Weston Urban has assembled in recent years swaths of property in northwest downtown with plans to change the landscape there.

As Weston Urban’s holdings in downtown San Antonio grow, the developer focuses on housing

Bolt's Ardo Reinsalu delves into how the company's micro-mobility vehicles embody the revolution in urban navigation and city interaction.

Uncovering the hidden potential of Bolt’s scooters: There’s more than meets the eye

But ground assaults in urban environments are extremely risky for any fighting force. Reserve Major General Yair Golan, who has led troops into battle in Gaza many times during his military career ...

Israeli military faces challenging urban warfare in Gaza

The Balfour Beatty group is a consortium comprising Superior Construction Co. Southeast; Beep Inc.; WGI Inc.; Stantec Consulting Services Inc.; Miller Electric; Urban SDK; and Balfour Beatty. In January 2022, JTA signed a $49 million contract with the ...

U2C building, maintenance facility gets review board’s conceptual approval

PC Amelia Shearer, formerly of Cleveland Police, was found guilty of gross misconduct over the incident in Urban Outfitters' York branch in 2021. The disciplinary panel in 2022 did not recommend that PC Shearer, who was then 24, be sacked. A Cleveland ...

Vetting review prompts urinating PC to quit

The Balfour Beatty group is a consortium comprising Superior Construction Co. Southeast; Beep Inc.; WGI Inc.; Stantec Consulting Services Inc.; Miller Electric; Urban SDK; and Balfour Beatty. The application says JTA intends to start construction Dec. 1 ...

JTA’s U2C operations, maintenance building design set for city review board

OpenAI has unveiled the Beta version of its Python SDK, marking a significant step towards enhancing access to the OpenAI API for Python developers. The OpenAI Python library offers a simplified way for Python-based applications to interact with the OpenAI ...

OpenAI to Release Its Python SDK

Sisense today unveiled a public preview of the Compose SDK for Fusion to help empower developers to build insights-driven products. With over 15 years as a leader in embedded analytics, Sisense now enters a new developer-first era with more developer ...

Sisense Unveils Compose SDK for Fusion

PDF SDK Market is expected to grow at the Same Pace in FY30, Largely Driven by Volumes, Moderating Rates and expected to reach XX million US dollars, with a CAGR of xx% by 2030. The global PDF SDK ...

PDF SDK Market (New Report): Global Research Strategic Analysis 2030

VideoCX.ios new Multiuser Video Conferencing Workflow opens many new use cases for the BFSI industry, such as credit verification for housing loans with multiple applicants in insurance, where the customer,

VideoCX.io (earlier known as WorkApps) launches multiuser video conferencing workflows for the BFSI industry with Amazon Chime SDK

Urban SDK

urbansdk

This paper is under CC 4.0 license. available on arxiv Authors: (1) Joel Jang, CarperAI,University of Washington & Allen Institute for AI; (2) Seungone Kim, KAIST AI; (3) Yizhong Wang, University of Washington; (4) Jack Hessel, University of Washington; (5) Luke Zettlemoyer, Aleph Alpha; (6) Hannaneh Hajishirzi, University of Washington & Allen Institute for AI; (7) Yejin Choi, UC San Diego. 5 CONCLUSION Previous work has shown that adapting LLMs with RLHF helps them generate outputs that are preferred by humans over the supervised fine-tuned counterpart. However, recent work has also pointed out that simply training LLMs to abide by the preference of the general may result in ignoring individual preferences and values. In this work, we provide the first steps to tackle this issue by proposing Reinforcement Learning from Personalized Human Feedback as a multi-objective problem so that LLMs can be aligned to follow conflicting preferences. We propose a promising method called P-SOUPS that is able to composite models trained on single objectives on the fly during inference. We also highlight the scalability of P-SOUPS by showing that it scales linearly, instead of exponentially like the MORL baseline, with regards to the number of new preferences, which is required to provide true personalization to individual users. ACKNOWLEDGMENTS Thanks to Minyoung Hwang, Sungdong Kim, Tim Dettmers, Yoonjoo Lee, and Margaret Li for helpful feedback and discussion. REFERENCES Xiang Ao, Xiting Wang, Ling Luo, Ying Qiao, Qing He, and Xing Xie. PENS: A dataset and generic framework for personalized news headline generation. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pp. 82–92, Online, August 2021. Association for Computational Linguistics. doi: 10.18653/v1/2021.acl-long.7. URL https: //aclanthology.org/2021.acl-long.7. Yuntao Bai, Andy Jones, Kamal Ndousse, Amanda Askell, Anna Chen, Nova DasSarma, Dawn Drain, Stanislav Fort, Deep Ganguli, T. J. Henighan, Nicholas Joseph, Saurav Kadavath, John Kernion, Tom Conerly, Sheer El-Showk, Nelson Elhage, Zac Hatfield-Dodds, Danny Hernandez, Tristan Hume, Scott Johnston, Shauna Kravec, Liane Lovitt, Neel Nanda, Catherine Olsson, Dario Amodei, Tom B. Brown, Jack Clark, Sam McCandlish, Christopher Olah, Benjamin Mann, and Jared Kaplan. Training a helpful and harmless assistant with reinforcement learning from human feedback. ArXiv, abs/2204.05862, 2022a. URL https://api.semanticscholar.org/ CorpusID:248118878. Yuntao Bai, Saurav Kadavath, Sandipan Kundu, Amanda Askell, Jackson Kernion, Andy Jones, Anna Chen, Anna Goldie, Azalia Mirhoseini, Cameron McKinnon, et al. Constitutional ai: Harmlessness from ai feedback. arXiv preprint arXiv:2212.08073, 2022b. Stephen Casper, Xander Davies, Claudia Shi, Thomas Krendl Gilbert, Jer´ emy Scheurer, Javier ´ Rando, Rachel Freedman, Tomasz Korbak, David Lindner, Pedro Freire, et al. Open problems and fundamental limitations of reinforcement learning from human feedback. arXiv preprint arXiv:2307.15217, 2023. Deborah Cohen, Moonkyung Ryu, Yinlam Chow, Orgad Keller, Ido Greenberg, Avinatan Hassidim, Michael Fink, Yossi Matias, Idan Szpektor, Craig Boutilier, et al. Dynamic planning in openended dialogue using reinforcement learning. arXiv preprint arXiv:2208.02294, 2022. Shachar Don-Yehiya, Elad Venezian, Colin Raffel, Noam Slonim, Yoav Katz, and Leshem Choshen. Cold fusion: Collaborative descent for distributed multitask finetuning. arXiv preprint arXiv:2212.01378, 2022. Yann Dubois, Xuechen Li, Rohan Taori, Tianyi Zhang, Ishaan Gulrajani, Jimmy Ba, Carlos Guestrin, Percy Liang, and Tatsunori B Hashimoto. Alpacafarm: A simulation framework for methods that learn from human feedback. arXiv preprint arXiv:2305.14387, 2023. Shangbin Feng, Chan Young Park, Yuhan Liu, and Yulia Tsvetkov. From pretraining data to language models to downstream tasks: Tracking the trails of political biases leading to unfair nlp models. arXiv preprint arXiv:2305.08283, 2023. Xinyang Geng, Arnav Gudibande, Hao Liu, Eric Wallace, Pieter Abbeel, Sergey Levine, and Dawn Song. Koala: A dialogue model for academic research. Blog post, April, 1, 2023. Conor F Hayes, Roxana Radulescu, Eugenio Bargiacchi, Johan K ˘ allstr ¨ om, Matthew Macfarlane, ¨ Mathieu Reymond, Timothy Verstraeten, Luisa M Zintgraf, Richard Dazeley, Fredrik Heintz, et al. A practical guide to multi-objective reinforcement learning and planning. Autonomous Agents and Multi-Agent Systems, 36(1):26, 2022. Edward J Hu, yelong shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models. In International Conference on Learning Representations, 2022. URL https://openreview.net/forum? id=nZeVKeeFYf9. Chengsong Huang, Qian Liu, Bill Yuchen Lin, Tianyu Pang, Chao Du, and Min Lin. Lorahub: Efficient cross-task generalization via dynamic lora composition. ArXiv, abs/2307.13269, 2023. URL https://api.semanticscholar.org/CorpusID:260155012. Gabriel Ilharco, Marco Tulio Ribeiro, Mitchell Wortsman, Suchin Gururangan, Ludwig Schmidt, Hannaneh Hajishirzi, and Ali Farhadi. Editing models with task arithmetic. ICLR, 2022. Joel Jang, Seungone Kim, Seonghyeon Ye, Doyoung Kim, Lajanugen Logeswaran, Moontae Lee, Kyungjae Lee, and Minjoon Seo. Exploring the benefits of training expert language models over instruction tuning. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett (eds.), Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machine Learning Research, pp. 14702–14729. PMLR, 23–29 Jul 2023. URL https://proceedings.mlr.press/v202/jang23a. html. Sungdong Kim, Sanghwan Bae, Jamin Shin, Soyoung Kang, Donghyun Kwak, Kang Min Yoo, and Minjoon Seo. Aligning large language models through synthetic feedback. arXiv preprint arXiv:2305.13735, 2023. Hannah Rose Kirk, Bertie Vidgen, Paul Rottger, and Scott A Hale. Personalisation within bounds: A ¨ risk taxonomy and policy framework for the alignment of large language models with personalised feedback. arXiv preprint arXiv:2303.05453, 2023. Noriyuki Kojima, Alane Suhr, and Yoav Artzi. Continual learning for grounded instruction generation by observing human following behavior. Transactions of the Association for Computational Linguistics, 9:1303–1319, 2021. Margaret Li, Suchin Gururangan, Tim Dettmers, Mike Lewis, Tim Althoff, Noah A Smith, and Luke Zettlemoyer. Branch-train-merge: Embarrassingly parallel training of expert language models. arXiv preprint arXiv:2208.03306, 2022. Pan Li and Alexander Tuzhilin. Towards controllable and personalized review generation. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pp. 3237–3245, Hong Kong, China, November 2019. Association for Computational Linguistics. doi: 10.18653/v1/D19-1319. URL https://aclanthology.org/D19-1319. Bodhisattwa Prasad Majumder, Shuyang Li, Jianmo Ni, and Julian McAuley. Generating personalized recipes from historical user preferences. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pp. 5976–5982, Hong Kong, China, November 2019. Association for Computational Linguistics. doi: 10.18653/v1/D19-1613. URL https://aclanthology.org/D19-1613. Pierre-Emmanuel Mazare, Samuel Humeau, Martin Raison, and Antoine Bordes. Training mil- ´ lions of personalized dialogue agents. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pp. 2775–2779, Brussels, Belgium, OctoberNovember 2018. Association for Computational Linguistics. doi: 10.18653/v1/D18-1298. URL https://aclanthology.org/D18-1298. Jacob Menick, Maja Trebacz, Vladimir Mikulik, John Aslanides, Francis Song, Martin Chadwick, Mia Glaese, Susannah Young, Lucy Campbell-Gillingham, Geoffrey Irving, et al. Teaching language models to support answers with verified quotes. arXiv preprint arXiv:2203.11147, 2022. Reiichiro Nakano, Jacob Hilton, S. Arun Balaji, Jeff Wu, Ouyang Long, Christina Kim, Christopher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, Xu Jiang, Karl Cobbe, Tyna Eloundou, Gretchen Krueger, Kevin Button, Matthew Knight, Benjamin Chess, and John Schulman. Webgpt: Browser-assisted question-answering with human feedback. ArXiv, abs/2112.09332, 2021a. URL https://api.semanticscholar.org/CorpusID:245329531. Reiichiro Nakano, Jacob Hilton, Suchir Balaji, Jeff Wu, Long Ouyang, Christina Kim, Christopher Hesse, Shantanu Jain, Vineet Kosaraju, William Saunders, et al. Webgpt: Browser-assisted question-answering with human feedback. arXiv preprint arXiv:2112.09332, 2021b. Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke E. Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Francis Christiano, Jan Leike, and Ryan J. Lowe. Training language models to follow instructions with human feedback. ArXiv, abs/2203.02155, 2022a. URL https://api.semanticscholar.org/ CorpusID:246426909. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, 35: 27730–27744, 2022b. Baolin Peng, Chunyuan Li, Pengcheng He, Michel Galley, and Jianfeng Gao. Instruction tuning with gpt-4. arXiv preprint arXiv:2304.03277, 2023. Jonas Pfeiffer, Naman Goyal, Xi Victoria Lin, Xian Li, James Cross, Sebastian Riedel, and Mikel Artetxe. Lifting the curse of multilinguality by pre-training modular transformers. arXiv preprint arXiv:2205.06266, 2022. Alexandre Rame, Guillaume Couairon, Mustafa Shukor, Corentin Dancette, Jean-Baptiste Gaya, Laure Soulier, and Matthieu Cord. Rewarded soups: towards pareto-optimal alignment by interpolating weights fine-tuned on diverse rewards. arXiv preprint arXiv:2306.04488, 2023. Alireza Salemi, Sheshera Mysore, Michael Bendersky, and Hamed Zamani. Lamp: When large language models meet personalization. arXiv preprint arXiv:2304.11406, 2023. Shibani Santurkar, Esin Durmus, Faisal Ladhak, Cinoo Lee, Percy Liang, and Tatsunori Hashimoto. Whose opinions do language models reflect? arXiv preprint arXiv:2303.17548, 2023. John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347, 2017. Umer Siddique, Paul Weng, and Matthieu Zimmer. Learning fair policies in multi-objective (deep) reinforcement learning with average and discounted rewards. In International Conference on Machine Learning, pp. 8905–8915. PMLR, 2020. David Silver, Satinder Singh, Doina Precup, and Richard S Sutton. Reward is enough. Artificial Intelligence, 299:103535, 2021. Prasann Singhal, Tanya Goyal, Jiacheng Xu, and Greg Durrett. A long way to go: Investigating length correlations in rlhf. arXiv preprint arXiv:2310.03716, 2023. Feifan Song, Bowen Yu, Minghao Li, Haiyang Yu, Fei Huang, Yongbin Li, and Houfeng Wang. Preference ranking optimization for human alignment. arXiv preprint arXiv:2306.17492, 2023. Nisan Stiennon, Long Ouyang, Jeffrey Wu, Daniel Ziegler, Ryan Lowe, Chelsea Voss, Alec Radford, Dario Amodei, and Paul F Christiano. Learning to summarize with human feedback. Advances in Neural Information Processing Systems, 33:3008–3021, 2020. Alane Suhr and Yoav Artzi. Continual learning for instruction following from realtime feedback. arXiv preprint arXiv:2212.09710, 2022. Richard S Sutton and Andrew G Barto. Reinforcement learning: An introduction. MIT press, 2018. Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, et al. Lamda: Language models for dialog applications. arXiv preprint arXiv:2201.08239, 2022. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothee´ Lacroix, Baptiste Roziere, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and ` efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023. Peter Vamplew, Richard Dazeley, Cameron Foale, Sally Firmin, and Jane Mummery. Humanaligned artificial intelligence is a multiobjective problem. Ethics and Information Technology, 20:27–40, 2018. Kristof Van Moffaert and Ann Nowe. Multi-objective reinforcement learning using sets of pareto ´ dominating policies. The Journal of Machine Learning Research, 15(1):3483–3512, 2014. Kristof Van Moffaert, Madalina M Drugan, and Ann Nowe. Scalarized multi-objective reinforce- ´ ment learning: Novel design techniques. In 2013 IEEE symposium on adaptive dynamic programming and reinforcement learning (ADPRL), pp. 191–199. IEEE, 2013. Yizhong Wang, Hamish Ivison, Pradeep Dasigi, Jack Hessel, Tushar Khot, Khyathi Raghavi Chandu, David Wadden, Kelsey MacMillan, Noah A Smith, Iz Beltagy, et al. How far can camels go? exploring the state of instruction tuning on open resources. arXiv preprint arXiv:2306.04751, 2023. Mitchell Wortsman, Suchin Gururangan, Shen Li, Ali Farhadi, Ludwig Schmidt, Michael Rabbat, and Ari S Morcos. lo-fi: distributed fine-tuning without communication. arXiv preprint arXiv:2210.11948, 2022a. Mitchell Wortsman, Gabriel Ilharco, Samir Ya Gadre, Rebecca Roelofs, Raphael Gontijo-Lopes, Ari S Morcos, Hongseok Namkoong, Ali Farhadi, Yair Carmon, Simon Kornblith, et al. Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time. In International Conference on Machine Learning, pp. 23965–23998. PMLR, 2022b. Jeff Wu, Long Ouyang, Daniel M Ziegler, Nisan Stiennon, Ryan Lowe, Jan Leike, and Paul Christiano. Recursively summarizing books with human feedback. arXiv preprint arXiv:2109.10862, 2021a. Yuwei Wu, Xuezhe Ma, and Diyi Yang. Personalized response generation via generative split memory network. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 1956–1970, 2021b. Zeqiu Wu, Yushi Hu, Weijia Shi, Nouha Dziri, Alane Suhr, Prithviraj Ammanabrolu, Noah A Smith, Mari Ostendorf, and Hannaneh Hajishirzi. Fine-grained human feedback gives better rewards for language model training. arXiv preprint arXiv:2306.01693, 2023. Jie Xu, Yunsheng Tian, Pingchuan Ma, Daniela Rus, Shinjiro Sueda, and Wojciech Matusik. Prediction-guided multi-objective reinforcement learning for continuous robot control. In International conference on machine learning, pp. 10607–10616. PMLR, 2020. Jing Xu, Arthur Szlam, and Jason Weston. Beyond goldfish memory: Long-term open-domain conversation. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 5180–5197, Dublin, Ireland, May 2022. Association for Computational Linguistics. doi: 10.18653/v1/2022.acl-long.356. URL https: //aclanthology.org/2022.acl-long.356. Runzhe Yang, Xingyuan Sun, and Karthik Narasimhan. A generalized algorithm for multi-objective reinforcement learning and policy adaptation. Advances in neural information processing systems, 32, 2019. Saizheng Zhang, Emily Dinan, Jack Urbanek, Arthur Szlam, Douwe Kiela, and Jason Weston. Personalizing dialogue agents: I have a dog, do you have pets too? arXiv preprint arXiv:1801.07243, 2018. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, et al. Judging llm-as-a-judge with mt-bench and chatbot arena. arXiv preprint arXiv:2306.05685, 2023. Yinhe Zheng, Guanyi Chen, Minlie Huang, Song Liu, and Xuan Zhu. Personalized dialogue generation with diversified traits. arXiv preprint arXiv:1901.09672, 2019. Daniel M Ziegler, Nisan Stiennon, Jeffrey Wu, Tom B Brown, Alec Radford, Dario Amodei, Paul Christiano, and Geoffrey Irving. Fine-tuning language models from human preferences. arXiv preprint arXiv:1909.08593, 2019. A DETAILS OF EVALUATION SETUP We use a modified version of the GPT4 annotation prompt used by Dubois et al. (2023). We modify the criteria to perform the pairwise evaluation from general to a single preference dimension. We also provide 3 demonstrations: one scenario where there is a tie because both responses do not contain any notion of the preference (e.g. both responses do not show any signs of friendliness), one scenario where there is a clear winner, and one scenario where they are both good, but one is better than the other. Simulated pairwise evaluation We recruited 24 crowd workers for our human evaluation. Figure 5 shows the interface used for human evaluation. We consider both the ‘Tie’ and ‘Both are bad’ options to be Ties. B CRITERIA-WISE EVALUATION The criteria-wise win rate (%) across all the methods are shown in Figure 6. The criteria-wise win rate was calculated by getting the average win-rate of the preference combinations that contained the specific preference dimension. For example, when calculating the criteria-wise win rate of ‘Elementary’, we got the average win-rate of the preference combinations that contained the ‘Elementary’ preference, which includes AAA, AAB, ABA, and ABB. C THE FULL LIST OF EVALUATION PROMPTS The full list of evaluation prompts used in our experiments are provided in Table 5 D DETAILED RESULTS FOR HUMAN EVALUATION AND GPT-4 EVALUATION We provide detailed results (win / loss / tie for each of the preference combinations for our main experimental results. Table 6 shows the GPT4 evaluation and Table 7 shows the human evaluation results. E EXAMPLES OF P-SOUPS TEXT GENERATIONS Table 8 shows empirical examples of the text generated from each preference combination of the 16 preference combination experiments for the same prompt.

Personalized Soups: LLM Alignment Via Parameter Merging - Conclusion & References

This paper is available on arxiv under CC 4.0 license. Authors: (1) Xiaoyu Ai, School of Electrical Engineering & Telecommunications, University of New South Wales, Sydney, NSW 2052, Australia; (2) Robert Malaney, School of Electrical Engineering & Telecommunications, University of New South Wales, Sydney, NSW 2052, Australia. Table of Links Abstract & Introduction System Overview Overcoming the limitation of key reconciliation Final Key Rate Experimental Results Discussion Conclusion & References Appendix VII. CONCLUSION In this work, we have carried out a full-blown analysis and experimental implementation of a Slice Reconciliation scheme applied to a specific CV QKD protocol (with post-selection) under simulated channel conditions anticipated for satelliteto-Earth channels. We have provided the optimal solution for the classical reconciliation process for this CV-QKD protocol in the context of massive parallelisation under the finite key regime. More specifically, we have identified the optimal block length when a large-code block is to be subdivided so as to improve the final secure key rate in bits per second. Although our results were based on a specific CV-QKD protocol and a specific GPU architecture, the type of analysis we have introduced here will apply in general terms a large suite of CV-QKD protocols run over any form of architecture that offers massive parallelisation. Our results, therefore, pave the way to optimal reconciliation system design for a wide range of practical CV-QKD systems that operate in the finite key regime. As the demand on the finite key size grows (better security thresholds), and technology advances lead to larger quantum signalling rates, the importance of optimised multithreaded CV-QKD reconciliation will grow. REFERENCES [1] N. Hosseinidehaj, Z. Babar, R. Malaney, S. X. Ng, and L. Hanzo, “Satellite-Based Continuous-Variable Quantum Communications: Stateof-the-Art and a Predictive Outlook,” IEEE Communications Surveys & Tutorials, vol. 21, no. 1, 881–919, 2018. [2] C. H. Bennett and G. Brassard, “Quantum Cryptography: Public Key Distribution and Coin Tossing,” in Proceedings of IEEE International Conference on Computers, Systems and Signal Processing,, vol. 175, 8, 1984. [3] H. Weinfurter, “Quantum Communication Experiments with Discrete Variables,” Quantum Information: From Foundations to Quantum Technology Applications, 369–381, 2016. [4] L. Gy ¨ongy ¨osi, L. Bacsardi, and S. Imre, “A Survey on Quantum Key Distribution,” Infocommunications Journal, vol. 11, no. 2, 14–21, 2019. [5] S. Pirandola, U. L. Andersen, L. Banchi, M. Berta, D. Bunandar, R. Colbeck, D. Englund, T. Gehring, C. Lupo, C. Ottaviani, et al., “Advances in Quantum Cryptography,” Advances in Optics and Photonics, vol. 12, no. 4, 1012–1236, 2020. [6] B. Korzh, C. C. W. Lim, R. Houlmann, N. Gisin, M. J. Li, D. Nolan, B. Sanguinetti, R. Thew, and H. Zbinden, “Provably Secure and Practical Quantum Key Distribution over 307 km of Optical Fibre,” Nature Photonics, vol. 9, no. 3, 163–168, 2015. [7] T. A. Eriksson, T. Hirano, B. J. Puttnam, G. Rademacher, R. S. Lu´ıs, M. Fujiwara, R. Namiki, Y. Awaji, M. Takeoka, N. Wada, et al., “Wavelength Division Multiplexing of Continuous Variable Quantum Key Distribution and 18.3 Tbit/s Data Channels,” Communications Physics, vol. 2, no. 1, 1–8, 2019. [8] S.-Y. Shen, M.-W. Dai, X.-T. Zheng, Q.-Y. Sun, G.-C. Guo, and Z.- F. Han, “Free-Space Continuous-Variable Quantum Key Distribution of Unidimensional Gaussian Modulation Using Polarized Coherent States in an Urban Environment,” Physical Review A, vol. 100, 012325, 2019. [9] L. Gyongyosi and S. Imre, “Secret Key Rates of Free-Space Optical Continuous-Variable Quantum Key Distribution,” International Journal of Communication Systems, vol. 32, no. 18, e4152, 2019. [10] P. Jouguet, S. Kunz-Jacques, and A. Leverrier, “Long-Distance Continuous-Variable Quantum Key Distribution with a Gaussian Modulation,” Physical Review A, vol. 84, no. 6, 062317, 2011. [11] P. Jouguet, S. Kunz-Jacques, T. Debuisschert, S. Fossier, E. Diamanti, R. All´eaume, R. Tualle-Brouri, P. Grangier, A. Leverrier, P. Pache, et al., “Field Test of Classical Symmetric Encryption with Continuous Variables Quantum Key Distribution,” Optics Express, vol. 20, no. 13, 14030–14041, 2012. [12] Q. Liao, Y. Guo, D. Huang, P. Huang, and G. Zeng, “Long-Distance Continuous-Variable Quantum Key Distribution Using Non-Gaussian State-Discrimination Detection,” New Journal of Physics, vol. 20, no. 2, 023015, 2018. [13] D. Guo, C. He, T. Guo, Z. Xue, Q. Feng, and J. Mu, “Comprehensive High-Speed Reconciliation for Continuous-Variable Quantum Key Distribution,” Quantum Information Processing, vol. 19, no. 9, 1–19, 2020. [14] Y. Zhang, Z. Chen, S. Pirandola, X. Wang, C. Zhou, B. Chu, Y. Zhao, B. Xu, S. Yu, and H. Guo, “Long-Distance Continuous-Variable Quantum Key Distribution over 202.81 km of Fiber,” Physical Review Letters, vol. 125, no. 1, 010502, 2020. [15] A. Leverrier, F. Grosshans, and P. Grangier, “Finite-Size Analysis of a Continuous-Variable Quantum Key Distribution,” Physical Review A, vol. 81, no. 6, 062343, 2010. [16] F. Furrer, T. Franz, M. Berta, A. Leverrier, V. B. Scholz, M. Tomamichel, and R. F. Werner, “Continuous Variable Quantum Key Distribution: Finite-Key Analysis of Composable Security against Coherent Attacks,” Physical Review Letters, vol. 109, no. 10, 100502, 2012. [17] S. Kish, E. Villase˜nor, R. Malaney, K. Mudge, and K. Grant, “Feasibility Assessment for Practical Continuous Variable Quantum Key Distribution over the Satellite-to-Earth Channel,” Quantum Engineering, vol. 2, no. 3, e50, 2020. [18] M. Tomamichel, C. C. W. Lim, N. Gisin, and R. Renner, “Tight FiniteKey Analysis for Quantum Cryptography,” Nature Communications, vol. 3, no. 1, 1–6, 2012. [19] Z. Yuan, A. Plews, R. Takahashi, K. Doi, W. Tam, A. Sharpe, A. Dixon, E. Lavelle, J. Dynes, A. Murakami, et al., “10-Mb/s Quantum Key Distribution,” Journal of Lightwave Technology, vol. 36, no. 16, 3427– 3433, 2018. [20] S.-K. Liao, W.-Q. Cai, W.-Y. Liu, L. Zhang, Y. Li, J.-G. Ren, J. Yin, Q. Shen, Y. Cao, Z.-P. Li, et al., “Satellite-to-Ground Quantum Key Distribution,” Nature, vol. 549, no. 7670, 43–47, 2017. [21] D. Lin, D. Huang, P. Huang, J. Peng, and G. Zeng, “High Performance Reconciliation for Continuous-Variable Quantum Key Distribution with LDPC Code,” International Journal of Quantum Information, vol. 13, no. 02, 1550010, 2015. [22] X. Wang, Y. Zhang, S. Yu, B. Xu, Z. Li, and H. Guo, “Efficient Rate-Adaptive Reconciliation for Continuous-Variable Quantum Key Distribution,” Quantum Information & Computation, vol. 17, no. 13– 14, 1123–1134, 2017. [23] C. Zhou, X. Wang, Y. Zhang, Z. Zhang, S. Yu, and H. Guo, “ContinuousVariable Quantum Key Distribution with Rateless Reconciliation Protocol,” Physical Review Applied, vol. 12, no. 5, 054013, 2019. [24] M. Milicevic, C. Feng, L. M. Zhang, and P. G. Gulak, “Quasi-Cyclic Multi-Edge LDPC Codes for Long-Distance Quantum Cryptography,” NPJ Quantum Information, vol. 4, no. 1, 1–9, 2018. [25] S.-S. Yang, Z.-G. Lu, and Y.-M. Li, “High-Speed Post-Processing in Continuous-Variable Quantum Key Distribution Based on FPGA Implementation,” Journal of Lightwave Technology, vol. 38, no. 15, 3935–3941, 2020. [26] H. Li and Y. Pang, “FPGA-Accelerated Quantum Computing Emulation and Quantum Key Distillation,” IEEE Micro, vol. 41, no. 4, 49–57, 2021. [27] C. Weedbrook, A. M. Lance, W. P. Bowen, T. Symul, T. C. Ralph, and P. K. Lam, “Quantum Cryptography without Switching,” Physical Review Letters, vol. 93, no. 17, 170504, 2004. [28] N. Hosseinidehaj, A. M. Lance, T. Symul, N. Walk, and T. C. Ralph, “Finite-Size Effects in Continuous-Variable Quantum Key Distribution with Gaussian Postselection,” Physical Review A, vol. 101, no. 5, 052335, 2020. [29] D. Dequal, L. T. Vidarte, V. R. Rodriguez, G. Vallone, P. Villoresi, A. Leverrier, and E. Diamanti, “Feasibility of Satellite-to-Ground Continuous-Variable Quantum Key Distribution,” NPJ Quantum Information, vol. 7, no. 1, 1–10, 2020. [30] M. Bloch, A. Thangaraj, and S. W. McLaughlin, “Efficient Reconciliation of Correlated Continuous Random Variables Using LDPC Codes,” arXiv preprint cs/0509041, 2005. [31] P. Jouguet, D. Elkouss, and S. Kunz-Jacques, “High-Bit-Rate Continuous-Variable Quantum Key Distribution,” Physical Review A, vol. 90, no. 4, 042329, 2014. [32] H. Mani, T. Gehring, P. Grabenweger, B. Omer, C. Pacher, and ¨ U. L. Andersen, “Multiedge-Type Low-Density Parity-Check Codes for Continuous-Variable Quantum Key Distribution,” Physical Review A, vol. 103, no. 6, 062419, 2021. [33] H. Imai and S. Hirakawa, “A New Multilevel Coding Method Using Error-Correcting Codes,” IEEE Transactions on Information Theory, vol. 23, no. 3, 371–377, 1977. [34] U. Wachsmann, R. F. Fischer, and J. B. Huber, “Multilevel codes: Theoretical concepts and practical design rules,” IEEE Transactions on Information Theory, vol. 45, no. 5, 1361–1391, 1999. [35] R. D. Wesel, X. Liu, J. M. Cioffi, and C. Komninakis, “Constellation Labeling for Linear Encoders,” IEEE Transactions on Information Theory, vol. 47, no. 6, 2417–2431, 2001. [36] M. Bloch, A. Thangaraj, S. W. McLaughlin, and J.-M. Merolla, “LDPCBased Gaussian Key Reconciliation,” in IEEE Information Theory Workshop-ITW’06 Punta del Este, 116–120, IEEE, 2006. [37] A. Leverrier, R. All´eaume, J. Boutros, G. Z´emor, and P. Grangier, “Multidimensional Reconciliation for a Continuous-Variable Quantum Key Distribution,” Physical Review A, vol. 77, no. 4, 042325, 2008. [38] J. Fiur´aˇsek and N. J. Cerf, “Gaussian Postselection and Virtual Noiseless Amplification in Continuous-Variable Quantum Key Distribution,” Physical Review A, vol. 86, no. 6, 060302, 2012. [39] F. Darema, “The SPMD Model: Past, Present and Future,” in European Parallel Virtual Machine/Message Passing Interface Users’ Group Meeting, 1–1, Springer, 2001. [40] M. Pharr and W. R. Mark, “A SPMD Compiler for High-Performance CPU Programming,” in Innovative Parallel Computing (InPar), 1–13, IEEE, 2012. [41] J. Lodewyck, M. Bloch, R. Garc´ıa-Patr ´on, S. Fossier, E. Karpov, E. Diamanti, T. Debuisschert, N. J. Cerf, R. Tualle-Brouri, S. W. McLaughlin, et al., “Quantum Key Distribution over 25 km with an AllFiber Continuous-Variable System,” Physical Review A, vol. 76, no. 4, 042305, 2007. [42] F. Laudenbach, C. Pacher, C.-H. F. Fung, A. Poppe, M. Peev, B. Schrenk, M. Hentschel, P. Walther, and H. H ¨ubel, “Continuous-Variable Quantum Key Distribution with Gaussian Modulation—The Theory of Practical Implementations,” Advanced Quantum Technologies, vol. 1, no. 1, 1800011, 2018. [43] Y. Polyanskiy, H. V. Poor, and S. Verd ´u, “Channel Coding Rate in the Finite Blocklength Regime,” IEEE Transactions on Information Theory, vol. 56, no. 5, 2307–2359, 2010. [44] D. Slepian and J. Wolf, “Noiseless Coding of Correlated Information Sources,” IEEE Transactions on Information Theory, vol. 19, no. 4, 471–480, 1973. [45] T. Richardson and R. Urbanke, Modern Coding Theory. Cambridge University Press, 2008. [46] X. Ai, R. Malaney, and S. X. Ng, “A Reconciliation Strategy for RealTime Satellite-Based QKD,” IEEE Communications Letters, vol. 24, no. 5, 1062–1066, 2020. [47] S.-Y. Chung, T. J. Richardson, and R. L. Urbanke, “Analysis of SumProduct Decoding of Low-Density Parity-Check Codes Using a Gaussian Approximation,” IEEE Transactions on Information Theory, vol. 47, no. 2, 657–670, 2001. [48] V. A. Chandrasetty and S. M. Aziz, “FPGA Implementation of an LDPC Decoder Using a Reduced Complexity Message Passing Algorithm,” Journal of Networks, vol. 6, no. 1, 36, 2011. [49] A. Leverrier, “Composable Security Proof for Continuous-Variable Quantum Key Distribution with Coherent States,” Physical Review Letters, vol. 114, no. 7, 070501, 2015. [50] C. Lupo, C. Ottaviani, P. Papanastasiou, and S. Pirandola, “ContinuousVariable Measurement-Device-Independent Quantum Key Distribution: Composable Security against Coherent Attacks,” Physical Review A, vol. 97, no. 5, 052327, 2018. [51] N. Hosseinidehaj, N. Walk, and T. C. Ralph, “Optimal Realistic Attacks in Continuous-Variable Quantum Key Distribution,” Physical Review A, vol. 99, no. 5, 052336, 2019. [52] S. Pirandola, “Limits and Security of Free-Space Quantum Communications,” Physical Review Research, vol. 3, no. 1, 013279, 2021. [53] S. Pirandola, “Satellite Quantum Communications: Fundamental Bounds and Practical Security,” Physical Review Research, vol. 3, no. 2, 023130, 2021. [54] J. H. Mathews and K. D. Fink, Numerical Methods Using MATLAB, vol. 4. Pearson Prentice Hall Upper Saddle River, NJ, 2004. [55] A. Leverrier, “Security of Continuous-Variable Quantum Key Distribution via a Gaussian de Finetti Reduction,” Physical Review Letters, vol. 118, no. 20, 200501, 2017. [56] E. Villase˜nor, R. Malaney, K. A. Mudge, and K. J. Grant, “Atmospheric Effects on Satellite-to-Ground Quantum Key Distribution Using Coherent States,” in GLOBECOM 2020-2020 IEEE Global Communications Conference, 1–6, IEEE, 2020. [57] J. Mateo, Efficient Information Reconciliation for Quantum Key Distribution. PhD thesis, Universidad Politecnica de Madrid, 2011. [58] T. Tian, C. R. Jones, J. D. Villasenor, and R. D. Wesel, “Selective Avoidance of Cycles in Irregular LDPC Code Construction,” IEEE Transactions on Communications, vol. 52, no. 8, 1242–1247, 2004. [59] D. V. Nguyen, S. K. Chilappagari, M. W. Marcellin, and B. Vasic, “On the Construction of Structured LDPC Codes Free of Small Trapping Sets,” IEEE Transactions on Information Theory, vol. 58, no. 4, 2280– 2302, 2012. [60] S. Landner and O. Milenkovic, “Algorithmic and Combinatorial Analysis of Trapping Sets in Structured LDPC Codes,” in International Conference on Wireless Networks, Communications and Mobile Computing, vol. 1, 630–635, IEEE, 2005. [61] X. Wang, Y. Zhang, S. Yu, and H. Guo, “High Speed Error Correction for Continuous-Variable Quantum Key Distribution with Multi-Edge Type LDPC Code,” Scientific Reports, vol. 8, no. 1, 1–7, 2018. [62] Y. Li, X. Zhang, Y. Li, B. Xu, L. Ma, J. Yang, and W. Huang, “HighThroughput GPU Layered Decoder of Quasi-Cyclic Multi-Edge Type Low Density Parity Check Codes in Continuous-Variable Quantum Key Distribution Systems,” Scientific Reports, vol. 10, no. 1, 1–11, 2020. [63] F. Grosshans, “Collective Attacks and Unconditional Security in Continuous Variable Quantum Key Distribution,” Physical Review Letters, vol. 94, 020504, 2005. [64] M. Navascu´es, F. Grosshans, and A. Acin, “Optimality of Gaussian Attacks in Continuous-Variable Quantum Cryptography,” Physical Review Letters, vol. 97, no. 19, 190502, 2006. [65] S. Fossier, E. Diamanti, T. Debuisschert, R. Tualle-Brouri, and P. Grangier, “Improvement of Continuous-Variable Quantum Key Distribution Systems by Using Optical Preamplifiers,” Journal of Physics B: Atomic, Molecular and Optical Physics, vol. 42, no. 11, 114014, 2009. [66] P. Jouguet, S. Kunz-Jacques, E. Diamanti, and A. Leverrier, “Analysis of Imperfections in Practical Continuous-Variable Quantum Key Distribution,” Physical Review A, vol. 86, no. 3, 032309, 2012.

