interactions are a nice bonus, but the real payoff is making brief, yet meaningful connections you will make with some awesome people with whom you share this planet. Follow John’s startup-oriented Twitter feed here: @johngreathouse

The Secret To Getting Free Drinks, Upgrades And Genuine Smiles When You Travel

And why it matters for building real-world AI systems, especially in healthcare. And why it matters for building real-world AI systems, especially in healthcare. It started with a question. Not a difficult one on the surface, just something a curious interviewer tossed my way during a discussion on AI systems: “So what’s the difference between a chatbot and an AI agent?” “So what’s the difference between a chatbot and an AI agent?” “So what’s the difference between a chatbot and an AI agent?” I smiled. Because if you’ve worked in the AI space, you know this is a loaded question. The kind of question that hides a vast philosophical and technical shift inside a few simple words. But what struck me wasn’t that the question was hard. It’s that it was everywhere from investor meetings to product demos, even casual conversations with my family. And so I decided to answer it in the most human way possible: with a story. everywhere Let me take you on a journey not just to clarify the difference, but to help you feel why it matters, with some example scenarios. Let’s imagine two cars. The first is a remote-controlled toy. Cute, functional, and responds to commands. You press left, it goes left. Push forward, it rolls ahead. It's predictable, simple, and entertaining in short bursts. Now picture a real car. But not just any car. This one knows where you want to go. It reads road signs. It senses traffic. It can reroute, park itself, even anticipate a coffee stop if it notices you're tired. The toy car is a chatbot. The real car? That’s an AI agent. Same idea of “motion,” completely different degree of autonomy, awareness, and intelligence. And yet, both are powered by an engine much like both chatbots and AI agents are powered by large language models (LLMs). What makes the difference is what's built around that engine. what's built around that engine. So What is a Chatbot, Really? The Helpful Machines That Never Grew Up? So What is a Chatbot, Really? If you’ve ever interacted with a customer support window that greeted you cheerfully—only to send you into an endless loop of “Press 1 for billing” or “Please rephrase that”—then you already know what a chatbot feels like. They’re like vending machines with a friendly face. You press a button (or type a query), and they try to serve you something that fits. Under the hood, chatbots are built on predefined scripts or rule-based flows. Some of the more modern ones might be powered by lightweight language models, but the logic is still fundamentally reactive. They recognize keywords, follow a decision tree, and hand you the most relevant pre-programmed response. And you know what? For a long time, that was enough. They did their jobs well—helping users reset passwords, book tickets, check account balances, and reduce call center load. In fact, they’ve become ubiquitous in industries from banking to e-commerce. But here’s the thing: they’re not built to think. They don’t learn from your past interactions. They don’t remember what you said last week—or even two minutes ago. They can’t recognize nuance or context, let alone emotion. If you deviate from their carefully structured path, the conversation crumbles. Imagine you're on a hospital website. You type: “I’m having abdominal cramps and irregular bowel movements. What should I do?” A chatbot might respond politely with: A chatbot “I’m sorry to hear that. Would you like to book an appointment? Here are some FAQs.” Let’s say you type: “Hey, I’ve been rescheduling my therapy sessions every week for the past month because I’ve been too anxious to talk. I think I need help.” “Hey, I’ve been rescheduling my therapy sessions every week for the past month because I’ve been too anxious to talk. I think I need help.” The chatbot might say: “Would you like to schedule an appointment?” “Would you like to schedule an appointment?” That’s a response. response But what you really needed was recognition. Context. Compassion. And the chatbot can’t offer that, not because it’s flawed, but because it was never designed to. What