In this episode of Gradient Dissent, Lukas Biewald talks with Jarek Kutylowski, CEO and founder of DeepL, an AI-powered translation company. Jarek shares DeepL’s journey from launching neural machine translation in 2017 to building custom data centers and how small teams can not only take on big players like Google Translate but win. They dive into what makes translation so difficult for AI, why high-quality translations still require human context, and how DeepL tailors models for enterprise use cases. They also discuss the evolution of speech translation, compute infrastructure, training on curated multilingual datasets, hallucinations in models, and why DeepL avoids fine-tuning for each individual customer. It’s a fascinating behind-the-scenes look at one of the most advanced real-world applications of deep learning. Timestamps: [00:00:00] Introducing Jarek and DeepL’s mission [00:01:46] Competing with Google Translate & LLMs [00:04:14] Pretraining vs. proprietary model strategy [00:06:47] Building GPU data centers in 2017 [00:08:09] The value of curated bilingual and monolingual data [00:09:30] How DeepL measures translation quality [00:12:27] Personalization and enterprise-specific tuning [00:14:04] Why translation demand is growing [00:16:16] ROI of incremental quality gains [00:18:20] The role of human translators in the future [00:22:48] Hallucinations in translation models [00:24:05] DeepL’s work on speech translation [00:28:22] The broader impact of global communication [00:30:32] Handling smaller languages and language pairs [00:32:25] Multi-language model consolidation [00:35:28] Engineering infrastructure for large-scale inference [00:39:23] Adapting to evolving LLM landscape & enterprise needs 🎙 Get our podcasts on these platforms: Apple Podcasts: http://wandb.me/apple-podcasts Spotify: http://wandb.me/spotify YouTube: http://wandb.me/youtube Follow Weights & Biases: https://twitter.com/weights_biases https://www.linkedin.com/company/wandb
We were thrilled to host the third edition of our annual AI conference, Fully Connected. This year, we expanded to two days and focused on bringing AI from prototype to production, empowering attendees to deploy and scale with confidence. We brought together two days packed with real-world insights, cutting-edge demos, and networking with the people building the next generation of production-ready AI and the applications built on top of the foundation models powering it all. Customers shared their journeys building innovative AI applications and models, revealing breakthrough insights and lessons learned.
Weights & Biases Weave is a lightweight toolkit for tracking and evaluating LLM applications, built by Weights & Biases. With Weave, you can log and debug LLM inputs, outputs, and traces, build rigorous evaluations for LLM use cases, and organize all the information generated across the LLM workflow, from experimentation to evaluation to production. Subscribe to this channel for demos on how to get started and use Weave.
Welcome to Gradient Dissent, a podcast series that explores the intricacies and opportunities in AI, one conversation at a time. Hosted by Lukas Biewald and powered by Weights & Biases, this series bring you insights from industry leaders like NVIDIA, Meta, Google, Lyft, OpenAI, and more. In each episode we explore the complexities of AI, celebrate its advancements, and offer a candid look at the challenges of integrating AI into real-world applications. From the power of AI in transforming search engines to the role of AI in reshaping industries, each episode is a treasure trove of knowledge, narrated by those who are leading the charge in this dynamic field. Hit the subscribe button, turn on notifications, and get ready to be inspired with every episode.

Shorts

Our newest course created in collaboration with Cohere and Weaviate focuses on practical RAG techniques for engineers: learn production-ready solutions from industry experts to optimize performance, cut costs, and enhance the accuracy and relevance of your applications. Enroll at: wandb.me/rag-yt

NotebookLM: RAG++ course

Weights & Biases
1.1K views9 months ago

RAG++ course: Welcome to the course

Weights & Biases
419 views9 months ago

RAG++ course: Hybrid search with Weaviate

Weights & Biases
636 views9 months ago
Fully Connected 2024 was a conference for the builders pioneering the Generative AI industry. Learn from foundation model builders, enterprises fine-tuning LLMs, and developers deploying GenAI applications. This event took place on April 18, 2024. Fullyconnected.com
This year the first AI track for the popular Meta Hacker Cup programming competition, designed to assess the capabilities of Generative AI in performing autonomous code generation tasks is being run by NeurIPS. It aims to test the limits of AI in complex coding challenges and measure the performance gap between AI systems and human programmers. This live lecture series is run in collaboration with the HAC organizers at PyTorch and Microsoft and aims to get participants up to speed on the challenge, how its structured and how to develop solutions to the challenge. Join the NeurIPS Hacker Cup Challenge discord here: https://discord.gg/b4SmXYMS

RAG for Code Generation, an AI Hacker Cup example

Weights & Biases
2.6K viewsStreamed 11 months ago

NeurIPS Hacker Cup AI: FineTuning

Weights & Biases
924 viewsStreamed 11 months ago

NeurIPS Hacker Cup AI: SWEAgent

Weights & Biases
3.1K viewsStreamed 11 months ago

NeurIPS Hacker Cup AI: Reinforcement learning

Weights & Biases
892 viewsStreamed 11 months ago
Weights & Biases is trusted by 800,000 users and 1000+ companies from the most cutting-edge and innovative AI startups and research institutions to the biggest brands around the world. Watch the testimonials videos to learning why customers choose Weights & Biases for their AI developer platform.
Overcome model chaos, automate key workflows, ensure governance, and streamline the end-to-end model lifecycle. This course will provide you with the concepts, best practices, and tools to level up your model management and drive success.
Weights and Biases (W&B) は機械学習の開発者向けツールを構築し、実験のトラッキング、モデルの最適化、データセットのバージョニング等を支援します。このチャンネルでは、機械学習とディープラーニングのチュートリアル、アプリの使い方やデモ、業界の専門家との会話を特集したポッドキャスト「Gradient Dissent」、そして研究者が最新の研究成果を発表する様子などを公開する予定です。 W&Bはアカデミックやオープンソースプロジェクトに対しては、常に無料でご利用いただけます。私たちのYouTubeチャンネルをいいね！して、購読してください。」