DeepPavlov

20 Aug 2024 17:24

DeepPavlov 🎅 поможет в цифровой трансформации сотрудникам российского производителя чая и кофе МАЙ 🔥

Сотрудничество команд направлено на реализацию совместного проекта, цель которого - создание прототипа цифрового помощника на основе больших языковых моделей для помощи сотрудникам компании 🤖

Подробнее о проекте читайте по ссылке.

DeepPavlov

31 May 2024 17:21

В эти выходные встречаемся на Data Fest 2024 🚀

Ждем вас в воскресенье в зале «Мулен Руж» 🔥 Мария Молчанова, исследователь DeepPavlov, расскажет про разметку данных с помощью экспертов, краудсорсеров и больших языковых моделей🤖

🔹рассмотрим три подхода к организации разметки данных на примере задачи разметки диалогов на сложные лингвистические классы
🔹поделимся методами и инструментами для упрощения процесса разметки, чтобы сделать его доступным для исполнителей без специализированных знаний
🔹расскажем про способы автоматизации разметки с помощью больших языковых моделей.

❗️Дата: 2 июня в 17.30

Регистрация и нетворкинг 👉 https://ods.ai/events/fest2024-yandex-msc

DeepPavlov

11 Oct 2023 14:09

Hello everyone!

Recently we published an article “Memory enhances ChatGPT performance in multi-hop QA”. The article discusses the approach for ChatGPT memory augmentation and compares it to the task-tuned language model on multi-hop QA tasks.

Check out the full article, and don’t hesitate to contact us if you have any questions or comments!

#AI #NLP #DeepPavlov #ChatGPT #MHQA

DeepPavlov

24 Aug 2023 13:26

Hello everyone!

We are excited to announce the #release of DeepPavlov 1.3.0 🎅. Here is a sneak peak of what's included in the new release:
◾️ A few-shot text classification model that allows you to operate with a small amount of training data and supports OOS examples.
◾️ A refactored ODQA model optimized to consume less RAM during training.
◾️ A long-anticipated syntax parser and morphotagger.

As a reminder, you can check our multilingual models at the DeepPavlov Demo page. Be the first to try out the DeepPavlov 1.3.0 release and don’t hesitate to share your feedback with us!

DeepPavlov

26 Jul 2023 12:03

Hello everyone,

We are back with some fantastic news: a brand new demo of DeepPavlov Library just came out, featuring a lot of new, much needed multilingual models. Be the first to try it out and don’t hesitate to share your feedback with us!

As you probably know, the NLP field is highly dynamic, and with the current pace of change we’d like to hear from you to learn how else we could help you solve your everyday natural language processing tasks; if you’re up for it, we’d highly appreciate your filling out a a short survey!

DeepPavlov

17 Jul 2023 11:56

TIME FOR TUTORIALS ON DREAM 📣

Learn how to create a simple and light-weight 🤖 that specializes in discussing movies and answering factoid questions, utilizing the existing Dream components.

#AI #NLP #DeepPavlov #GenerativeAI

https://lnkd.in/eMy5xWzZ

DeepPavlov

21 Jun 2023 11:29

Ever wondered how #ChatGPT stacks up against traditional #NLP models? 🤔 The results might surprise you! 🚀 Dive into this insightful analysis to discover the strengths and weaknesses of ChatGPT on Q&A tasks. Will it outshine the rest? 🌟

#NLProc #ai

Find out all the details 👉🏻 https://artificialcorner.com/how-good-is-chatgpt-on-qa-tasks-953a10236ec7?source=friends_link&sk=2afb1ccb4717935d2b15efccd4f98057

DeepPavlov

09 Mar 2023 09:10

CV-Multimodality Engineer at DeepPavlov.ai

At DeepPavlov.ai, our mission is to simplify the development process for chatbots, dialog systems, and Multiskill assistants worldwide. We've already created multimodal Open Source dialog systems that can speak, write, see, and understand their surroundings.

We are looking for an experienced CV engineer to join our team and work primarily with computer vision models in multimodal text-audio-vision models. The primary goal of this role is to integrate the vision modality into both virtual and embodied dialog agents.

If you have 2+ years of experience with CV models, proficiency in tools such as docker, bash, git, and pytorch, and a talent for writing clean and efficient code, we believe you might be a perfect candidate.