Makes an AI Agent Different? What Makes an AI Agent Different? Now enter the AI agent. TIf a chatbot is a vending machine, then an AI agent is more like a trusted colleague—one who not only listens but learns, adapts, and acts on your behalf, sometimes even before you know what you need. Unlike chatbots that live in the moment, AI agents remember. They carry context. They recognize patterns across time, understand your preferences, anticipate your needs, and—critically—execute. At the heart of both chatbots and agents lies the same foundational engine: large language models like GPT-4 or Claude. But what elevates an agent is what wraps around that engine—agency. Agency is the ability to: Make decisions
Plan ahead
Call tools
Access external data
Collaborate with other systems or agents
Follow through on goals—not just answer queries Make decisions Plan ahead Call tools Access external data Collaborate with other systems or agents Follow through on goals—not just answer queries Let me show you how that plays out in a real-world scenario. Imagine you're recovering from surgery. It's been a difficult journey—pain, fatigue, inconsistent sleep, digestive issues. You open an app and say: “I didn’t sleep well last night, and I’ve had severe bloating since lunch.” “I didn’t sleep well last night, and I’ve had severe bloating since lunch.” “I didn’t sleep well last night, and I’ve had severe bloating since lunch.” A chatbot might respond with: “Would you like to schedule a consultation?” “Would you like to schedule a consultation?” “Would you like to schedule a consultation?” Polite. Functional. But limited. Now let’s say the same interface is powered by an AI agent. Instead of offering an appointment, it draws from your symptom history, your dietary logs, and recent microbiome analysis. Then it replies: “It looks like last night’s meal triggered a reaction similar to your March flare-up. I’ve updated your nutrition plan and sent your log to your clinician. Want me to monitor your sleep pattern this week and adjust your hydration levels?” “It looks like last night’s meal triggered a reaction similar to your March flare-up. I’ve updated your nutrition plan and sent your log to your clinician. Want me to monitor your sleep pattern this week and adjust your hydration levels?” “It looks like last night’s meal triggered a reaction similar to your March flare-up. I’ve updated your nutrition plan and sent your log to your clinician. Want me to monitor your sleep pattern this week and adjust your hydration levels?” Let’s take it a step further. You're browsing a hospital website late at night, feeling anxious. You type: “I’m having abdominal cramps and irregular bowel movements. What should I do?” “I’m having abdominal cramps and irregular bowel movements. What should I do?” “I’m having abdominal cramps and irregular bowel movements. What should I do?” The chatbot, once again, means well: “I’m sorry to hear that. Would you like to book an appointment? Here are some FAQs.” “I’m sorry to hear that. Would you like to book an appointment? Here are some FAQs.” “I’m sorry to hear that. Would you like to book an appointment? Here are some FAQs.” It’s trying. But it’s still just handing you options. An agent, by contrast, sees the whole picture. It connects the dots from your symptom history, flags potential risks, and acts accordingly. Its response? “Based on your past symptoms and logs, it looks like your condition might be flaring. I’ve alerted your primary physician and updated your food sensitivity data. Should I go ahead and schedule a follow-up for tomorrow or notify your caregiver?” “Based on your past symptoms and logs, it looks like your condition might be flaring. I’ve alerted your primary physician and updated your food sensitivity data. Should I go ahead and schedule a follow-up for tomorrow or notify your caregiver?” “Based on your past symptoms and logs, it looks like your condition might be flaring. I’ve alerted your primary physician and updated your food sensitivity data. Should I go ahead and schedule a follow-up for tomorrow or notify your caregiver?” One offers links. The other initiates care. This distinction is foundational. Especially in my field, where I’m building agentic systems to support cancer survivors managing complex post-treatment syndromes like LARS. These are deeply personal, often invisible