DEV Community: Dmitry Spodarets

Weekly AI Highlights Review: November 5–12

Dmitry Spodarets — Thu, 14 Nov 2024 18:04:28 +0000

Two of the most impactful news this week concern OpenAI and multiple reports stating the company is leaving no stone unturned to on the one hand, keep up with the computing demands of its products and figure out new ways to enhance its models' performance on the other.

Regarding the first one, recent reports claim that OpenAI has put its chip foundry ambitions on hold to focus on more immediate ways to support its massive computing infrastructure requirements: the startup is reportedly partnering with Broadcom to design a new in-house chip that will enable it to reduce its dependence on Nvidia's GPUs. According to the reports, it was through Broadcom that OpenAI managed to secure manufacturing with TSMC, which is tentatively set to begin in 2026. In the meantime, OpenAI is also considering additional alternatives, including using AMD chips through Microsoft Azure.

Nvidia has become a key player and the biggest benefactor of the AI boom. Estimates place the company's market share at an astounding 80%, while companies like AMD and chip design startups like Fractile are vying for a bigger slice of the market. Perhaps the most telling indicator of the power and influence Nvidia has gained over the market is that this week, Nvidia became the world's largest company after surpassing Apple's $3.38 trillion capitalization at market close.

Even if it has proven nearly impossible for tech giants and AI startups alike to escape from relying on Nvidia's products, companies are keen to find at least some respite from their dependence on the chipmaker's timelines and manufacturing and shipping capabilities. In addition to the report that OpenAI may be looking to design and manufacture a custom chip, this week also saw Amazon consider making another multi-billion investment in Anthropic as long as the latter considers incorporating more servers powered by Amazon's proprietary chips.

According to the reports about this matter, Anthropic agreed to adopt AWS as its primary cloud provider for essential workloads, including model training and safety research, after Amazon committed to the $4 billion investment completed earlier this year. However, it appears that Anthropic has since preferred to use Amazon's Nvidia-powered servers. Adopting Amazon's custom chips would likely cost Anthropic some degree of freedom, at least in its choice of cloud computing providers, as the hardware is not as ubiquitous and its software stack is not as well-supported as Nvidia's.

In addition to OpenAI's hardware woes, there have been some reports that Orion, allegedly OpenAI's next-generation GPT model has not shown a degree of performance improvement comparable to the leap from GPT-3 to GPT-4, despite the corresponding increase in computing resources and data availability. The matter is so serious that some claim this new model may not be reliably better than GPT-4 in domain-specific tasks like coding.

There had been a myriad of opinions regarding this development, ranging from questioning the so-called "scaling laws", to accepting that given the exponential and relatively smooth performance increases seen up to date, it was expected that a difficulty like this one would eventually come along.

Other headlines that caught our eye this week include:

The secret to Google's efficiency? Over 25% of its code is now AI-generated: During Google's 2024 Q3 earnings call, CEO Sundar Pichai revealed that AI now generates over 25% of Google's new code (which is then reviewed by engineers), highlighting the company's commitment to AI integration and development efficiency.

Meta announced the availability of its Llama models for U.S. government agencies and contractors: Following reports of China using Meta's Llama 2 model for military intelligence, Meta announced it would make its AI models available to US government agencies and contractors despite ongoing debates about the safety and usefulness of commercial models in military contexts.

Research Grid secured $6.5M to automate clinical trials' administrative work: Research Grid, a startup founded by Dr. Amber Hill that uses AI to streamline clinical trial processes through its TrialEngine and Inclusive platforms, has secured $6.5 million in seed funding led by Fuel Ventures to expand its presence in the US and Asia and enhance its AI capabilities.

Predikt raised €750K to support CFOs in their decision-making processes: Belgian-Swiss startup Predikt has secured €750,000 in funding to develop its AI-powered financial forecasting platform. Predikt combines internal financial data with millions of macroeconomic indicators to help large companies make more accurate financial predictions.

Mistral has released batch and moderation APIs: a moderation API that leverages a Ministral 8B fine-tune to classify text according to nine categories describing common harms, and a batch API that enables developers to process high volumes of data for 50% of the cost of equivalent synchronous API calls.