Find out more about CV-Multimodality Engineer here:
https://deeppavlov.ai/careers/tpost/82g6ab53k1-cv-multimodality-engineer

#jobs

DeepPavlov

06 Mar 2023 09:34

Hello, everyone!

Recently we published an article "Knowledge-Grounded Response Generation in the DREAM Socialbot". We explored the integration of knowledge bases into socialbot conversations to improve user engagement and conversation quality.

In this article you’ll find out how to use a pre-trained neural conversational model to generate relevant utterances that contain new information from given knowledge facts. It helps to evolve the discussion and encourage the user to continue the dialogue with the chatbot.

Check out the full article, and don't hesitate to contact us if you have any questions or comments!

DeepPavlov

08 Feb 2023 11:35

Dear DeepPavlov Community!

We will be holding a live DFF demo on February 27th at 9am (PST). Join us and enhance your chatbot development skills with DFF.

For those of you who are not aware, this is perfect for AI Developers, Data Scientists, and Software Developers seeking to integrate advanced ML/DL NLP models with intuitive scripting dialog systems.

Discover Dialog Flow Framework (DFF), the cutting-edge open-source technology that streamlines the creation of flexible dialog graphs using a powerful combination of Python-native DSL and DL models. Brought to life by the DeepPavlov team during the Amazon Alexa Prize Socialbot Grand Challenge 4.

During the demo, you will learn:

▪️The ins and outs of DFF's functionality
▪️How to build a simple FAQ system and live chat support with goal-oriented services (using ChatGPT type models)
▪️The latest updates, including an asynchronous processing pipeline and a more user-friendly interface, based on your feedback.

Sign up now and receive a link to the demo

DeepPavlov

12 Jan 2023 13:24

Hello everyone!

We are back from holidays with some fantastic news! The very new 1.0.2 release of DeepPavlov Library has just come out.

We were excited to deliver v1.0.0 of the DeepPavlov Library, with full migration to PyTorch and python 3.8, 3.9 support.

Now get ready for v1.0.2!

In the updated version we’ve added support to use our Python pipelines, a native mechanism to build pipelines on top of the models in the library, together with an example showcasing the use of python inference pipelines. We also made model requirements installation more comfortable, and delivered several bug fixes and improvements.

Be the first to try the new version. Don’t hesitate to share your feedback with us!

DeepPavlov

23 Dec 2022 12:55

Hello, everyone!

The DeepPavlov team built a platform for designing multiskill AI assistants – DeepPavlov Dream. It works with different environments, from texts to entire virtual worlds. We made several demo to show you how you can benefit from developing your own multiskill AI assistant with DeepPavlov Dream.

1. First video demonstrates how ConvAI can help the user in Minecraft. Our virtual assistant DeepPavlova is an AI-driven NPC building the blockhenges in Minecraft following the user’s lead. It perfectly navigates through the environment, recognizes different objects, and acts. DeepPavlova can keep a conversation in an open domain and perform simple tasks.

2. Second video shows how virtual assistants can be used in the metaverses.The user can simply talk to a virtual assistant about the Beatles, and learn their history in the Roomful metaverse. The idea of the project was to create a Virtual Assistant that could keep up a conversation on any Beatles-related topic and take the user on a museum tour to expand the Roomful user experience.

3. Third video shows how our virtual assistant can control the actions of the robot in real life. The robot does all tasks correctly: goes back and forward, rotates clockwise and counterclockwise.

It’s only 3 examples of how you can build virtual assistants on top of DeepPavlov Dream. Don’t hesitate to build your own virtual assistant!

DeepPavlov

19 Dec 2022 12:01

Hello, everyone!

We are happy to share with you our most complete curriculum for the development of multiskill AI assistants using the #Dream Platform 🔥

In this article you’ll find out how to get started building your multiskill ConversationalAI assistants with the DeepPalov Dream. We collected a full list of all essential materials that will guide you through the development of your own AI Assistant.

Follow the link to read the article, and don't hesitate to make your own Multiskill ConvAI assistant with DeepPavlov Dream!

DeepPavlov

22 Nov 2022 11:46

Hello, everyone!

Daniel Kornev, CPO at DeepPavlov.ai, will share his expertise at AI Journey 2022. AI journey is an international conference for discussing the development of artificial intelligence and machine learning technologies

Daniel Kornev will take part in the Science track with the presentation on the topic “Making a small step beyond text and voice-based Conversational AI with DeepPavlov Dream Builder and Dream for Robotics”. He will talk about the current state and vision of the development of the conversational AI-platform, the DeepPavlov project, and the developments in the field of robotics.

Also, researchers from DeepPavlov.ai will present scientific posters in the field of artificial intelligence/machine learning:
◾️Kseniya Petukhova, Junior Researcher II at DeepPavlov.ai, will present the topic: “Goal Tracker & Dialog Games for Human Goals in Open-Domain Conversations”. In this work, we propose a first goal-aware prototype of DREAM Socialbot.
◾️ Dilyara Zharikova, Senior Researcher at DeepPavlov.ai, and Dmitry Kosenko, Junior Researcher I at DeepPavlov.ai, will present the topic: “Utilize knowledge and persona for dialogue response generation”. The proposed approach achieved a state-of-the-art result at The 1st Workshop on Customized Chat Grounding Persona @ COLING 2022.

Date: 23d November 2022
Join us at AI Journey and register below:
https://ai-journey.ru/

DeepPavlov

15 Nov 2022 08:35

Hi, friends!
This is our regular #ConvAInews!

◾️ Last week we made a 1.0.0 release of DeepPavlov.ai Library. In this version TensorFlow-based models were replaced with PyTorch-based ones, some models were renamed, and various models and components were removed. Also, we improved several models: Entity Linking, Context Question Answering, NER, and Classifier models. Moreover, we added Python 3.8/3.9 support, case-agnostic distill NER for DREAM, DeepPavlov Topics Classifier model, and Russian SuperGLUE models. You can find more about it here.

◾️ An article about BLOOM, A 176B-Parameter Open-Access Multilingual Language Model, has been finally published. It was designed and built thanks to a collaboration of hundreds of researchers. BLOOM achieves competitive performance on a wide variety of benchmarks, with stronger results after undergoing multitask prompted finetuning. You can find more about it here.

◾️ Google has Beta-launched its text-to-image AI model after the wave of text-to-image models whose creations have flooded the internet this year. It’s called Imagen. But unlike those released by Meta, Microsoft, and OpenAI, the internet giant is taking a cautious approach by offering users a limited taste inside its AI Test Kitchen. The version available inside AI Test Kitchen has two functions for users to play with, City Dreamer and Wobble.

◾️ MTV India introduced AI-powered virtual rapper ‘Bot Hard’ as its latest star. Bot Hard uses a specially trained version of OpenAI’s GPT-3 language model to produce its raps. The developers filtered the AI’s training specifically to look for rhymes and rhythms that mimic current rap stars, with continued use continuing to refine its creations.

◾️ Developer combined three AI systems - Stable Diffusion, Whisper and GPT-3 - to create a futuristic design assistant. Developers can have simple conversations with the assistant and give it instructions for image ideas. The assistant confirms the instructions and either carries them out directly or asks queries to clarify them.

◾️ Marketing content creation firm Simplified introduced an AI text-to-image generator. The new tool combines OpenAI’s DALL-E and Stability AI’s Stable Diffusion synthetic media services enabling its clients to produce visuals from a written prompt.

Stay tuned!

DeepPavlov

01 Jul 2024 15:33

3 июля - пройдёт бесплатная онлайн-конференция DataStart по Data Science, машинному обучению и нейросетям 🔥

Мария Молчанова, исследователь DeepPavlov, выступит с докладом: Как получить больше от LLM: Продвинутые стратегии и реальные кейсы ❗️

🔹Какие задачи можно решать с помощью LLM и что такое промпт
🔹Для чего нужен Prompt Engineering - Таксономия методов промпт инжиниринга
🔹Как ставить эксперименты - Практические рекомендации
🔹Несколько полезных продвинутых методов работы с промптами

Подробнее о мероприятии и регистрация 👉 https://datastart.ru/

DeepPavlov

12 Apr 2024 15:32

Научные хакатоны возвращаются вместе с DeepHack!