challenges that can’t be solved with drop-down menus or rule-based trees. What our users need isn’t another bot that talks at them. They need an intelligent companion that can walk with them, think with them, and sometimes, even actfor them. for This is the moment conversational AI is growing up. And agents are leading that evolution. A Real-Life Use Case: Patient Support in Oncology When AI Becomes a Colleague A Real-Life Use Case: Patient Support in Oncology When AI Becomes a Colleague The shift from chatbot to agent is the shift from interface to partner. Let me bring you into my world for a moment. I’m building a healing system for colorectal cancer survivors—people navigating post-treatment challenges that are often invisible, unpredictable, and deeply personal. A chatbot might check symptoms. An AI agent might: Interpret a week’s worth of symptom logs,
Recognize subtle deterioration patterns,
Forecast risk escalation using predictive models,
Suggest interventions,
And connect the survivor with their clinician before things spiral. Interpret a week’s worth of symptom logs, Recognize subtle deterioration patterns, Forecast risk escalation using predictive models, Suggest interventions, And connect the survivor with their clinician before things spiral. before This isn’t hypothetical. This is already in motion. And this is why I believe the difference between chatbot and agent is not just a technical evolution—it’s a human one. The Technology Behind the Curtain The Technology Behind the Curtain So what makes agents possible? It’s not just one thing. It’s the orchestration of many moving parts: LLMs provide the language understanding and generation.
Tool use allows agents to interact with APIs, databases, and applications.
Memory helps retain facts about you and recall them in future interactions.
Reasoning and planning enable the agent to break down tasks and act step by step.
Multi-agent coordination (like with CrewAI or AutoGen) allows multiple agents to collaborate toward a shared goal. LLMs provide the language understanding and generation. Tool use allows agents to interact with APIs, databases, and applications. Memory helps retain facts about you and recall them in future interactions. Reasoning and planning enable the agent to break down tasks and act step by step. Multi-agent coordination (like with CrewAI or AutoGen) allows multiple agents to collaborate toward a shared goal. Frameworks like LangChain help engineers design these systems modularly stringing together LLMs, retrievers, memory components, agents, and external tools in what we now call “chains.” LangChain Why the Distinction Matters More Than Ever Why the Distinction Matters More Than Ever We’re in a pivotal moment in tech. People no longer just want answers. They want solutions. They want systems that understand, adapt, and act. And that’s exactly where agents come in. As a data scientist, founder, and someone building mission-critical AI in healthcare, I believe we’re entering the agentic era where chatbots become the entry point, but agents define the experience. This shift also means our roles are evolving. If you're a: Product manager, you’ll need to learn about agent behavior loops and goal-driven design.
Software engineer, you’ll start architecting environments for autonomous flows, not just REST APIs.
Data scientist, your models will be called upon by agents not just humans and will need to work inside multi-step reasoning pipelines. Product manager, you’ll need to learn about agent behavior loops and goal-driven design. Software engineer, you’ll start architecting environments for autonomous flows, not just REST APIs. Data scientist, your models will be called upon by agents not just humans and will need to work inside multi-step reasoning pipelines. Agents will not replace us, but they will work with us. And maybe, one day, they’ll even anticipate what we need before we do. The future of AI is not in better answers. It’s in better actions. The future of AI is not in better answers. It’s in better actions. Let’s Keep the Conversation Going Let’s Keep the Conversation Going I’m currently building intelligent healing agents for cancer recovery and quality-of-life optimization. If this space excites you, whether you’re an investor, a colorectal cancer surgeon, an AI engineer, researcher, or survivor yourself, Let’s connect.