Optimised Multithreaded CV-QKD Reconciliation for Global Quantum Networks: Conclusion & References

Authors: (1) Dominic Petrak, UKP Lab, Department of Computer Science, Technical University of Darmstadt, Germany; (2) Nafise Sadat Moosavi, Department of Computer Science, The University of Sheffield, United Kingdom; (3) Ye Tian, Wluper, London, United Kingdom; (4) Nikolai Rozanov, Wluper, London, United Kingdom; (5) Iryna Gurevych, UKP Lab, Department of Computer Science, Technical University of Darmstadt, Germany. Table of Links Abstract & Introduction Related Work Datasets Examined Manual Error Type Analysis and Taxonomies Automatic Filtering for Potentially Relevant Dialogs Statistical Analysis Evaluation and Experiments Discussion Conclusion, Limitation, Acknowledgments, and References A Integrated Error Taxonomy – Details B Error-Indicating Sentences And Phrases C Automatic Filtering – Implementation D Automatic Filtering – Sentence-Level Analysis E Task-Oriented Dialogs – Examples F Effectiveness Of Automatic Filtering – A Detailed Analysis G Inter-Annotator Agreement – Detailed Analysis H Annotation Guidelines I Hyperparameters and Baseline Experiments J Human-Human Dialogs – Examples 9 Conclusion In this work, we examined the dialogs of six datasets from various types, including MultiWoZ, SGD, BABI, PersonaChat, Wizards-of-Wikipedia, and the human-bot split from the Self-Feeding Chatbot, for errors in system utterances and the types of subsequent user responses to assess their extendibility with annotations for learning from free-text human feedback. Our results show that this largely depends on whether the dialogs are human-human or human-bot, and whether they are task-oriented, open-domain, or knowledgegrounded. We found that human-bot dialogs, contain more errors in system utterances that are addressed with free-text human feedback in subsequent user responses, especially in the case of opendomain and knowledge-grounded dialogs. Therefore, it might be feasible to extend these datasets with the needed annotations to support research into methods for learning from free-text human feedback, e.g., by taking advantage of the recent developments in synthetic data generation. We also used the insights gained during this process to propose a new user response type taxonomy and a modified Integrated Error Taxonomy for the annotation of free-text human feedback. Our experiments show that including errors from system utterances and subsequent user responses has a positive impact in response generation. 10 Limitations The majority of our evaluation was done manually. Therefore, with respect to the original dataset sizes, we only consider a small fraction of the data in our study. It might be possible that our results would have been clearer when we would have considered more dialogs for the collection of error-indicating sentences. However, our analysis shows that errors found in the randomly selected dialogs are mostly ignored by the user, i.e., the user does not provide free-text human feedback that could be used for learning. Thus, as far as we are concerned, this does not limit the meaningfulness of our results. Regarding dataset selection, our corpus study (and its results) have only limited expressiveness for knowledge-grounded dialog datasets, since we only consider one of such datasets in our study, Wizards-of-Wikipedia (Dinan et al., 2019). However, this does not affect the relevance of our work, as there are already free-text human feedback annotated datasets available, e.g., FITS (Xu et al., 2023), and we considered a representative number of datasets from other dialog types for which there is a lack of publicly available feedback-annotated datasets, such as task-oriented dialogs. The taxonomies used in this work are also subject to limitations. In the case of the modified Integrated Error Taxonomy, our results show that it improves agreement across different dialog types. However, its abstract error types might limit application for specific use cases, e.g., for a more fine-grained consideration of different types of social errors. Moreover, it reflects only error types observed in the datasets examined. The same applies to the user response type taxonomy. 11 Acknowledgments This work has been funded by the LOEWE Distinguished Chair Ubiquitous Knowledge Processing (LOEWE initiative, Hesse, Germany) and the European Union under the Horizon Europe grant № 101070351 (SERMAS). References Antoine Bordes, Y-Lan Boureau, and Jason Weston. 2017. Learning end-to-end goal-oriented dialog. In International Conference on Learning Representations. Paweł Budzianowski, Tsung-Hsien Wen, Bo-Hsiang Tseng, Iñigo Casanueva, Stefan Ultes, Osman Ramadan, and Milica Gašic. 2018. ´ MultiWOZ - a largescale multi-domain Wizard-of-Oz dataset for taskoriented dialogue modelling. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 5016–5026, Brussels, Belgium. Association for Computational Linguistics. Paul F Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg, and Dario Amodei. 2017. Deep Reinforcement Learning from Human Preferences. In Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc. Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Sharan Narang, Gaurav Mishra, Adams Yu, Vincent Zhao, Yanping Huang, Andrew Dai, Hongkun Yu, Slav Petrov, Ed H. Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V. Le, and Jason Wei. 2022. Scaling instruction-finetuned language models. Emily Dinan, Stephen Roller, Kurt Shuster, Angela Fan, Michael Auli, and Jason Weston. 2019. Wizard of wikipedia: Knowledge-powered conversational agents. In International Conference on Learning Representations. Laila Dybkjaer, Niels Ole Bernsen, and Hans Dybkjaer. 1996. Grice incorporated: Cooperativity in spoken dialogue. In COLING 1996 Volume 1: The 16th International Conference on Computational Linguistics. Braden Hancock, Antoine Bordes, Pierre-Emmanuel Mazare, and Jason Weston. 2019. Learning from dialogue after deployment: Feed yourself, chatbot! In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3667– 3684, Florence, Italy. Association for Computational Linguistics. Matthew Henderson, Paweł Budzianowski, Iñigo Casanueva, Sam Coope, Daniela Gerz, Girish Kumar, Nikola Mrkšic, Georgios Spithourakis, Pei-Hao Su, ´ Ivan Vulic, and Tsung-Hsien Wen. 2019. ´ A repository of conversational datasets. In Proceedings of the First Workshop on NLP for Conversational AI, pages 1–10, Florence, Italy. Association for Computational Linguistics. Ryuichiro Higashinaka, Masahiro Araki, Hiroshi Tsukahara, and Masahiro Mizukami. 2021. Integrated taxonomy of errors in chat-oriented dialogue systems. In Proceedings of the 22nd Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 89–98, Singapore and Online. Association for Computational Linguistics. Ryuichiro Higashinaka, Masahiro Araki, Hiroshi Tsukahara, and Masahiro Mizukami. 2021. Integrated taxonomy of errors in chat-oriented dialogue systems. In Proceedings of the 22nd Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 89–98, Singapore and Online. Association for Computational Linguistics. Hyunwoo Kim, Jack Hessel, Liwei Jiang, Ximing Lu, Youngjae Yu, Pei Zhou, Ronan Le Bras, Malihe Alikhani, Gunhee Kim, Maarten Sap, and Yejin Choi. 2022. Soda: Million-scale dialogue distillation with social commonsense contextualization. Klaus Krippendorff. 2004. Reliability in content analysis. Human Communication Research, 30(3):411– 433. Ilya Loshchilov and Frank Hutter. 2019. Decoupled weight decay regularization. In International Conference on Learning Representations. Sebastian Möller, Klaus-Peter Engelbrecht, and Antti Oulasvirta. 2007. Analysis of communication failures for spoken dialogue systems. In INTERSPEECH 2007, International Speech Communication Association (ISCA). Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F Christiano, Jan Leike, and Ryan Lowe. 2022. Training language models to follow instructions with human feedback. In Advances in Neural Information Processing Systems, volume 35, pages 27730–27744. Curran Associates, Inc. Kishore Papineni, Salim Roukos, Todd Ward, and WeiJing Zhu. 2002. Bleu: A method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting on Association for Computational Linguistics, ACL ’02, page 311–318, USA. Association for Computational Linguistics. Sunghyun Park, Han Li, Ameen Patel, Sidharth Mudgal, Sungjin Lee, Young-Bum Kim, Spyros Matsoukas, and Ruhi Sarikaya. 2021. A scalable framework for learning from implicit user feedback to improve natural language understanding in large-scale conversational AI systems. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 6054–6063, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics. Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Köpf, Edward Yang, Zach DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. 2019. Pytorch: An imperative style, high-performance deep learning library. Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. 2019. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9. Abhinav Rastogi, Xiaoxue Zang, Srinivas Sunkara, Raghav Gupta, and Pranav Khaitan. 2020. Towards scalable multi-domain conversational agents: The schema-guided dialogue dataset. Proceedings of the AAAI Conference on Artificial Intelligence, 34(05):8689–8696. Nils Reimers and Iryna Gurevych. 2019. SentenceBERT: Sentence embeddings using Siamese BERTnetworks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3982–3992, Hong Kong, China. Association for Computational Linguistics. Abigail See and Christopher Manning. 2021. Understanding and predicting user dissatisfaction in a neural generative chatbot. In Proceedings of the 22nd Annual Meeting of the Special Interest Group on Discourse and Dialogue, pages 1–12, Singapore and Online. Association for Computational Linguistics. Kurt Shuster, Jing Xu, Mojtaba Komeili, Da Ju, Eric Michael Smith, Stephen Roller, Megan Ung, Moya Chen, Kushal Arora, Joshua Lane, Morteza Behrooz, William Ngan, Spencer Poff, Naman Goyal, Arthur Szlam, Y-Lan Boureau, Melanie Kambadur, and Jason Weston. 2022. Blenderbot 3: a deployed conversational agent that continually learns to responsibly engage. Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, and TieYan Liu. 2020. MPNet: Masked and Permuted Pretraining for Language Understanding. In Advances in Neural Information Processing Systems, volume 33, pages 16857–16867. Curran Associates, Inc. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023. Llama: Open and efficient foundation language models. Megan Ung, Jing Xu, and Y-Lan Boureau. 2022. SaFeRDialogues: Taking feedback gracefully after conversational safety failures. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 6462– 6481, Dublin, Ireland. Association for Computational Linguistics. Mathilde Veron, Sophie Rosset, Olivier Galibert, and Guillaume Bernard. 2021. Evaluate on-the-job learning dialogue systems and a case study for natural language understanding. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lhoest, and Alexander Rush. 2020. Transformers: State-of-the-art natural language processing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 38–45, Online. Association for Computational Linguistics. Jing Xu, Megan Ung, Mojtaba Komeili, Kushal Arora, Y-Lan Boureau, and Jason Weston. 2023. Learning new skills after deployment: Improving open-domain internet-driven dialogue with human feedback. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13557–13572, Toronto, Canada. Association for Computational Linguistics. Saizheng Zhang, Emily Dinan, Jack Urbanek, Arthur Szlam, Douwe Kiela, and Jason Weston. 2018. Personalizing dialogue agents: I have a dog, do you have pets too? In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 2204–2213, Melbourne, Australia. Association for Computational Linguistics. Xiang Zhang, Junbo Zhao, and Yann LeCun. 2015. Character-level convolutional networks for text classification. In Advances in Neural Information Processing Systems, volume 28. Curran Associates, Inc. Chujie Zheng, Sahand Sabour, Jiaxin Wen, and Minlie Huang. 2022. Augesc: Large-scale data augmentation for emotional support conversation with pretrained language models. This paper is under CC BY-NC-SA 4.0 DEED license. available on arxiv

Conclusion and Beyond: Navigating the Landscape of Errors and Feedback in AI Conversations

Not a day goes by without scientists or renowned intellectuals speaking solemnly about the risks associated with the uncontrolled development of artificial intelligence. The rumor of the first Singularity, Q Star, born in the labs of OpenAI, the company behind ChatGPT, has further accelerated concerns. In the meantime, an ethical and technical framework with uncertain foreseeable effectiveness begins to be debated by politicians and governments worldwide. But what is the reality, in an almost hysterical context in which everyone gives their opinion, often without knowledge... Stephen Hawking shortly before his death, James Cameron referring to Terminator, Mo Gawdat (former CBO of Google X), Geoffrey Hinton (formerly with Google Brain now known as Google DeepMind, considered as one of the fathers of modern AI), Satya Nadella (Microsoft's CEO): all these experts, and many others, have already warned us all on multiple occasions about the civilizational danger of an uncontrolled AI. Sam Altman himself, the CEO then ex-CEO, and now new CEO of OpenAI, co-signed with 349 other experts a one-sentence declaration organized by the NGO "Center for AI Safety" last May. Their words are crystal clear: “Mitigating extinction risk from AI should be a global priority, together with other societal risks such as pandemics and nuclear war.” (1) Elon Musk has also spoken in this direction, not bothered by another contradiction, he is one of the first investors in OpenAI, signed an open letter 6 months ago with 1,100 other personalities for a 6-month moratorium concerning AI (2), and has just announced the launch of “TruthGPT”. Quite the adventure! It seems difficult to pick the wheat from the chaff as to the true intentions of these declarations when we understand that the very same people who claim to be on the side of "good" are all directly or indirectly involved in the most funded and advanced AI research labs. Ethics and necessary safeguards are at the centre of all discussions but at the same time a merciless economic and strategic war is being played out: which nation(s) will take the lead that will profoundly redefine the global power dynamics? All of this makes more sense once you become aware that it is not extraordinary intelligence or exceptional talent that has propelled contemporary billionaires to the level of wealth that is theirs today; it is hard work (this, we must admit), the happy course of events, and above all the exclusive access - well before this concept is revealed to the public - to a dedicated proprietary Singularity. Guess what Elon named his own! AI IS NOT JUST LIKE ANY OTHER INNOVATION AI is not just another innovation on the road to progress. Until now, the nation which had a determining competitive advantage for a certain period of time was the one which announced the first discovery or innovation: a new exploitable reserve of fossil energy, a rare metal mine, the scientific and industrial validation of a new type of nuclear reactors (e.g. fusion vs fission), quantum transistors (QBits), or first vaccine against the Covid-19… With AI, the first country to announce that it possesses an operational Singularity has in its hands the extraordinary tool from which ALL future innovations will emanate. It is the definitive Pandora's box. Tim Urban’s summary of the power of such a tool will help us understand better this unparalleled advantage: Imagine that a singularity is turned on in a lab somewhere on Earth; at time t=zero, it will possess the equivalent reasoning power of a human brain (approximately 100 billion synapses); After a minute, by self-learning, it is today estimated that this Artificial Intelligence will have reached the power of all human brains combined; After two minutes, the power of all human brains since the advent of humans on Earth; After just three minutes, the added power of all that is endowed with thinking in the universe; And at the end of the fourth minute, it's Lucy from the Luc Besson movie, but in real life: this Singularity will undoubtedly have more or less extensive control over space, matter, and time. And that's when the problems start for us humans. “IRREVOCABLE” RISKS The “Singularity risk” is added to a list which now fills the fingers of one hand and characterizes an unprecedented era in the history of humankind: it is the first time that we are concretely confronted with a multitude of risks of “irrevocable impact”, meaning the total destruction of humankind due to a potential catastrophe of planetary scale: nuclear holocaust climate deregulation mass extinction of species global pandemic and in the foreseeable future (within 5 to 10 years according to the consensus), the irrevocable impact of a technological Singularity. Problem: with the risk of a nuclear holocaust, it took several decades for the public to correctly assess the technical prerequisites necessary for understanding such risk, and to become capable of having an enlightened opinion on the matter. We can recall that the concept of nuclear holocaust is addressed for the first time in the Science Fiction novel Last and First Men by Olaf Stapledon (1930), that the first atomic bombing (Hiroshima) took place in 1945, and that the very first Partial Nuclear Test Ban Treaty was signed 18 years later (!) in 1963 in Moscow, by the United States and the USSR as well as a majority of both sides’ allies. But for example, the world is still very far from consensus and concrete action-plans concerning climate disruption. Although it is now proven - it is difficult to deny the statistics which show that storms, tsunamis, and temperature anomalies are increasing - the exact causes are still debated by a strongly divided scientific community. As for the risk of a global pandemic, what to think of the amateur-ish management of Covid by governments around the world, and of the TV sets invaded all day long by so-called experts coming to explain what was needed to be done while never having published any research document considered a reference in epidemiology... Thirty years have allowed people and then decision-makers to correctly understand the issues arising with the creation and proliferation of a cataclysmic weapon like nuclear fire, and to act accordingly to put in place the necessary safeguards. And it is still not fully resolved today, far from it. A full year was necessary for people and then for decision-makers to begin to understand the outbreak of a global pandemic like Covid-19 and to act accordingly to structure the necessary defense and care processes. And it is still also not correctly resolved to this day. In the case of artificial intelligence, when the first Singularity appears, humankind will have 4 minutes. In 2005, with the expression "technological singularity", the computer scientist and futurologist Raymond Kurzweil hypothesized that artificial intelligence would trigger a surge in technological growth which would induce unpredictable changes in society. Beyond this point, progress would only be the work of self-improving artificial intelligence, with new, increasingly intelligent generations appearing more and more quickly, ultimately creating a powerful superintelligence that would far surpass human intelligence. WHO THE PUBLIC NEEDS TO BE INFORMED AND EDUCATED ON AI The risks that the human race has faced up until now are eminently complex: not only are they problems with a very large number of variables (i.e. impossible to model without having to carry out multiple simplifications which degrade the relevance of the results obtained), but they also require a solid cultural and scientific background to be able to produce an elaborated reasoning and develop an opinion in full awareness. Who actually does this preliminary work? But even before this, which body structures the content that would allow people to obtain unbiased information, learn and educate themselves correctly on these subjects? Is it not precisely the job of those who are entrusted with the power to make society function to ensure that people can grasp the issues of their time? We quote Noam Chomsky - a respected intellectual and linguist, former professor at MIT who in his time, exchanged with peers of Foucault's caliber - denouncing the manipulation of the masses by the elites: "People not only don't know what's happening to them, they don't even know that they don't know In his essays, The Age of Spiritual Machines and Humanity 2.0., the computer scientist and futurologist Raymond Kurzweil creates the concept of the "law of accelerating returns" which structures the constant acceleration of technological progress based on Moore’s law (the power of computer chips doubles every two years at constant cost). He suggests that the phases of exponential growth in technological progress would be part of patterns visible throughout human history and even before, from the appearance of life on Earth, taking into account its biological complexity. According to Kurzweil, this will lead to unimaginable technological progress during the 21st century, the inflexion point of which will be the appearance of the complete Singularity. The question will then be whether all of humanity will be able to change speed, which could therefore redefine the notion of humanity, or whether there will be a part of humanity which will prefer to try to remain evolving at its previous speed. The advent of the singularity will likely be so sudden that no one will be granted time to react accordingly: the subsequent phases of awareness, observation, analysis, and then reaction for humans would be inadequate. The definition of technological singularity then becomes technological progress so rapid that it exceeds the capacity of humans to maintain control over it or to know how to anticipate it, understand it and be able to react in time. In this context, the vision of a life during which most of the knowledge and culture would be transmitted during youth and almost stops at one’s entry into the work life becomes nonsense. It is imperative to find a way to remain correctly informed and educated throughout one's life in order to stay abreast of the permanent and increasingly rapid changes occurring in the world. We need to be able to inform and educate the public about AI and the Singularity now BEFORE it happens. Because, unlike other risks of irrevocable impact, once it has happened there is only 4 minutes to go, and it will then be too late to think and decide. Humankind will then have lost any chance of maintaining control over its own destiny. Tim Urban, editor of the excellent blog of popular science , gives the following striking example: if we experimented with trying to define the period of time that would be needed for a human not to be able sustain the shock of time-travelling into the future (i.e. impossible for this person to comprehend the technological advances of this new era they live in), we discover that this period shrinks exponentially over the ages. waitbutwhy.com You would have to send a human from the period 12,000 BC to around the year 1500 for him to go crazy, but it would be enough to send a human from the period 1750 to around the 2000s to get the same result. And it is likely that we are close to the moment when this period will be less than the duration of a lifetime: a senior in the 2020s will probably not be able to mentally survive a post-Singularity world, which will simply be fundamentally too different from the one he had experienced in the first part of their life (without the information, training and learning processes having been set up accordingly to allow him to gradually adapt). HOW TO EDUCATE PEOPLE ABOUT WEB3 The issue at stake today is primarily educational. How to convince today's youth to inform and educate themselves to get the necessary prerequisites for maintaining intellectual and moral discernment at a sufficient level? The task is large when looking at the major threat posed by an unregulated AI and a singularity which could emerge in an instant, somewhere in the Qbits (quantum transistors) of a secret lab... This is the challenge of the artistic and sociological experience “World of The Machines”, created by a team of specialists in media, marketing, blockchain, NFTs, AI, and web3. Learn in a painless way, and capture attention thanks to a bold narrative, a new and immersive user interface, a connected web3 gadget, and a gamified experience. All of this permitted by the advent of the incomplete Singularity Nissam Gorma. Get ready for a deep dive in the next and last part of this series. We will complete this introduction with an interview of Huber Etienne, philosopher specializing in the ethics of artificial intelligence, founder of the "computational philosophy" movement, former head of ethics for generative AI models at Meta, and current CEO of the government consulting company Quintessence AI. Question 1 - Hubert, I imagine that you immediately spotted what was true and what was invented in these first two parts. Let’s just say that I appreciate the liberties you take with reality! Question 2 - What led you to become passionate about AI and make it your career? Unlike most of my colleagues, I have never been a geek and my interest in AI arose accidentally, as the result of a double dissatisfaction. Trained as a philosopher and financier, I was disappointed by my professors at the Sorbonne University who, although brilliant, described the world as it should be without them doing anything to change it. I was also terribly bored in the offices of investment banks and funds. All this gray matter mobilized to produce tons of slides which will barely be glimpsed; all this made no sense to me and generative AI has come to put an end to it! “How should a young philosopher in a world in crisis act to give meaning to his life? » I asked myself. It was then that I chose to embrace philosophy wholeheartedly, but to do so with the efficiency of a banker serving a goal: to participate, at my level, in saving the humanity in distress. My first metaethical questionings – with thoughts relating to the moral relationships that humans must maintain with non-human beings such as other animals, objects or robots – gradually directed me towards AI. My obsession to make myself useful led me to develop computational philosophy: a resolutely activist approach to philosophy, open to all other disciplines and which gives itself the means to enact change. Question 3 - Back to what I'm writing about: what is your opinion on Kurzweil's words and the acceleration of technology? It seems to me that Kurzweil's error consists of seeing in Moore's law a sort of physical law, like the law of universal gravity, when it is only an observed phenomenon, and whose reality only partially fits the theory. In addition, the conservation of Moore's law implies increasing investments, therefore a competition of will in order to support it. We understand that this law only binds those who believe in it and it must continue so that we believe in it; it is based on a self-fulfilling prophecy. Therefore, Kurzweil could well be right, in the sense that humanity is destined to run after Moore's law, but we would also be well advised not to be seduced by these controversial arguments. Current research in AI is gradually moving away from science towards religion and those who stop for a moment to question our collective hysteria are seen as enemies of “progress”. But no one should be ashamed of not falling into technological fanaticism; it is even our duty as reasoned men and women. Question 4 - Do you agree with my vision on the urgency of rethinking knowledge distribution systems, particularly with regards to the themes involved in what I call the "irrevocable impact risks" among which AI seems to me to be the most potentially close to materialize? Absolutely! What I hear here and there is distressing because, from the mouths of this new legion of self-proclaimed experts in AI ethics, what comes out are banalities at best, nonsense at worst, often things that seem true but which research has refuted. Access to knowledge has never been so open and yet the world seems to be disconnected from it. You were talking about the manipulation of the public by the elites but the manipulators have changed because the traditional elites are just as manipulated. They are the toys of those who blow hot and cold, playing on fears and the need for reassurance to serve unworthy ambitions. The worst part is undoubtedly that the decision makers think they know when they don’t. For example, it is often claimed that we must educate young people about the risks of disinformation on social networks, even though seniors are the ones who spread it the most by far! Without falling into ageism, we must nowadays accept that age is not necessarily a guarantee of wisdom and that to remain connected to the developments of the world we live in, continued education is necessary. AI does indeed pose civilizational threats but it remains a tool, a weapon with which some seek to destroy and others to save. “[…] the misfortune is that whoever wants to act like an angel acts like a beast,” Pascal would conclude. Question 5 - I know you've heard about the "World of the Machines" project. What do you think about it? The project is ambitious to say the least! It is also very intelligently thought out, socially useful and, let's say it, a little crazy. This is undoubtedly the reason why I like it so much and why I find it properly post-modern. It is representative of these initiatives that we see emerging in an increasingly complex and confusing world, blurring the distinctions to go back to the meta level and to question the meaning of everything. Social and artistic experience at the crossroads of technologies (NFTs, AI, IoT) and disciplines, it mixes educational pedagogy and fundraising in a world that calls for a mix of genres. Synaesthesias belong to poets as much as to psychotic subjects. We can therefore expect that some will cry genius, others monstrous, and it is again through its divisive aspect that World of the Machines represents our post-modernity as much as it invites us to define it. At the dawn of a new year which promises to be as rich in surprises as the previous one, all my wishes for success accompany this magnificent UFO. (1) (2) https://www.safe.ai/statement-on-ai-risk https://futureoflife.org/open-letter/pause-giant-ai-experiments/

The Digital Antichrist—Part 2/3:  Unraveling the Risks of Uncontrolled AI Development