📌 В рамках хакатона DeepHack.Agents 🤖 вам предстоит создать своего ИИ-ассистента для научных исследований с помощью фреймворка GigaChain

Дата: 24-28 апреля
Формат: гибридный

В последний день хакатона у вас будет возможность пообщаться с экспертами вживую и собраться вместе с другими участниками! 🔥

Регистрироваться можно как готовой командой, так и собрать её на хакатоне.

❗Успейте подать заявку до 22 апреля по ссылке

DeepPavlov

05 Sep 2023 11:53

Hello everyone!

We are excited to announce our new article about multi-task models in DeepPavlov. These models allow combining many tasks into one backbone while retaining or even increasing the model quality.

Check out the full article, and don’t hesitate to contact us if you have any questions or comments!

#AI #NLP #DeepPavlov

DeepPavlov

01 Aug 2023 13:42

📣 TIME FOR NEXT TUTORIAL ON DREAM!

Create a distribution with reasoning and API calls performing utilizing OpenAI ChatGPT to think of actions required to handle user requests and to choose the suitable API to use.

#AI #NLP #DeepPavlov #GenerativeAI

https://clck.ru/35AVb2

DeepPavlov

19 Jul 2023 11:23

📣 TIME FOR NEXT TUTORIAL ON DREAM!

Discover how to create a generative bot 🤖 with a predefined persona, using the existing Dream ☁️ Persona distribution that utilizes OpenAI ChatGPT.

#AI #NLP #DeepPavlov #GenerativeAI

https://lnkd.in/eKgdAa3s

DeepPavlov

10 Jul 2023 14:38

📣 Calling all #ACL2023 attendees! 🚀 Don't miss our System Demo on the open-source DeepPavlov Dream, designed for building complex, generative AI Assistants. 🤖💬

📌 10 July @ 14:00 Demo Session 2 @ Frontenac Ballroom, Queens Quay&Bay

#AI #NLP #DeepPavlov #GenerativeAI

DeepPavlov

14 Mar 2023 11:19

Hello everyone,

Here is our newsletter with the top industry events and the most up-to-date developments in the conversational AI area.

◾️ DeepPavlov team has explored the integration of knowledge bases into socialbot conversations to improve user engagement and conversation quality. We published an article "Knowledge-Grounded Response Generation in the DREAM Socialbot". Read the full text at the following link and don't hesitate to contact us if you have any questions or comments!

◾️ Microsoft has announced an exciting new Power Platform that allows end users to seamlessly incorporate AI into their workflows to boost productivity. The automation platform comes equipped with next-generation AI features for Power Virtual Agents and AI Builder.

◾️ The continuous boom in conversational AI and LLMs forces companies to rethink and present AI assistants again, including companies that had previously put on hold or shut down development of AI assistants. Salesforce has brought Einstein back to life with generative AI provided by OpenAI. The new Einstein GPT employs large language models and its own databases to compose and edit emails to customers, draft digital advertisements, and generate computer code for developers.

◾️ Grammarly has announced GrammarlyGO, a collection of generative AI writing and editing tools that serve as a productivity aid for communication. This expands on the spelling, grammar, and style guidance already offered by the company's apps and plug-ins.

That's all for today. Stay tuned for more exciting news and developments in the world of Conversational AI!

DeepPavlov

06 Mar 2023 16:18

Hello, everyone!

Exciting news: Mikhail Burtsev, the Head of DeepPavlov.ai, and Daniel Kornev, CPO of DeepPavlov.ai, will be speaking at OpenTalks.AI tomorrow! OpenTalks.AI is a leading independent Open conference on Artificial intelligence.

Mikhail will be reviewing the main results in Natural Language Processing for 2022, discussing achievements and trends.

Meanwhile, Daniel will be sharing his insights on redesigning DeepPavlov Dream around Large Language Models. He'll also give tips on how to build your own Generative AI Assistants with DeepPavlov Dream.

Please join us and learn more about simple and complex chatbots’ development with DeepPavlov.ai.

For more details, click on the link below:

https://opentalks.ai/en/timetable2023#!/tab/540721099-3

DeepPavlov

17 Feb 2023 13:13

Hi, friends!

Welcome to Conversational AI News! If you are looking forward to hear about the latest developments in AI research and technology, we've got you covered.