An AI robot's portrait of Alan Turing made history after being auctioned for $1M: A portrait of Alan Turing created by Ai-Da, the world's first ultra-realistic robot artist, just made history by selling for $1.08 million at Sotheby's— becoming the first artwork by a humanoid robot to be sold at auction.

Panjaya's BodyTalk is a dubbing platform that synchronizes speakers' voices, lips, and bodies: Panjaya's new AI dubbing platform BodyTalk synchronizes lip movements and body gestures for natural-looking video translations. BodyTalk early adopter TED reports doubled viewer completion rates and a 115% increase in views of dubbed content.

Conflixis secured $4.2M in seed funding to support smarter financial decisions in healthcare: Healthcare startup Conflixis raises $4.2M in seed funding to use AI and advanced analytics to help healthcare organizations better manage and understand conflicts of interest in their decision-making processes.

AI coding assistant Cursor's parent company has triggered a bidding war: AI coding assistant Cursor's parent company Anysphere is fielding unsolicited offers from major VC firms at valuations up to $2.5 billion, following explosive revenue growth from $4M annually to $4M monthly in less than a year.

The Beatles' final AI-produced song snagged two Grammy nominations: AI technology was used to clean up the vocals from John Lennon's 1978 "Now and Then" demo to create The Beatles' final song, which received two nominations for the 2025 Grammy Awards: Record of the Year and Best Rock Performance.

Webinar "Exploring Infrastructure Management for GenAI Beyond Kubernetes"

Dmitry Spodarets — Tue, 05 Mar 2024 07:28:06 +0000

Data Phoenix team invites you all to our upcoming webinar that’s going to take place on March 14th, 10 am PST.

- Topic: Exploring Infrastructure Management for GenAI Beyond Kubernetes
- Speaker: Andrey Cheptsov (Founder & CEO at dstack)
- Participation: free (but you’ll be required to register)

Enterprises, inspired by OpenAI and recent advancements in the open-source AI community, are exploring training and deploying GenAI models on their infrastructure. The traditional Kubernetes stack is not tailored for the AI era and presents numerous drawbacks hindering enterprises from adopting GenAI.

In this webinar, we’ll discuss:

The significance of open source for infrastructure management
Drawbacks of the Kubernetes stack in the context of AI
Introducing dstack, a new open-source orchestration engine
Exploring how dstack addresses the drawbacks of Kubernetes in the training and deployment of GenAI models
Highlighting dstack’s role in simplifying and optimizing infrastructure management processes

Speaker:
Andrey is a core contributor to dstack. He is passionate about open-source, developer tools, and the cloud. Previously, Andrey worked at JetBrains with the PyCharm team.

Evaluating LLM Models for Production Systems: Methods and Practices

Dmitry Spodarets — Fri, 23 Feb 2024 00:26:24 +0000

Data Phoenix team invites you all to our upcoming webinar that’s going to take place on February 27th, 10 am PST.

Topic: Evaluating LLM Models for Production Systems: Methods and Practices
Speaker: Andrei Lopatenko
Participation: free (but you’ll be required to register)

This webinar is designed to offer a comprehensive understanding of the evaluation processes for LLMs, particularly in the context of preparing these models for deployment in production environments.

Key Highlights of the webinar:

In-Depth Analysis of LLM Evaluation Methods: Gain insights into a variety of methods to evaluate LLM models, understanding their strengths and weaknesses.
End-to-End Evaluation Techniques: Explore how LLM augmented systems are assessed from a holistic perspective.
Pragmatic Approach to System Deployment: Learn practical strategies for applying these evaluation techniques to systems intended for real-world application.
Focused Overview on Critical LLM Aspects: Receive an overview of various evaluation techniques that are essential for assessing the most crucial elements of modern LLM systems.
Simplifying the Evaluation Process: Understand how to streamline the evaluation process, making the work of LLM scientists more efficient and productive.