Chatbot vs AI Agent: The Difference Everyone Talks Around but Rarely Gets Right

Today, Skype officially bids farewell. If you're still using it, you're part of a rare group. For most of us, the blue "S" has already faded from our desktops and memories. And yet, as it disappears into the archives of internet history, it's worth taking a moment to remember just how deeply Skype once shaped the way we communicated. A Revolution That Started with Voice When Skype launched in 2003, it wasn't just another chat app. It was a revolution in internet communication. Built by the team behind Kazaa, Skype was one of the first applications that made internet voice calls feel both simple and reliable. It didn't require tech know-how, expensive gear, or even much bandwidth. All you needed was a computer and a mic, and you could talk to someone across the planet as if they were in the next room. For international students, long-distance couples, digital nomads, and remote workers long before they were called that, Skype was a lifeline. It saved money. It brought people closer. And for many, it brought the future into the present. Skype was also ahead of its time in many ways. It introduced peer-to-peer technology into everyday communication, giving it a robust and scalable infrastructure that could handle millions of users with relatively low latency. It was lean, fast, and free when phone calls were still expensive. Suddenly, you didn’t need to explain where you were calling from. You could just talk. From Startup Darling to Microsoft Property In 2005, eBay bought Skype for $2.6 billion. It didn’t really know what to do with it. The synergy wasn't there. In 2011, Microsoft stepped in with an $8.5 billion acquisition, folding Skype into its software suite and integrating it with Windows, Xbox, and Office. Microsoft saw Skype as a foundational piece for the future of communication. There were plans to integrate it into enterprise workflows and consumer products alike. Skype for Business was supposed to become the gold standard for online meetings, outshining even its own predecessors like Lync. For a while, Skype was still prominent. It became the default tool for business meetings, online tutoring, podcast interviews, and even family reunions. It had name recognition. It had global reach. It was synonymous with calling someone online. If you had a webcam and a microphone, you probably had Skype installed. But it also had something else creeping in: bloat. But it also had something else creeping in: bloat. A Slow Decline Skype's interface grew heavier. New features were added without a clear purpose. Updates felt less like improvements and more like interruptions. The app, once known for its clean utility, began to feel cluttered. Users found themselves struggling with confusing menus, inconsistent notifications, and unexpected crashes. Meanwhile, competitors were moving faster and cleaner. Zoom came out of nowhere and quickly became the de facto standard for online meetings. WhatsApp and FaceTime dominated mobile. Discord took over for casual voice and video. Even Microsoft Teams began to overshadow Skype within Microsoft's own walls. Skype's decline wasn’t an overnight failure. It was a slow fade from relevance. Some blame mismanagement. Others point to missed opportunities, like not capitalizing on mobile-first strategies or integrating better with browsers. But in the end, it was a classic story of software entropy. The world evolved, and Skype didn't evolve fast enough. The Final Sign-Off Microsoft has now made it official. Skype will be retired. Support will wind down, and updates will stop. The servers will eventually go dark. For many, this moment passed without notice. The app that once defined digital communication has quietly become a footnote. This is more than a product phase-out. It's the end of an era. Skype was part of internet culture. It was there during pivotal global events, it kept families in touch across continents, and it let indie creators collaborate in real-time. It was the medium through which countless first remote job interviews were conducted. It was often the first app someone would install after setting up a new laptop. But it's worth remembering that Skype walked so the rest could run. Before Zoom fatigue was even a phrase, Skype gave us the power to connect across oceans, borders, and time zones. It helped people stay close when the world felt far apart. So, What Comes Next? The landscape of digital communication keeps shifting. AI-powered assistants, spatial audio, real-time translation, and virtual reality meetings are becoming part of our new reality. And yet, even as we step into this more immersive and complex era, there's a strange comfort in simplicity. Skype never aimed to be flashy. It just worked. Skype’s legacy isn’t measured by the features it offered in its final years, but by the standard it set when it was at its best. It taught us what internet calling could be. It brought the world closer together long before remote work and virtual classrooms became the norm. We may not miss the glitches, the confusing redesigns, or the endless sign-ins. But we will miss what Skype represented: the possibility of a more connected world, made accessible with just a username and a smile on a grainy webcam. It didn’t just start a trend. It started a way of life. So thank you, Skype. You weren’t perfect, but you mattered. And today, we say goodbye. Have memories of using Skype? Share them in the comments or tag us with your favorite Skype story. Have memories of using Skype? Share them in the comments or tag us with your favorite Skype story. Have memories of using Skype? Share them in the comments or tag us with your favorite Skype story.