◾️ Discover the art of crafting Generative AI Assistants with the aid of DeepPavlov Dream open-source multiskill AI Assistant platform. Our most recent version allows for the integration of open-source Generative AI Models such as GPT-J and BLOOM within your DeepPavlov Dream-based AI assistants. Our latest article, "Building Generative AI Assistants with DeepPavlov Dream," presents a comprehensive guide to building your very own Generative AI Assistants. Take the first step towards developing cutting-edge conversational AI by following our instructions.

◾️Explore the intricacies of building a custom chatbot with DeepPavlov Dream in our latest article. This piece delves into the nuanced decision of selecting the ideal LLM (language model) to suit your project's needs. We evaluate and compare the performance of eight distinct large language models, each with varying sizes and architectures, in response to three unique natural language prompts designed to evaluate their abilities as a dialog system. Unveil the key insights and gain a deeper understanding of the best practices in language model selection for your next AI project.

◾️ Join the live demo of Dialog Flow Framework (DFF) on Feb, 27th at 9 am (PST) to streamline your chatbot development with cutting-edge open-source technology. DFF combines Python-native DSL and DL models to create flexible dialog graphs. Discover the ins and outs of DFF's functionality, learn how to build a simple FAQ system and live chat support, and hear about the latest updates and features, including an asynchronous processing pipeline and a more user-friendly interface.

◾️ Microsoft AI Research is breaking new ground with their proposed eXtensible Prompt (X-Prompt) technology which takes large language models beyond natural language. Now, these models can discover correlations and patterns in text that were previously impossible. It's a fascinating breakthrough that's sure to have major implications for the field of AI.

◾️ The Mycroft Mark II is a privacy-focused open-source device designed as an alternative to smart speakers and displays. The developers launched a crowdfunding campaign for it, but had to end it due to the lack of funds, leaving thousands of backers without devices. Mycroft CEO Michael Lewis confirmed the company could not continue to manufacture and distribute Mark II units to its backers.

◾️ Microsoft has launched an updated AI-powered Bing search engine and Edge browser designed to handle complex search queries. The new features include interactive chat experiences, detailed summarized answers, and complete results. The goal is to solve the problem of half of the 10 billion daily search queries going unanswered.

That's all for now. Stay tuned for more exciting news and developments in the world of Conversational AI!

DeepPavlov

31 Jan 2023 15:38

Hi, friends!
Welcome to #ConvAInews!

◾️We just launched v1.0.2 of DeepPavlov Library. This update supports using native Python pipelines and features a built-in mechanism for creating pipelines on top of the Library models. We added an example of Python inference pipelines, too. Plus, installing model requirements is now easy-to-use, and we’ve fixed bugs and made improvements. Get the updated version here.

◾️ SymbolicAI is a framework that can implant human thoughts, reasoning, and behavior into a computer program. It combines the strengths of Symbolic AI and Large Language Models.

◾️ Microsoft Invested $10 Billion in ChatGPT Maker OpenAI. OpenAI intends to give Microsoft access to some of the most popular and advanced artificial intelligence systems.

◾️ Apple introduced a new version of its HomePod smart speaker. This second generation HomePod is a match in size of its predecessor, but with upgraded hardware and sensors. Siri, its built-in voice assistant, can identify the sound of smoke and carbon monoxide alarms, sending a phone alert about it.

◾️ Microsoft has introduced VALL-E, a novel language model method for text-to-speech synthesis (TTS). It can employ audio codec codes as intermediate representations and can replicate anyone's voice after listening to just three seconds of audio recording.

Unleash your creativity and bring AI to your projects!

DeepPavlov

31 Dec 2022 12:23

Hello, everyone!

DeepPavlov.ai team wants to wish you a great celebration of the New Year and Christmas. Enjoy these holidays together with your loved ones, and don’t forget about your dreams in the Conversational AI space!

We’ll continue to please you with new developments, posts and articles in 2023!

P.S. We generated this Christmas card with the help of AI!

DeepPavlov

21 Dec 2022 11:52

Hi, friends!
Welcome to #ConvAInews again!

◾️ Recently, we published an article with our most complete curriculum for the development of multiskill AI assistants. In this article, you’ll find out how to get started building your multiskill ConversationalAI assistants with the DeepPalov Dream. Follow the link to read the article.