Speaker: Dr. Andrei Lopatenko is a seasoned expert and executive leader with over 15 years of experience in the tech industry, focusing on search engines, recommendation systems, and large-scale AI, ML, and NLP applications. He has contributed significantly to major companies like Google, Apple, Walmart, eBay, and Zillow, benefiting billions of customers. Dr. Lopatenko earned his PhD in Computer Science from the University of Manchester. He played a key role in developing Google's search engine, initiating Apple Maps, co-founding a Conversational AI startup acquired by Facebook/Meta, and leading Search, LLM, and Generative AI at Zillow.

Webinar "Gorilla: Large Language Model Connected with Massive APIs"

Dmitry Spodarets — Mon, 09 Oct 2023 05:22:57 +0000

Data Phoenix team invites you all to our upcoming webinar that’s going to take place on October 12, 8 am PST / 5 pm CET.

Topic: Gorilla: Large Language Model Connected with Massive APIs
Speaker: Shishir Patil (Ph.D. student at UC Berkeley)
Participation: free (but you’ll be required to register)

The Gorilla project is designed to connect large language models (LLMs) with a wide range of services and applications exposed through APIs. Imagine if ChatGPT could interact with thousands of services, ranging from Instagram and Doordash to tools like Google Calendar and Stripe, to help you accomplish tasks. This may be how we interact with computers and even the web in the future. Gorilla is an LLM that we train using a concept we call retriever-aware training (RAT), which picks the right API to perform a task that a user can specify in natural language. Gorilla also introduces an Abstract Syntax Tree (AST) based sub-tree matching algorithm, which for the first time allows us to measure hallucination of LLMs!

Speaker:
Shishir is a Ph.D. student in Computer Science at UC Berkeley advised by Joseph Gonzalez and Prabal Dutta affiliated with the Sky Computing Lab (previously RISE), Lab11, and Berkeley AI Research (BAIR). He is broadly interested in ML-Systems, and LLMs. Previously he has interned at Google Brain, and Amazon Science, and was at Microsoft Research as a Research Fellow before.

Data Phoenix Digest — ISSUE 2.2023

Dmitry Spodarets — Thu, 26 Jan 2023 18:53:59 +0000

Video recording of our webinar about dstack and reproducible ML workflows, AVL binary tree operations, Ultralytics YOLOv8, training XGBoost, productionize ML models, introduction to forecasting ensembles, domain expansion of image generators, Muse, X-Decoder, Box2Mask, RoDynRF, AgileAvatar and more.

VIDEO
dstack — a command-line utility to provision infrastructure for ML workflows

Video recording of our webinar about dstack and reproducible ML workflows by Andrey Cheptsov.

If you have interesting topics or projects that you would like to share with the world in our webinars, you can submit them here.

ARTICLES

AVL Binary Tree Operations

In this article, the author described AVL trees and operations you can perform on them, such as inserting a node in different variants (for example, left, right or right, right).

Ultralytics YOLOv8: State-of-the-Art YOLO Models

YOLOv8 is the latest family of YOLO based Object Detection models from Ultralytics providing state-of-the-art performance. This article looks into the latest improvements and features added to YOLOv8, and provides a guide on using it in practice.

End-to-End MLOps with Snowpark Python and MLFlow

How would you leverage Snowpark Python for operationalizing your machine learning models within the flow of your existing MLOps processes? The article provides detailed answers and looks into end-to-end MLOps, from A to Z.

Building a Predictive Maintenance Solution Using AWS AutoML and No-Code Tools
Industrial machine, equipment, and vehicle operators need to reduce maintenance costs while operating under strict constraints. This article presents a predictive maintenance solution built using AutoML and no-code tools powered by AWS. Check it out!

PAPERS & PROJECTS

Zero-Shot Text-Guided Object Generation with Dream Fields

Dream Fields can generate the geometry and color of a wide range of objects without 3D supervision. It combines neural rendering with multi-modal image and text representations to synthesize diverse 3D objects solely from natural language descriptions. Take a look!

AgileAvatar: Stylized 3D Avatar Creation via Cascaded Domain Bridging

AgileAvatar is a novel self-supervised learning framework to create high-quality stylized 3D avatars with a mix of continuous and discrete parameters. To ensure the discrete parameters are optimized, a cascaded relaxation-and-search pipeline is implemented.

RoDynRF: Robust Dynamic Radiance Fields

In this work, the authors address the robustness issue of dynamic radiance field reconstruction methods by jointly estimating the static and dynamic radiance fields along with the camera parameters (poses and focal length). Learn how they do it!