Goodbye, Skype: A Quiet Sign-Off for a Communication Icon

Reading has always been a source of enjoyment and frustration for me. Recently AI lit an even larger “What For” sign over it (Spoiler: there’s a good answer) To Read: An Arrogant Fool’s Pursuit? I have always been an avid reader. My parents had a floor-to-ceiling library (which they read). My elementary school and hometown libraries were real places we went to. In my 20s I shifted from thrillers and sci-fi to the classics, modernists and later to non-fiction. I used to read at a pace of about a book per month. they read In a clever 2017 survey, Emily Temple analyzed reading trends in the US and broke down the results into three categories: the average reader (about 12 books a year), the voracious reader (50 books a year) and the super reader (80 books a year). She was interested in how many books a person will read in a lifetime–but as I will soon argue–I think she was asking the wrong question. 2017 survey My “Emily Temple” reader-group allocation changed when solitary sports activities started consuming longer hours in my 30s. I discovered Audible and later Delphin and sped up to around 2 to 4 books per week (depending on run/swim mileage). Delphin A quick calculation puts me somewhere between 1,000 to 2,000 books to date and in the reputable club of Bill Gates (50/year), Barack Obama (30-50/year) and Mark Zuckerberg (30/year), while Buffett, Munger and Roosevelt (hundreds of books per year) smile at us from the Olympus. But here comes the question: What the heck for? My often honest answer? Little traceable value. If you asked about most of them, I probably recall the key concept, if that. The detailed arguments, supporting evidence, even important points—most faded away. This is a troubling realization. If so much of what we read evaporates from memory, is reading worth the effort? Is it different from passing time entertaining on Instagram and Netflix, only with an aura of high-brow activity? Taken a step further – if once reading built a “map of the world” to better navigate life, does the new all-knowing AI make it as obsolete as navigation using a map in the age of GPS? A Negligible Human Activity Turned Craze In terms of human history, reading is a last-minute thing. Until around 1800, only 1 in 10 people worldwide could read (better numbers in Europe) becoming 9 out of 10 today (with most growth occurring in the 20th century). Although recent years have seen a decline in the time spent reading for pleasure (from 23 minutes to 16 minutes), a growing number of people that do not read at all skew these numbers. Those who do, keep at it—for over an hour a day. Today, more books are being printed than ever before. This is one field where innovation is in slow adoption. Though in existence for decades, e-books and audiobooks take a small portion of the market, each about 10%. Their use case and benefits are not at the order of magnitude required for a “revolution”. As for content, from the introduction of Gutenberg’s movable press and later improvements, new book categories emerged from the original religious, scholarly and bureaucratic works. Romance, thrillers, sci-fi, fantasy, children and of course non-fiction. This last category–which interests us most in the context of this article–occupies a varying portion of the market in different cultures. Non-fiction accounts for almost 50% of book sales in the US, compared to as little as 6% in the UK (10% globally). From Strategic Forgetting to Trained Models So many people, books, knowledge, reading and so much forgetting. Are we all mistaking entertainment for education? Yes and no. The truth about the theories, anecdotes and facts we consume is that they mostly fade from memory. Looking at Hermann Ebbinghaus’ Forgetting Curve we can see that most of what we encounter (and consume) fades away. The good news is that it is not a bug. It is a feature. Forgetting improves our cognitive efficiency (freeing “CPU”), maintains our emotional health (with the assistance of active forgetting during sleep), keeps our minds agile, enables creativity and most importantly for our discussion – fosters prioritization and generalization. Hermann Ebbinghaus’ Forgetting Curve active forgetting Psychologists distinguish between explicit memory (facts we can recall) and implicit memory (knowledge that shapes our thinking without direct recall). Reading, even when forgotten, alters our cognitive structures. It builds mental models, strengthens associative thinking and expands our perspectives. explicit memory implicit memory If this sounds familiar from a recent article you read about AI, it might not be a coincidence. There are similarities between a “trained AI model” and a “well-read human” (leaving “intelligence” aside as I think it has little to do with either). The more information you feed both – the better they become at generalizing about the world. But the similarities should not be over stretched. A trained AI model “remembers” more details as a byproduct of being trained on real text while the human brain does not. It only retains the “trainness”. a recent article AI Reading Overtaking Humans? Not So Fast. Hold on. If AI’s reading is improving its generalization abilities (implicit memory), similar to humans and simultaneously emerging its know-it-all capabilities (explicit memory), where does it leave us? The trash-heap of evolution? emerging Probably not. Because AI’s reading byproducts (generalizing, logic, facts) are not accompanied by motivation, curiosity, initiative, necessity and invention. AI might have answers, but it does not know what to ask. Even with details completely forgotten, or maybe because of it, a well-read person will be in an interesting position when coming to work with an AI. Humans possessing innate drives and a trained worldview, will more effectively harness an all-knowing logic machine. A trained human gets better results using Google Maps because what counts is not the ability to read a map, use a compass, count steps and remember road numbers. What counts is having a will to go, knowing what you look for and being able to recognize a good location. Not On The Same Book There is one more thing we ignored until now. AIs do not read the same books humans do. The reason is most books are proprietary. AIs are not allowed to read them for training (copyright for personal and not commercial use). Let’s assume for a minute this is the case. We have a new paradox. Humans train their mental models on books and then forget the details, while AIs train on public data, building a model that retains different details. How can a human with a model and motivation consult a tool on specifics it does not know? this is the case That is a question that remains to be answered…

Stories, News & Mentions about SMILE

Categories

Trending Topics

Stories, News & Mentions about SMILE

Smile’s Stories on HackerNoon

Categories

Trending Topics