◾️ Amazon released new Alexa features for the Echo Show — smart display allowing users to control the voice assistant through gestures and typing. For instance, users can now dismiss timers with a wave of their hand, tap out words that the voice assistant will speak on their behalf, or enable universal captioning in one command.

◾️ DeepBrain AI launched an interactive, synthetic version of the America’s Got Talent host — Howie Mandel. “AI Howie” replicates how the real actor looks, speaks, and moves, allowing for realistic videos and even real-time interactions with his digital counterpart in web browsers, metaverse virtual worlds.

◾️ SoapBox Labs and Early Bird Education partnered on a project to spot and diagnose dyslexia in children. Soapbox will augment EarlyBird’s dyslexia screener with its speech recognition AI to measure a child’s oral language skills and potentially figure out any obstacles to their learning how to read even before they begin.

Stay tuned!

DeepPavlov

07 Dec 2022 15:25

Hi, friends!
Welcome to #ConvAInews again!

◾️ Recently, we published an article about DeepPavlov Scenario-Driven Skills. In this article, we highlighted the role of Speech Functions in making dialogs with DeepPavlov Dream Socialbot smoothly and naturally. Follow the link to read the article.

◾️ We are happy to share a new article about Russian SuperGLUE models using the DeepPavlov.ai library. This article describes how Russian SuperGLUE solutions can be used by academy and IT companies. Feel free to click on the link to find out more (available in Russian).

◾️ Amazon released new “Create with Alexa tool” which can synthesize animated stories from children's prompts. It plays the stories on Echo Show smart displays using a child’s responses to a few questions when they ask the voice assistant to “make a story” for them.

◾️ New Caltech research shows how devices implanted into people’s brains, called brain-machine interfaces (BMIs), could one day help patients who have lost their ability to speak. A new brain-machine interface is the most accurate to date at predicting a person’s internal monologue.

◾️ Last week, OpenAI opened for testing ChatGPT. This chatbot can write jokes, computer code, college-level essays, and much more. During the first week of release, more than a million people signed up to test it, and hundreds of screenshots of ChatGPT conversations went viral on Twitter.

Stay tuned!

DeepPavlov

17 Nov 2022 07:46

Hello, everyone!

Daniel Kornev, CPO at DeepPavlov.ai, and Denis Kuznetsov, Researcher at DeepPavlov.ai, will share their expertise at Conversations 2022, one of the most popular conferences in ConvAI organized by our friends from Just AI. They will tell about the approaches to the automatic construction of dialog scripts together with controlled generation.

Conversations is an annual conference about voice assistants, chatbots, and the technologies behind them. More than 40 experts and visionaries from academia and business will speak at the event.

Daniel Kornev and Denis Kuznetsov will take part in the Technology track with the presentation on the topic “Using Deep Learning to create controlled scalable dialog systems”.

The presentation will be focused on a scenario-based approach. This is the oldest approach for dialogue systems development that has its pros and cons. DeepPavlov.ai experts will highlight both and reveal tactics you can use to combine the controllability of the scenario-based approach with scaling up possibilities by using generative language models, as well as automatic script construction..

Date: 2nd December 2022
Join us at Conversations 2022 and register below:
https://conversations-ai.com/

DeepPavlov

28 Oct 2022 17:49

Hi, friends!
This is our regular #ConvAInews!

◾️ In october we published a new article about the #Dream AI Assistant Platform. In this article, we highlighted the role of of Knowledge Base-specific Annotators for the utterances’ analysis within the Dream AI Platform. Follow the link to read the article.

◾️ Children’s speech recognition technology startup SoapBox Labs has become the first edtech company to receive a stamp of approval for its work against racial bias in technology. You can find more about it here

◾️ Synthetic voice developer Play.ht published an imaginary conversation between Apple founder Steve Jobs, who died 11 years ago, and Joe Rogan this week. AI-generated voice clones of Rogan and Jobs gave deepfake voice to a script composed using OpenAI’s GPT-3 language model for the first episode of the new “Podcast.ai” show.

◾️ Google Assistant will be improved and augmented with new features as part of its integration into the new Pixel Watch and Pixel 7 smartphone line. Google introduced the upgrades, showcasing better voice typing and new smart home connections for its latest products.

Stay tuned!