Box2Mask: Box-supervised Instance Segmentation via Level-set Evolution

Box2Mask is a novel single-shot instance segmentation approach, which integrates the classical level-set evolution model into deep neural network learning to achieve accurate mask prediction with only bounding box supervision. Check the paper out!

If you enjoyed this content make sure to subscribe to our newsletter and share it with others who may be interested. Follow us on social networks (Telegram, Facebook, Twitter, LinkedIn, YouTube) to stay updated about the upcoming webinars and have more interesting content.

Read the full digest here.

Charity MLOps Webinar "Vertex AI Pipelines infrastructure with Terraform"

Dmitry Spodarets — Wed, 14 Dec 2022 18:41:35 +0000

Data Phoenix team invites you all to our upcoming "The A-Z of Data" charity webinar that’s going to take place on December 21, 2022 at 16.00 CET.

Topic: “Vertex AI Pipelines infrastructure with Terraform”
Speaker: Alona Slastin, ML Engineer at SoftServe
Participation: free (but you’ll be required to register)
Karma perk: donate to our charity initiative

REGISTER 👉 https://www.eventbrite.com/e/charity-ai-webinar-vertex-ai-pipelines-infrastructure-with-terraform-tickets-481912522327

ABOUT THE SPEAKER AND TOPIC

Alona is an ML Engineer with experience in Data Analysis and Data Science with knowledge of statistics, statistical analysis, advanced analytics and forecasting. Qualified in working with Google Cloud AI services for building continuous ML training pipelines and implementing CI/CD technologies.

Vertex AI is a Google Cloud Platform service specified on building, deploying and scaling ML models with with pre-trained and customizable models. It provides tools for every step of the machine learning workflow across different model types and for different levels of machine learning expertise.
Like any cloud platform, it requires the preliminary deployment of an infrastructure, the enabling of specific services and setting up the permissions. You can do it as a code with open-source software tool Terraform, that allows you to safely and predictably create, change, and improve infrastructure.
During this session we will go through the best practices for developing Vertex AI Pipelines and connect different services together. You will learn how to wrap your infrastructure with Terraform and set up automatic deployment of models using Vertex AI and GCP services.

ABOUT OUR CHARITY INITIATIVE

This webinar is planned as a charity event. As we did before, we do our best to raise funds for KOLO — A project created by Ukrainian IT experts to help Ukraine fight the war against Russia by supplying high-tech equipment to the front lines.

This cause is very special to us. So, please donate any amount you can to help us save Ukrainian lives. Every donation is a huge help!

Donate 👉 https://bit.ly/dataphoenix-koloua

Please note that participation is free but registration is required! We are also looking forward to any donations that you’re willing to give to our charity initiative.

REGISTER 👉 https://www.eventbrite.com/e/charity-ai-webinar-vertex-ai-pipelines-infrastructure-with-terraform-tickets-481912522327

Charity AI webinar "Learning through machine learning: how we built a recommendation system from scratch"

Dmitry Spodarets — Fri, 28 Oct 2022 06:00:27 +0000

The Data Phoenix team invites you all to our upcoming "The A-Z of Data" charity webinar that’s going to take place on November 9, 2022 at 16.00 CET.

Topic: “Learning Through Machine Learning: How We Built a Recommendation System from Scratch”.
Speaker: Dora Petrella, Senior Data Scientist at METRO.digital
Language: English
Participation: free (but you’ll be required to register)
Karma perk: donate to our charity initiative

Registration 👉 https://www.eventbrite.com/e/webinar-how-we-built-a-recommendation-system-from-scratch-tickets-452539968227

ABOUT THE SPEAKER

Dora Petrella, Senior Data Scientist at METRO.digital

Dora is a curiosity-driven Data Scientist with 4 years of experience. Her background is in Engineering and Computer Science, and she is currently based in Germany.

She chose to work for Metro.digital, the tech software company daughter of the international wholesaler Metro. Her Italian origins and her passion for everything around Machine Learning fit well in a company where food and technology meet together.

