The conversation explores the emergence of DeepSeek as a significant player in the AI landscape, particularly in the context of open-source models and the competitive dynamics with established companies like OpenAI and NVIDIA.

The discussion delves into the implications of open-source strategies, the challenges faced by NVIDIA, and the innovation landscape in China.

Miguel Neves, Clara Gadelho and Gabriel Gonçalves address the ethical considerations surrounding AI training, the user experience of AI products, and the future of personal assistants in AI.

Overall, the dialogue highlights the evolving nature of AI technology and the importance of adaptability in a rapidly changing market.

Chapters

00:00 Emergence of New AI Players

04:37 The Hardware Landscape and NVIDIA's Dominance

08:56 Innovation vs. Copying: The Chinese AI Landscape

13:38 Reducing AI project costs with diverse open-source LLMs

21:39 OpenAI's Shift: From LLMs to Product Focus

26:07 User Experience and Future of AI Assistants

35:00 Regional Perspectives on AI Development

36:53 The Global AI Race: Competition and Innovation

38:29 Emerging Players in the AI Landscape

40:21 Ethics and Fairness in AI Training

Join Miguel Neves, Clara Gadellu, and Gabriel Gonçalves as they dive deep into OpenAI's groundbreaking o1 model. They explore how this shift toward reasoning-based AI differs from previous models, its implications for programming, and whether this is the future of AI development. Tune in to hear their takes on how o1 and reasoning models are shaping the landscape.

• 00:01 - Introduction to o1

• 00:38 - Paradigm Shift

• 01:47 - The Way To GPT-5

• 03:16 - Reasoning for Specific Fields

• 05:29 - O1 in Real Use Cases

• 08:55 - Challenges and Limitations

• 23:16 - The Business Perspective

Join us for our latest discussion with Gad Benram and Charles Frye from Modal as they explore the strategic reasons behind companies choosing to host their own AI infrastructure versus relying on external cloud services. From controlling critical data to customizing AI applications, this episode is packed with valuable insights for anyone navigating the complex world of AI deployment.

Key topics include:

• 00:00 Introduction: Insights on AI Resources for Hosting AI Models

• 03:11 The Challenges of Existing Cloud Services

• 09:14 Introducing Modal: A Fast and Interactive Development Experience

• 15:13 Different Infrastructure Needs for Data Teams

• 19:42 Addressing Slowness in AI Services

• 26:20 Python and Notebooks for Data Scientists

• 33:35 Fast and Seamless Deployment with Modal

• 40:46 Future Directions and Closing Remarks

In this episode, Gad Benram and Charles Frye discuss the challenges of hosting AI models in production and the limitations of existing cloud services. They highlight the lack of resources and GPUs available for serving AI applications and the slow bootstrapping process. They introduce Modal, a serverless runtime for distributed applications built on top of cloud resources, as a solution to these challenges.

Modal offers fast deployment times, interactive development workflows, and support for large-scale models.

🔗 Visit our website for more resources and updates:

This episode on the value and potential of LLMs in business operations features Gabriel Gonçalves and Miguel Neves from TensorOps, along with our special guest, Morgan McGuire, Director of Growth ML at Weights & Biases. Together, they discuss the implications of automation, the need for realistic client expectations, and the limitations and future of generational AI models.

Key topics include:

🔗 Visit our website for more resources and updates:

This episode on the transformative impacts of AI on search technologies features Gad Benram and Gabriel Gonçalves , along with our special guest, Edward Zhou —who has recently led the search ranking team at Notion—and share his experiences and expert insights on AI's impact on search technologies.

Key topics include:

• 00:00 - Introductions and Opening Remarks

• 15:47 - Evaluating Search Systems and Techniques

• 20:40 - Scoring Algorithm and Semantic Searching

• 24:18 - Vector Space Model and Similarity Limitations

• 27:05 - Embedding Models and Relevance Challenges

• 32:00 - Addressing Search Bias Mitigation

• 36:25 - Evaluating Search Results and Language Models

• 41:49 - Language Models and Embedding Technologies

Throughout the episode, the team discussed the potential of AI-powered search tools, including the combination of traditional search algorithms with AI-powered language models, and the importance of evaluating search systems based on user actions and business outcomes. They also explored the workings of a scoring algorithm, the relevance of similarity in a vector space, and the challenges and potential solutions in incorporating embedding models for specific business domains. Additionally, they addressed the issues of position and click bias in search results, the difficulties in evaluating search results and language models, and the current and future state of language models and embedding technologies. Finally, they looked into the future of search systems, considering how advancements in AI and embeddings could revolutionize search experiences.

🔗 Visit our website for more resources and updates:

This one-hour session exploring the Secrets of Prompt Engineering , we'll discuss how prompt engineering resembles programming and what common design patterns they share

This episode offers a deep dive into the costs aspects of leveraging Large Language Models (LLMs) in production environments.

💲 Struggling with managing costs of LLMs in production? Find out about our workshop here: