EncodeAgent AI Digest #3

Inside OpenAI: How does ChatGPT Ship So Quickly?

https://newsletter.pragmaticengineer.com/p/inside-openai-how-does-chatgpt-ship

Gergely’s Pragmatic Engineer Newsletter features an exclusive interview with Evan Morikawa, who leads half of the 130-person Applied engineering team at OpenAI, the company behind the AI sensation ChatGPT. OpenAI, led by CEO Sam Altman, has seen ChatGPT reach 100 million weekly active users in less than a year since its launch in November 2022, with continuous updates and new features being released at a rapid pace. The article delves into the engineering culture at OpenAI, which has remained largely secretive until now. Morikawa discusses how the ChatGPT team operates with the agility of an independent startup, maintains tight integration with research, and focuses on long-term product and research thinking. He also highlights the importance of uncoupled, incremental releases, high talent density, and effective day-to-day habits. The team’s structure, which includes engineering, product, design, and research, is designed to foster innovation and rapid iteration, allowing OpenAI to quickly ship new features and maintain its position at the forefront of AI technology. The interview covers the formation of the Applied team, Morikawa’s career at OpenAI, and the company’s approach to shipping new products and features.

All You Need to Know About Google Gemini 1.5 (Hint: It’s More Important Than Sora)

https://thealgorithmicbridge.substack.com/p/all-you-need-to-know-about-google?utm_source=post-email-title&publication_id=883883&post_id=141698219&utm_campaign=email-post-title&isFreemail=true&r=39n6q&utm_medium=email

The article discusses the significance of Google’s Gemini 1.5, a new AI model, in comparison to OpenAI’s Sora, suggesting that despite Sora’s public attention, Gemini 1.5 represents a more profound breakthrough. Google’s recent release of Gemini 1.5 Pro, a multimodal sparse Mixture of Experts (MoE) model, is highlighted for its efficiency and performance, capable of maintaining low latency while being highly performant. A key feature of Gemini 1.5 is its unprecedented 1-million-token context window, extendable up to 10 million in research, which dwarfs the previous record held by Anthropic’s Claude 2.1. This large context window allows for more complex and nuanced understanding and generation of text, which the author believes is a more significant advancement than OpenAI’s offerings. The article speculates on strategic reasons behind Google’s release schedule and suggests that both Gemini 1.5 Ultra and OpenAI’s GPT-5 may be released at the end of 2024. The author emphasizes that the 1–10 million-token context window is the true innovation of Gemini 1.5, making it a more important development than OpenAI’s Sora.

Find the AI Approach That Fits the Problem You’re Trying to Solve

https://hbr.org/2024/02/find-the-ai-approach-that-fits-the-problem-youre-trying-to-solve?utm_medium=email&utm_source=newsletter_weekly&utm_campaign=weeklyhotlist_Inactive&deliveryName=NL_Hotlist_20240219

The article emphasizes the importance of selecting the right analytics tool for a business problem rather than being swayed by the allure of generative AI or other advanced technologies. Leaders are advised to start by identifying the problem they need to solve and then consider the range of available analytics tools, asking the right questions to technical experts to determine the best fit. The article provides examples of companies like Sysco, KLM, Boeing, and Airbus, which have successfully applied various analytics techniques, including traditional automation, AI, and machine learning, to address specific business challenges. It outlines four categories of advanced analytics: generative AI, traditional deep learning, econometrics, and rule-based automation, each with its own strengths and limitations. The article also presents five critical questions to guide the selection of analytics approaches, considering factors such as the cost of being wrong, the need for explainability, repeatability, data accuracy, and the reflection of operational conditions. It concludes by suggesting that companies should focus on the problem at hand and use AI and analytics tools judiciously, building on each solution for future innovations.