Dora contributed to different Machine Learning projects in the wholesale sector, such as recommendation systems, ranking problems in the context of product search, personalization in the context of marketing, and customer segmentation. Her main tasks are researching solutions, contributing to the implementation and monitoring of machine learning-based algorithms, and presenting results even to non-technical audiences.

We at Metro.digital have been building data science products for years and have gathered a fair amount of experience, especially in the area of recommendation systems. Over the last few months, my team’s been focusing on an exciting new challenge: building from scratch a machine learning-based recommendation system for real-time identification of substitute products. The resulting benefits are multiple: customers quickly discover more from our assortment; Metro employees reduce their time spent on manual search; and, Metro itself increases revenue and retains customers. The aim of this talk is to tell the story of our journey from concept to production, with a focus on our learnings, mistakes, and results.

ABOUT OUR CHARITY INITIATIVE

This cause is very special to us. So, please donate any amount you can to help us save Ukrainian lives. Every donation is a huge help!

Please note that participation is free but registration is required! We are also looking forward to any donations that you’re willing to give to our charity initiative.

Registration 👉 https://www.eventbrite.com/e/webinar-how-we-built-a-recommendation-system-from-scratch-tickets-452539968227

Charity AI webinar "The promising role of synthetic data to enable responsible innovation"

Dmitry Spodarets — Wed, 12 Oct 2022 14:58:30 +0000

Data Phoenix Events team invites you all on October 19 to our "The A-Z of Data" charity AI webinar. The topic - "The promising role of synthetic data to enable responsible innovation".

Good quality FAIR data is fundamental for enhancing data reuse. When we discuss data quality in the FAIR context, we often focus on the metadata level quality attributes like accessibility and reuse conditions rather than the semantic ones like imbalances, outliers, and duplicates. In practice, ensuring both the metadata and semantic levels of data quality is crucial but also challenging. One solution for this challenge is synthetic data. MIT technology review names synthetic data as one of the ten tech breakthroughs of 2022 citing it as a solution for training AI models when faced with inadequate quality, or incomplete data or biased data. Synthetic data improves data quality and helps accelerate AI projects enabling responsible innovation. Let's understand how it works in practice with the experience of the co-founder of a synthetic data company and how to check for data quality at scale using open-source libraries, as well as metrics required to measure the ensuing synthetic data quality.

Speaker
Shalini Kurapati - Co-founder and CEO Clearbox AI. Shalini leads the strategy, operations and business development at Clearbox AI with her multidisciplinary expertise at the intersection of Technology, Policy and Management. Shalini holds a PhD from Delft University of Technology with a strong R&D background and practical expertise in data management, data privacy and data stewardship. She specialises in transparency, privacy and fairness issues across data life cycles as well as algorithms. Shalini has a wide-ranging international professional experience in the Netherlands, Sweden, India, United States and most recently Italy and is also a Certified Informational Privacy Professional/Europe (CIPP/E) with demonstrable knowledge of GDPR and European e-Privacy laws.

Registration 👉 https://dataphoenix.info/charity-ai-webinar-the-promising-role-of-synthetic-data-to-enable-responsible-innovation/

Charity AI webinar "Deploying DL models with Kubernetes and Kubeflow"

Dmitry Spodarets — Fri, 16 Sep 2022 06:39:19 +0000

The Data Phoenix Events team invites you all on September 28 to our "The A-Z of Data" charity AI webinar. The topic — deploying deep learning models with Kubernetes and Kubeflow.

In this talk, we'll learn about deploying Keras models. First, we'll see how to do it with TF-Serving and Kubernetes, and in the second part of the talk, we'll do it with KFServing and Kubeflow.

We resume our series of webinars as charity webinars to raise money for KOLO. This project was created by Ukrainian technology industry experts to help Ukraine fight the war against Russia by supplying high-tech equipment to the front lines.

Speaker
Alexey Grigorev - Principal Data Scientist at OLX Group, Founder at DataTalks.Club. Alexey wrote a few books about machine learning. One of them is Machine Learning Bookcamp — a book for software engineers who want to get into machine learning.

Registration 👉 https://dataphoenix.info/charity-ai-webinar-deploying-dl-models-with-kubernetes-and-kubeflow/