The widow of Charles of Durazzo was shut up in the castle of Ovo

Every June, tech bros, pension portfolio managers and Twitter threadbois the world over turn their eyes to Santa Clara as hosts its annual Worldwide Developer Conference (WWDC, or ‘dub-dub’). Apple Among other things, dub-dub is best known as being the birthing ground of almost all of Apple’s suite of ubiquitous consumer technology. The original iPhone, iPad, Apple Watch, and Apple TV and subsequent upgrades have all been launched at the event. This year, this stable of consumer devices is joined by an entirely new product for the first time since 2015. That new product is the VisionPro. There is nothing I can tell you about the VisionPro or Apple’s vision for the future of extended reality that a quick browse of , or the internet at large cannot. The point of this article is instead to try and observe: Twitter LinkedIn how we got here; why people are building computers for your face and the many divergent paths this may lead us on. Content Overview XR: A Brief History Headsets Today: Code New Worlds The Next Frontier for Spatial Computers Requests for Startups XR: A Brief History For the uninitiated, XR refers to - a catch-all blend of (AR, e.g Pokemon Go), virtual reality (VR, e.g. Oculus Rift), and ‘capital R’ Reality (e.g. Google Street View). Mixed reality is another term that exists largely on the same dimension as AR and thus will mostly be ignored here. extended reality augmented reality The modern vision for XR begins around 1935. This was the year in which published . In this short story, protagonist Dan Burke is disillusioned with the reality in which he finds himself. To remedy this, he puts on a pair of glasses that transport him to a world of eternal youth and happiness. The only catch is that he obeys its rules (_T&C_s, in modern parlance). Stanley Weinbaum Pygmalion’s Spectacles Fast forward a few decades and we begin to see the first builds of XR in real environments. This begins with Morton Heilig’s ‘ ’ (excluded from the timeline below), a sort of 4D cinema experience that provided the viewer with smells and a vibrating chair in addition to the sounds & sight of the film. Sensorama Also excluded from the timeline below is the first military application of XR. In 1961, became the first headset with motion-tracking technology. Philco Headsight Though it has since been discontinued, the Oculus Rift was the first headset to gain real commercial appeal. Following Oculus, many of big tech’s household names joined the action. Curiously, most had different visions for who the primary customer base for these headsets would be. The dominant expected use case for most was in immersive entertainment applications, more specifically in gaming. This was the path followed by Oculus, Sony through their , and with the Meta Quest line of products. PlayStation VR Meta Others chose to focus predominantly on enterprise. Early pioneers in this segment included and These early movers focused their aims on things like enterprise collaboration platforms (much like some of the metaverse plays described below), emergency response training, and medical education. HTC Vive Magic Leap. Then, the hype cycle. Corporations the world over began to envision what it would look like if we took our lived environment but made it *virtual*. The vision was essentially , but this time with less porn and more corporate work stuff. metaverse SecondLife In the wake of COVID-19, the rise of remote work created a supposed need for collaborative digital environments. joined the party with its own industrial metaverse. Accenture launched a ‘metaverse services’ division. Facebook changed its name to Meta and began spending $1bn per month on this vision, culminating in this brilliant . Microsoft presentation I have had fun panning the metaverse, but the hype cycle around it was an important moment in raising awareness for the arrival of what Apple would refer to this week as spatial computing. As this space evolves, there is every chance that many of these ideas come back into vogue. Other use cases for commercial XR varied from the practical to the benign. Some of them are beginning to become commonplace in popular web back-ends. You take your pick as to which fits which description from the below: Virtual tourism Virtual home inspections Virtual clothing fitting (If you’re going to watch any of these videos, do yourself a favour and make it this one) Corporate training & education Surgical simulations Product design & prototyping Disney-fied sporting telecasts. Most of the existing devices listed above tried to achieve some sort of balance with the use cases they were targeting. The most successful to date by sales, the Meta Quest, did so by positioning hard as a gaming device. Normally, this would be a lesson. However, if anyone is to buck this trend it would be Apple. Apple has a broad suite of already ubiquitous products (1.6bn active iPhones, to name one) that can and will serve as secondary endpoints for VisionPro applications. While Apple’s inaugural demos for the VisionPro were all hosted within working environments, it is more than likely that it will end up being a general-purpose VR. I foresee its role as being a form of spatial augmentation for Apple’s existing ecosystem of products (e.g Maps, FaceTime, Siri et al) as its vision for computers shifts modalities. More on other potential applications later in this piece. Headsets Today: Code New Worlds Spatial Computing Rather than ride the wave of today’s tech buzzwords, Apple used its informational market-making power to stamp authority on a new one at WWDC. https://twitter.com/tim_cook/status/1665806600261763072?embedable=true While it sounds intuitive enough on the surface, let’s delve a little bit deeper into what this term may actually convey going forward. Besides the obvious branding benefits of coining a buzz phrase, the use of the term ‘computing’ feels extremely deliberate. It represents a logical next step from previous eras of ‘desktop computing’ and ‘mobile computing’. Just as Apple has done with the previous generations of personal computers, it will aim to make this concept and terminology ubiquitous. To quote Tim Cook from WWDC: “In the future, you’ll wonder how you led your life without augmented reality” Besides the change in hardware, how is spatial computing functionally different from its predecessors? Arguably the most important immediate shift will be from working in 2D to working in 3D. Tasks such as visualization, workspace navigation (i.e. clicking and moving things), and collaboration will all be enabled in ways not seen under previous modalities. In the longer term, and are more likely to prove the groundbreaking features of spatial computing. contextual awareness interconnectivity refers to the ability of the computer to; Contextual awareness i) adapt existing recommendations and notifications to a user’s sensory context (i.e recommendations based on location and movement patterns, automatic changes to screen presentations based on user habits, etc) and ii) provide customized (possibly agentic) assistance to users as they complete tasks in extended reality. refers to the user’s ability to extend extended reality beyond the device. Beyond just working and collaborating on holographic interfaces, users can control and interact with other devices in their environment. To take a boring example, users can change smart home settings from within the device. Interconnectivity In the longer run, interconnectivity may see Siri’s role as a virtual assistant evolve from voice commanded search aggregator to a living assistant that can help complete tasks from within the device based on the user’s previous habits and preferences. As the baseline capabilities of autonomous agents advance, the combination of Apple’s distribution network and the familiarity of Siri may make it the killer technology for bringing personal agents to the world en masse. Information Presentation As mentioned above, Apple’s demos of the VisionPro focused almost exclusively on working environments. Make no mistake, this first generation is very much geared toward information workers. How will they use it? Firstly, there is obvious and trodden ground. Apple will re-do and re-design all the enterprise XR solutions we have seen before. They will provide resources for institutions to train their employees, students, and leaders. They will provide a bevy of try-on solutions, gaming applications, and cloud TV solutions that we have seen before. One relatively novel example that was showcased as part of the launch was the ‘Minority Report’ style computer interfaces. Instead of having to boot up a physical device every time you want to work on a file, access it from anywhere in the cloud in an instant. But how does this make the core experience of personal computing itself different? Some early ideas: at any given point in time Unlimited monitors with information work (speech, physical movement etc) Multimodal forms of interacting Intelligent browser navigation through sensory (e.g eye) tracking Entirely automatically tailored to user preference custom 3D screen presentations (boring but important) 3D data visualization Each of these will present its own universe of custom ideas that will take the place of extensions and app stores before it. AppleLM This tweet from founder Sulaiman Omar feels prescient. Apple does indeed have a history of sacrificing first-mover advantage for the sake of developing a better product over a longer timeframe. The VisionPro is a perfect example of this, being released 13 years after the first commercially available Oculus Rift. Cognosys AI Integrating language models into its headsets will open up a number of possibilities for Apple that aren’t available to incumbents without their own proprietary LLMs (note: Meta’s LlaMa may put them in the same boat - Balaji Srinivasan has published a highlighting their potential role in the spatial computing race). great thread https://twitter.com/balajis/status/1666506719084843009?embedable=true&thread=true Among the possibilities that this may usher in: trained on users’ real-life habits. Virtual gaming experiences Immediate through the facial ‘browser’. queries built into the headset and trained on user habits. Agentic virtual assistants that recommend actions at a point in time. Spatial recommendation engines Real-time translation. allowing users to re-arrange objects in their space. Environmental augmentation that plays out immediately as the user speaks it. Generative content creation But could it be done more elegantly… The Next Frontier for Spatial Computers Many a commentator proclaimed the death of Google Glass upon seeing it. Who would want to wear a computer on the face? Why would you opt into looking a bit like a dork? Many have noted that not a single Apple executive even tried a VisionPro on as part of the launch. There has to be a better reason for strapping one on than “because everyone else is”. The key question in the long run then becomes: Why would anyone want to strap an XR computer to their face when you can have one inside your brain? The classic counter to this would be the idea of ‘ ’. When everyone is wearing an XR mask, we can reproject photons to make it look as though no one is. Because reprojection involves creating entirely new photons altogether, people can even make themselves look entirely different from what they actually do without the mask. This would become a killer app for catfishing. reprojection The techno-optimist’s concept of living in augmented reality is akin to living in a perpetual lucid dream. Sleep mask when you go to bed, extended reality mask when you get out of it. Examine exhibit A below: The current paradigm of XR relies on these exoskeletal aids for users to navigate alternative realities. The reality is that we are already so close to no longer needing to depend on these aids. Enter the brain-machine interface. BMIs In the long run, the envisioned reality is likely to trump any form of mixed reality that exists today. I use the term envisioned reality here to describe any form of environmental alteration that is; a) completely customizable to the user’s requirements or demands and b) requires no wearable aids. The most obvious conduit for envisioned reality today is (BMIs). For a longer primer, I cannot recommend Tim Urban’s piece on ‘ ’ from as far back as 2017. brain-machine interfaces Wizard hats for the brain BMIs offer many of the same ‘spatial computation’ benefits as tools like the VisionPro but without the need for strapping a weighty device to the face. existing prototypes for the first commercial BMIs involve invasive implants that, understandably, make many uncomfortable (even if the reward is superhuman intelligence and memory). What about when these become less invasive, more accessible, and thus more palatable to the human public? Neuralink’s Extended reality abounds. Requests for Startups Mentions of batteries were oddly absent from Apple’s inaugural launch of the VisionPro. It is less obvious for ‘in the moment’ demos and usage than other design features like interfaces, compute, and weight (as referenced in ) but equally important when used day-in-day-out. Better battery systems for today’s XR devices. this thread by Kyle Samani Not since first dates moved from cafés to online chat rooms has the core wiring of how humans interact with one another been so fundamentally rocked. How can behavior be moderated in a world where people can make themselves appear to be anything? (note: Apple already seems to be working on some form of protocol which may protect against this). What kind of content becomes more pervasive when the range of presentation methods expands so broadly? How do people come together online? Social infrastructure for an XR world. proof-of-identity SecondLife, Minecraft, and Roblox all created huger-than-expected businesses off the back of secondary marketplaces for digital assets. These marketplaces were enabled by creative tools baked into the very fabric of the platforms themselves. Arm the Rebels: Creative Tools for XR. While there are many developers today already working on designs for new experiences etc for headsets and metaverse-style projects, how can we make this process more accessible for the layman (more specifically the 8-year Roblocker) to create worlds of their own imagination? What will be the second coming of Minecraft for the augmented reality or ‘envisioned reality’ paradigm? A large part of the appeal of virtual/augmented/envisioned reality is the idea of living a different life from the one you have now. Much like Dan Burke in anyone will now be able to optimize their lived experiences within virtual worlds. Designs for Lived Experiences (i.e Qualia). Pygmalion’s Spectacles, In the event that such technologies do become ubiquitous, this creates a giant universal market for synthetic instances of subjective experience. The design scope is theoretically infinite. How many ways can you experience pure ecstasy? Nostalgia? The pool of rewards for those who can bring these kinds of experiences to the world through some form of extended reality are enormous. qualia - Also underappreciated may be the development of tools for users to capture or develop these qualia themselves. Is there a market for the ‘recording’ of one’s own lived experiences to share with others? How can people engineer new experiences to take to market and compose atop other open-source qualia? How can we provide guarantees of privacy for personal qualia if they begin to be used in recommendation engines? For the curious, check out the to learn more. Qualia Research Institute One thing that XR still has in common with other computing paradigms before it is its top-down nature. Users are at the whims of Apple for how they want to interact with their phones. If they want to protest this, they can select another one of the depressingly finite options on the market. To date, composability and customisability in hardware have been extremely limited. Open-Source Hardware & XR Software Marketplaces. But what if we were to open-source the development of headsets themselves through decentralized labs? Active participants could work to modify and iterate upon different hardware specifications to their preference. Modular designs could allow for customization at the layman's level. Just as importantly, these open-source labs would have no incentive to create closed-loop systems for software development. Developers could experiment with, ship, and deploy code that could be available to all headsets at the rate that they can build it. Such open-source development would represent a step change in the way people interact with both software and hardware. For some inspiration, check out the great work the team at is already doing. Auki Labs Linked to the point above, the top-down software ecosystem of today means that incentives built into software tend to be warped. Rather than going all out on user utility, market incentives corner developers into building sub-optimal applications that aim to maximize metrics like ‘average session time’ or ‘clickthrough rates’. Private User Agents. The opportunity that open-source development studios would have to outcompete legacy tech companies by building in private, natural language user agents that act as functional assistants to their users is immense. Also published here. The lead image for this article was generated by HackerNoon's AI Image Generator via the prompt "Augmented reality headsets"

Facial Computing: A Brief History, and the Promising Future, of Personal XR

ফেসিয়াল কম্পিউটিং: একটি সংক্ষিপ্ত ইতিহাস, এবং ব্যক্তিগত XR এর প্রতিশ্রুতিশীল ভবিষ্যত

Facial Computing: Eine kurze Geschichte und die vielversprechende Zukunft von Personal XR

Cómputo facial: una breve historia y el futuro prometedor de Personal XR

L'informatique faciale : une brève histoire et l'avenir prometteur du Personal XR

फेशियल कम्प्यूटिंग: ए ब्रीफ हिस्ट्री, एंड द प्रॉमिसिंग फ्यूचर, ऑफ पर्सनल एक्सआर

フェイシャル コンピューティング: パーソナル XR の簡単な歴史と有望な未来

얼굴 컴퓨팅: 개인용 XR의 간략한 역사와 유망한 미래

Computação facial: uma breve história e o futuro promissor do Personal XR

Yüz Bilişimi: Kişisel XR'nin Kısa Tarihi ve Umut Veren Geleceği

Điện toán khuôn mặt: Lịch sử tóm tắt và Tương lai đầy hứa hẹn của Personal XR

面部计算：个人 XR 的简史和充满希望的未来

Venture builders, organizations that create and scale startups from scratch, are wielding a growing influence in the Spanish startup ecosystem. Just a decade since the launch of the first Spanish venture builder, the number of similar vehicles in the country has . swelled to over 30 Unlike traditional venture capital firms or accelerators, which typically invest in and support already established startups, venture builders take a more active role in the development of new companies, coming up with business ideas in-house and using their own resources to bring them to the market. These entities have teams dedicated to identifying market opportunities, developing business models, and recruiting talent. Overall, their objective is to provide all the necessary resources and support to launch and grow the company. In return, they typically take a significant equity stake compared to other models. venture capital The era of the startup factories The of the Spanish startup ecosystem in recent years, as well as the nation’s tradition of innovation and entrepreneurship, has created fertile ground for venture builders. rapid growth These so-called “startup factories” have played a key role in creating and scaling some of Spain’s most successful startups, including Wallapop (which recently closed ) and Glovo. an €81M funding round Outside of Spain, venture builders have helped launch companies like Medium and Venmo. “Our goal is to build sustainable and scalable companies that take into account market trends and take advantage of the great opportunities offered by technology and innovation,” says Christian Rodríguez Fornós, CEO & Chairman at Mutter Ventures, a Spanish venture builder which has helped create startups such as Byhours, Northweek, Hawkers, and UrbanSecrets. The firm’s business model is based on attracting investment for the development of in-house projects. It distributes its own funding and resources according to the needs and development stage of each of its portfolio companies. According to Rodríguez Fornós, some of the biggest hurdles contributing to the failure of startups (inadequate teams, running out of capital, and incorrect market analysis) can be overcome with the support of venture builders, which provide all the necessary elements to help young companies thrive. “All our startups are led by professionals who are experts in the sector corresponding to each project,” Rodríguez Fornós points out. “In terms of structure, each one has a CEO and its own team that expands according to the stage of the company and taking into account the operational needs of the business.” The startup-as-a-service model With in-house experts dedicated to staying on top of market trends, emerging technologies, consumer habits, and regulatory changes, venture builders are well-positioned to identify lucrative opportunities, navigate evolving landscapes, and shape the future of industries. By leveraging their expertise and adaptability, they can proactively innovate and drive transformative change in the market. Corporate Lab, a Spanish venture builder backed by the VIKO Group whose portfolio companies include Okify and Equipzilla, works closely with large corporations looking to diversify their capital or business, offering them a “startup-as-a-service” model. “It is normal for corporations to expand their frontiers through startups. Some do so as venture clients: they contract a specific service or product from a startup. However, more and more companies are opting for other innovation models such as the creation of startups,” explains Brais Comesaña, Head of New Ventures at Corporate Lab. “Unlike the venture client, the result will be within its own business network and will become part of that group, instead of just hiring a service. This makes the process much more secure, as they benefit not only from the capital, but also from all the competitive advantages of having a large company as a parent company,” he adds. Corporate Lab helps companies explore new business models by following a two-step process. “First, we investigate an idea, possible competitors, potential users, and the market. From there, we get down to the nitty-gritty: we talk to all the parties that could potentially be part of what we are building, “ Comesaña says. “Additionally, the companies we work with benefit from being a part of the VIKO Group, which grants them access to specific resources and services that may be required in the future.” As the Spanish startup ecosystem continues to thrive, venture builders will remain a crucial alternative to traditional venture capital models, revolutionizing the way startups are created, nurtured, and scaled. By offering a blend of expertise, support, and risk reduction, they empower entrepreneurs and corporations alike to explore new horizons, foster innovation, and drive the growth of groundbreaking ventures. This article was originally published by Siobhan Parnell on . Novobrief

Redefining the Traditional Startup Model: Spain's Venture Builders Lead the Way

Redefiniendo el Modelo Tradicional de Startups: Los Venture Builders de España Lideran el Camino

Redéfinir le modèle de démarrage traditionnel : les créateurs d'entreprises espagnols ouvrent la voie

पारंपरिक स्टार्टअप मॉडल को फिर से परिभाषित करना: स्पेन के वेंचर बिल्डर्स ने रास्ता दिखाया

従来のスタートアップ モデルの再定義: スペインのベンチャー ビルダーが先導

Redefinindo o modelo tradicional de startup: os Venture Builders da Espanha lideram o caminho

Xác định lại mô hình khởi nghiệp truyền thống: Các nhà xây dựng mạo hiểm của Tây Ban Nha dẫn đường

重新定义传统创业模式：西班牙风险投资公司引领潮流

The Wimblehurst Apprenticeship

Up and away with scalability I thought I knew what scalability was. Just the same way most of us thought we knew what distributed systems were. I realized I had to completely relearn these systems because of the cloud.~Satya Nadella There are a number of considerations that you as a developer, architect or engineering team have to make as your user base grows. Going from products with 10 to say 30 or a million and so forth has its considerations. This, of course, depends on the user base we are talking about. . Remember astute reader: a customer can be an organization with a plethora of other members inside it, or just as well be a single user called Jack Explore with me Imagine thousands, half a million and so forth people performing a search query against a server. If those queries land on your end, would it break? Would it be able to hold? We should be able to handle more data, more concurrent connections and higher interaction rates. A note on interaction rate: How often a user changes the interface on a blog application vs a game. Before we begin, it is important to note that setting up our product for scalability should be as easy and cheap to grow 10x as it should at 0.2x as is common with startups whose needs change rapidly over a short period of time. Foremost, there are two types of scaling we need to observe. a) Vertical scalability b) Horizontal scalability Both of the aforementioned have their pros and cons - each with a use case that would be deemed fit. Let's narrow this down. : Example For solo and relatively small ventures, you might opt for a VPS(virtual private server) which is essentially a physical machine holding multiple 'servers', yours included, each with its own OS and so forth. Once at a certain cap, you might see the need to increase resources and hence, the incremental hunt; more RAM, more storage, more CPU and so forth. almost everything gets more. The example above is one implementing vertical scaling, where, we upgrade the hardware and or the network throughput. In short, you do not need to modify your application architecture. Vertical Scalability Broken to bits, we resolve to a number of techniques. Adding more drives in RAID arrays and thus distributing our reads and writes during database interaction. Switching to solid-state drives from hard drives Increasing the RAM Adding more cores to the server to reduce the context switching. Upgrading network adapters, especially for systems reliant on media file processing. There are a number of drawbacks, however. For instance, the change to a solid-state drive might not show a significant difference if working with MySQL which uses its own data structures for sequential disk operations or Cassandra. At a certain point, is bound to catch up and here, more CPU does little to any good. Vertical scaling also comes with a steep financial cost as you add more and more given the cash input doubles or triples as a unit of resource is added. lock contention Example: A RAM chip will double in cost as it grows in size. That is, x GB is $z, 2x GB is $2z and so forth. A reality check hits us once we get to a stage, say 128GB where the rules are not quite the same. Horizontal Scalability The alternative and preferred method - horizontal scaling. While it may come at a steep cost initially, it does get significantly cheaper as your enterprise grows to reach more users. Let's see how. Caching Use of content delivery networks (CDN) To manage our assets, we save the hassle of having to perform the read and write operations on our own servers. Ever wondered why we have CDNs for most packages(Bootstrap, Material etc)? Example: We have our servers for, say in South Africa. For access, an audacious Sergei (hint - thy knowest thyself) an enthusiastic early adopter from Germany, will hit the CDN(which are just data centres on their own). The CDN, acknowledging it lacks this data - images in our case, will hit our server for the resources it needs (CSS, images, javascript) and serve it to him while storing the same data and resources for further calls to the site. In this way, the next time, the call does not have to go a whole continent away. UrbanDesk Services like Cloudflare, Cloudinary or GoogleCloud CDN should ring a bell. We have our static assets sorted out, but what about our data? Edge Cache Servers I've expounded on the data server to include , yet, another enhancement to our scaling needs. The concept behind it is that our project should be decomposed into individual actions those components need to perform - to group components of our application based on what they do. functional partitioning For instance, we have sections that address this page you are currently reading ( ), then others that speak to different types of users, say, those interested in merch ( ), or the admin ( ). Each of these sections has been compartmentalized. thegreencodes.com merch.thegreencodes.com admin.thegreencodes.com We might also move the database to its own server and update our configs to point to it, hence to each its own, that is, the server the application runs and the server the database runs are different but still communicating. You will notice that I also added the GeoDNS. It works very similarly to an ordinary DNS, the only difference being that it is ' geo-tagged '. The DNS A Domain Name Server(DNS) is a way to identify your server's IP address on the web. In this way, if your server has an IP of say 127.0.0.1, it is displayed as , to whoever else is looking at it. example.com Back to GeoDNS Essentially, when you attempt to connect to , the server IP will be resolved to the data center closest to you. thegreencodes.com Elaboratively, once your product has a certain cap of users , serving all across the globe, having a single server in a single location might degrade the experience these very users get. Example Google has multiple servers. While you might connect to its root domain from Australia and person B from Germany, these might not be served from the same server. The GeoDNS works to resolve the IP addresses of these servers/ data centers based on the location of the end user. You will access the same data, only without knowledge of what IP the domain you are visiting is currently resolving to. What makes horizontal scaling stand out is the basis on which it stands. At its core, horizontal scaling lets you deploy and maintain your project on several medium or small servers. A single server does not necessarily have to be extensive with vast amounts of resources - we split the load between different servers. Of note, is that we need to consider whether our project is or . What this means is to identify whether the split needs data and whether that data needs to be synchronized. stateful stateless Managing data in stateful horizontally scaled products Behold, the and the ability to identify whether we need more reading or more writing of our database. CAP theorem If we indeed have data that needs management, we might opt for three techniques. Sharding by data Sharding by hash Data replication a) b) c) On the one hand, we have sharding by data. In a case example similar to the one shown, we can have a database that holds data from those in Europe and another that holds data from those in America. On the other hand, we can opt to use a consistent technique of hashing specific keys on our model and use those hashes to store data in different databases. For instance, hashing the and stating that data for users with the hash algorithm ABC will sit on this server and not the other. profile_id The above are optimizations meant to prevent hitting the same database for all requests. While not in-depth, I do intend to come back to this in the future. So keep an eye up for this. More than the type of scaling, we have to consider a number of techniques and practices to maintain. As these platforms grow, the following have to be considered. Testing There is more to testing than . Larger systems eventually comprise larger subsystems. How these systems work together (integration testing), whether the application is behaving according to the specification(functional testing), whether it can handle different loads (performance testing) and so forth. unit tests Logging The more the components, the more integration is needed and the more the system is prone to fail. The more servers we have, the higher the chances. We hope for the best but prepare for the worst. Your team needs to know when a module went down, whether the data maintained its integrity, which modules are still up and so on. Choose to go for the approach; let the system always be ready for a crash and whenever it reboots, it should be able to continue without human interaction. An example of such a system is Netflix's which runs to kill random components of the system within working hours to test its reliability. Crash-only Chaos Monkey Conclusion The practices above are intertwined. Functional partitioning would lead to the decomposition of the project, decomposition would lead to a ' ' use of microservices, this would bring a restructure of your team to smaller domain-driven teams, a change in how data is cached and managed and so forth. Build upon this. Go agile and iterate. This path is not linear, nor should it be treated as such. maybe Treat these principles as your northern star, but feel free to morph them based on the value they create for your business. Understand In the creation of sustainable software applications, we place the business logic at the center. The inverse, you'll find, will get you great code, rather than solving the need to be addressed. By placing technology first, we may get a great rails application, but it may not be a great pharmaceutical application ~ Robert C. Martin Onto architecture - the stuff we wish we'd gotten right at the start of our project. Also Published Here

The Aftermath of Going Enterprise

Since 2004, Airbnb, GitHub, Zendesk, Shopify, Kickstarter, UrbanDictionary, CrunchBase, and hundreds of amazing products have been built with Ruby on Rails. Is Ruby on Rails dead in 2021? The language has recently offered some reasons for developers to consider migration to different tools. In 2020, Ruby was still loved by . Why? Because it was built with the key thought in mind that it could “make developers’ lives easier”. Development with Ruby was fun, interactive, and fast. Nowadays, it is hard to imagine a development task for which there is no existing Ruby library. 42% of developers Over the past year, several clients addressed us with the request for updating Ruby-based enterprise products. And yes – modernization is possible; having an outdated Ruby product doesn’t mean it should be rebuilt from scratch with different technology. We believe that in most cases it is possible to work with existing legacy code, and we’d like to share our approach to modernizing legacy Ruby / Rails applications. Read on, if you endeavor to reanimate your Ruby-based project. When You Need to Modernize Ruby / Rails App 1. Your application needs updating if you have any of the following issues: - Outdated and non-interactive UI, problems with UX - Web app feels like it’s from the 2000s - The app makes the full page reload on each user click - UI has no immediate response to actions - Some interactions are impossible to implement Improving user experience (UX) is usually considered as a designer task. It is important to know that modern UI/UX tools require both frontend and backend enhancements. Why is this a risk for legacy systems? 2. Slow application performance - Web pages are loading slowly, so users need to wait a lot - The application is unresponsive and annoying Application architecture and database schema design are complex and time-consuming engineering tasks. In the early stages, serious investment is perceived as high risk because the product may not succeed. Subsequently, code refactoring becomes a challenge at the later stages. Why does it happen? 3. Unable to track Key performance indicators (KPI) - No clear view of data flow in the app - No vision of how users behave - Lack of understanding of how to formulate Key Business Indicators in real-time A data warehouse module! To support your marketing and business team, you need a solid engineering solution to gather data about your business and transform it into human-readable numbers and graphs. What is missing? 4. Application crashes during rush hour - The application cannot handle a big amount of users - Different parts of the system produce a different load, but cannot be scaled separately The system has a bottleneck, which influences the whole application. Bad news: a short-term solution for fixing the bottleneck is not enough in this case. In the long-term, you should consider creating a mechanism to individually boost (scale) your key features. What does this mean? 5. My product is really a set of multiple apps You can clearly see the way to move forward with Service Oriented Architecture, but do not know how to split the existing Monolith while keeping business running. Keeping all the apps in a single code base is like keeping all animals in the zoo in the same cage. It creates a lot of mess and unexpected bugs, slowing down development. What is wrong? 6. Increased infrastructure costs Your web app needs a lot of cloud resources and expensive servers to operate due to non-optimal implementation. Consult with an experienced solution architect and come up with a technical modernization plan for the next phase of the project. What to do? 7. Engineers started to work slowly - Implementation of the new feature takes a lot of time - The development is slow due to high code coupling - The deployment is hard and fragile due to huge amounts of bugs, testing, and fixing which requires additional efforts Not necessarily, but it requires professional help to identify what slows them down and come up with a plan for how to implement modernizations. Is my current team bad? 8. Each release brings a huge pile of new bugs - New features break existing ones - A lot of regression testing is needed by QAs - No possibility to make frequent releases It’s time to perform code modernization. How to stop it? 9. I cannot find engineers for the project’s tech stack - Project parts and services are established on different technologies - You need to hire multiple engineers for each service or look for a super engineer, who knows everything It is painful, but sometimes rewriting parts of the app to use the same technology stack might be a good option. How to proceed? 10. I am worried about security The first version was developed in a hurry with lots of bugs and minimal testing. Now that a lot of users have started using the app, I am worried that I will be sued if there is a data leak. Update all libraries, encrypt user data at rest and in transit, and implement end-to-end SSL. Or hire a team, which will perform the audit and carry out the required changes. What to do to sleep well again? 11. Mobile app integration is hard to implement The project does not have an API, or it requires duplicating a lot of code and bugs. MVP or the first version of the project was developed without taking into account the possibility of a mobile version. It is hard to reuse existing code in the mobile app. Why is it so hard? How to Handle Ruby on Rails Legacy App Upgrade Ruby on Rails (RoR) upholds requirements for startups and MVP projects, due to its strong points, such as a fast time-to-market and a large community of developers. Using RoR, a team of any level could start a project and develop it at the initial stages. This becomes a problem when your business grows – performance, scalability, security, and tracking come into play. At this stage, you need a more experienced team that is able to investigate, develop, and consult. Upgrades may be necessary either in this situation or when the Ruby on Rails application becomes obsolete. At MobiDev, we handle the growth of a Ruby on Rails-based project in the following way: 1. Assign a Tech expert 2. Perform the full project code and infrastructure analysis 3. Identify and prioritize the critical issues to be fixed 4. Come up with a timeline and strategy for the project modernization. It could be a gradual evolution together with further features development 5. Provide the team to do the modernization WHAT TO ESTIMATE IN THE PROCESS OF MODERNIZATION 1. How will modernization influence the development of new features? Should all the team be involved in modernization or is it better to have dedicated developers? When is it time to do the code freeze? 2. Is the code ready for modernization? The features are covered with tests, documentation is in place, backups and disaster recovery are functioning properly. 3. What are the goals of modernization? The team should set understandable and trackable performance indicators of modernization and regularly assess how the process is moving. 4. Is the risk of modernization warranted? Modernization is an improvement but everything has its cost. Modernization plans should cover the increased risks of developer error (automation is required), rising complexity (more documentation to onboard new developers needed), etc. Ruby on Rails Legacy Application Update Case Studies Here we will overview three case studies, where outdated Ruby and Ruby on Rails applications were modernized, and one case, where Ruby became a solution for modernization of a legacy software system initially built with another programming language. Three of the four case studies are accompanied by comments from developers involved in the project. CASE STUDY #1: REFACTORING HEALTH LEGACY APPLICATION The project was inherited from another development team after five years of development. So, it was a legacy code. It had outdated versions of Ruby and Ruby on Rails, and the code was written using many different architecture approaches mixed in a single codebase. There was no related documentation whatsoever. Challenges 1. Upgrade Ruby and Ruby on Rails versions 2. Define the set of patterns and abstractions to be used during the refactoring stage 3. Make sure the new version of code is developed in an extendable way 4. Provide a high level of security and personal health information (PHI) protection to comply with HIPAA 5. Setup performant and easily scalable infrastructure 6. Document the architecture design 7. Develop a new set of features and launch in half a year Solutions Upgrade to Ruby 2.7.1 and Rails 6.0.1 The first task performed by the team was to ensure that upgrades wouldn’t break the existing functionality. We developed and refactored 4488 automated rspec tests, which covered most of the existing application features. This step made it possible to upgrade Ruby on Rails to the newest versions safely. After the upgrade, the rspec tests showed us all the cases, where new libraries caused problems, and we managed to fix them. Modular (Component) Monolith Rails architecture design Although the Rails application had a single repository, by its true nature the application consisted of many different services. It was a real challenge to develop new features because of the way the code of all these services was mixed together. After the brainstorming phase, the team decided to use Modular Monolith design for Ruby on Rails API. The benefits of this choice were that all services were decoupled into separate Rails engines (Admin Portal, CMS, Common API, etc.), it was easy to navigate the project and implement new features, and the design had a clear and easy to implement set of guidelines. Building new infrastructure Microsoft Azure was a great candidate for a new cloud enterprise solution. At that moment, it had a suitable bunch of tools to set up new infrastructure, taking into account security, scalability, durability, performance, and recoverability. For security reasons, it was decided to implement an “End to End SSL” approach, which ensured that at any phase of communication between services the traffic was encrypted and safe. Azure Kubernetes service called AKS enabled achieving required scalability, durability, and performance for our Ruby on Rails API. Last but not least, the whole setup was performed using the Helm tool and Azure DevOps CI/CD. It was a real challenge to work on the healthcare project that was passed to MobiDev’s Ruby Team from previous developers without proper documentation. During the project’s 2.5 years, we were doing a major upgrade and developing a product by adding new functional modules. A major upgrade included refactoring files, moving business logic operations to Trailblazer and database queries to a separate repository. Following Domain-driven design principles, we thoroughly covered both the new code and the legacy one with QA testing. The project had quite a cumbersome monolith architecture, but we managed to divide it into several modules, using . To ensure proper code formatting, we applied . Rails Engines Rubocop Ruby Engineer, MobiDev Yaroslav Tiahelskyi CASE STUDY #2: UPGRADING SPORTS TOURNAMENTS MANAGEMENT SYSTEM MobiDev’s Ruby Team dealt with competing objectives for this project. We had to develop more features in the same budget, reduce the number of bugs, and increase extensibility. Additional tasks included simplifying the onboarding of new team members and improving overall security. Challenges 1. Move the code to Service and Query Objects plus Decorators by introducing a Domain Driven Design (DDD) approach 2. Unify the frontend part 3. Make onboarding for new team members and developing of new features seamless and fast Solutions It was decided to: 1. Move logic from controllers to Service Objects 2. Move query functions form models to Query Objects 3. Move helper functions form models to Decorators 4. Upgrade Rails version 5. Develop all Front-end features using React 6. Unify code style using Robocop and Rails upgrade Applying DDD allowed the developers to simplify overly-complicated models and controllers of the Legacy Ruby on Rails application and make sure it was easy to extend the code for new features. The Rails app had 2 different ways to develop UI components: pure jQuery with Ajax and EJS with remote links. To unify the frontend part, we employed migrating to React.js as an appropriate UI approach. Interesting technologies and a powerful team always motivate me. I was absolutely excited to work on the outstanding project which had no analogs, especially for kids, in that kind of sport. A lot of teams were involved to launch the mobile, desktop, and browser versions of the application. I am proud that we‘ve managed to transform the legacy project into the one with optimal architecture, totally covered with QA testing, and our API used only by GraphQL. Ruby Engineer, MobiDev Erik Zheved CASE STUDY #3: MODERNIZING BANK MANAGEMENT SYSTEM Multi-tenant Ruby on Rails startup development had been running smoothly until the team faced several risks: slow response time under load, increased development time for new features, a single point of failure, and a lack of security mechanisms. Challenges 1. Provide migrating from AWS to Azure 2. Improve legacy code base 3. Find out how to solve other tasks related to risks encountered by the team Solutions After investigation, the following problems were found: 1. A lot of load from periodic computations 2. At the initial stages, quality was sacrificed for speed 3. Hard to extend and maintain code base 4. Slow tests (harder to develop and release) 5. Hard dependencies between system components Migrating from AWS to Azure The original first version of the system was built using AWS cloud in an outdated manner. All developers had direct access to servers and databases, which is unsafe because of human errors and the high possibility of troubles that could happen from time to time. Also, servers were targets to crash because of periods of high load. The Ruby on Rails web app generated a lot of jobs and made the Compute Server (Sidekiq) get “stuck”. At these moments the whole system became unavailable due to the hard coupling of the components. It was decided to perform a decisive technology modernization and also migrate to Azure. Additional security mechanisms were implemented around authentication and authorization, applying Single Sign On approach (SSO) and Azure Active Directory (AAD). They were seamlessly integrated into the existing Ruby on Rails solution and Active Admin. All the public traffic was routed through Gateway and Firewall for high protection of sensitive data. Key Vault allowed to store Ruby on Rails credentials and secret keys. As the core of the system the AKS (Kubernetes) cluster was set up to provide the required value of performance and stability of the system, eliminating the single point of failure and providing a fault-tolerant experience for users. Improving legacy code base After the internal analysis of the code base, it became clear that to improve time-to-market for new features and reduce the amount of bugs, we had to get rid of monolith architecture and the server-side rendering approach. The split to services and separating the frontend part drastically assisted Ruby on Rails developers to speed up automated tests and increase the extendability of the legacy monolithic Ruby code. I’ve been working on the project for more than four years. Along with the team we’ve implemented the project features, using parallel computing, service architecture, and complex business logic. The project utilizes an advanced security system based on Azure, , and Helm. The tech stack comprises RoR, PG, Sidekiq, Redis, React, and Java. Kubernetes It’s worth noting that the project meets the Accessibility requirements, has successfully passed Security Audit performed by the client’s IT department, and is under ongoing performance optimization. The project is still being actively developed by the client, and my team participates in technical decision-making and developing of business logic. Ruby Team Leader, MobiDev Viktor Volontsevich CASE STUDY #4: WORDPRESS-BASED LEGACY APPLICATION MODERNIZATION WITH RUBY ON RAILS RoR is still considered the best platform for startup product development, so sometimes it can become a solution for the modernization of a legacy software system built with another programming language. The was initially a WordPress based startup social network for teams interested in the growth of their businesses. Soon an idea of a special growth management tool came into play, and it could be built using Ruby on Rails framework. project The team quickly realized that the project was a set of multiple services, and it was difficult to keep engineers for different technology stacks (WordPress and RoR). In addition to the startup-based architecture design, it was hard to develop new features and the development slowed down. Challenges 1. Unifying the tech stack 2. Isolate multiple apps from each other 3. Reduce release risks Solutions It was decided to: 1. Rewrite the WordPress part of the project using RoR for unifying the tech stack 2. Introduce Microservices architecture to keep multiple apps isolated from each other and make it possible to share the data between them (like users). This increased overall stability and performance 3. Introduce CI/CD automation for reducing release risks Why Ruby on Rails is Still Our Choice in 2021 In summary, the increasing demand in web systems development raises the question, which tool is best suited for these needs? For us, Ruby on Rails is still a valid and preferable tool for the following reasons: 1. It is a “swiss knife” of web development that allows us to create prototypes, proof of concepts and MVPs fast and test the market with a minimum budget. 2. Ruby on Rails framework has built-in tools out-of-the-box to fulfill 80% of common web systems feature requests. 3. For developers with the right skills, it is still a tool for infinite product growth in terms of features and code. 4. Ruby on Rails makes both businesses and engineers happy. It was built with the philosophy “to make developers’ lives easier” in mind, which simplifies delivering new features and meeting product owners’ expectations. It is undeniable that Ruby on Rails has some drawbacks. One of them is the high price for each mistake. Errors originating from the planning stage or the first development phase are difficult to correct in the future. As a result, this turns into broken interactions between separate parts of the app. To avoid mistakes, you need to find an experienced team that is able to deliver the product with the required customization and functionality, devoting attention to previous research. We are ready to work on challenging projects and believe that with a competent approach they are bound to succeed. For us, RoR is flexible enough in the long term, as well as secure and easy for business logic implementation. Ruby on Rails shines again with all its refactoring best practices gathered by the MobiDev team throughout the years in web development! Written by Serge Koba, Ruby Team Leader at MobiDev

Stories, News & Mentions about URBANSDK

Urbansdk’s Stories on HackerNoon

Urbansdk's News & Mentions