• AI KATANA
  • Posts
  • OpenAI reportedly preparing to launch new ‘Strawberry’ AI model

OpenAI reportedly preparing to launch new ‘Strawberry’ AI model

Also: Top 100 Gen AI Consumer Apps report by a16z

Author

AI KATANA
August 28, 2024

Morning!

In today's edition, we spotlight OpenAI's anticipated launch of its new "Strawberry" AI model, which is poised to surpass the capabilities of current systems and potentially advance the field toward artificial general intelligence (AGI). We also explore the latest a16z report highlighting the rapid growth of generative AI consumer apps, with creative tools leading the charge. Meanwhile, Cerebras unveils the "world's fastest" AI inference service, significantly enhancing real-time AI tasks. On a more cautionary note, the rise of AI-generated content raises concerns about the long-term quality of AI models, while Southeast Asia's strategic role in AI chip manufacturing underscores the geopolitical shifts shaping the industry. Additionally, Ray Dalio's AI-powered coaching tool aims to democratize his mentorship approach, and the legal tech startup Supio secures $25 million to revolutionize data analysis for lawyers using AI.

Sliced in this edition:

  • 🍓 OpenAI reportedly preparing to launch new ‘Strawberry’ AI model

  • 💯 Top 100 Gen AI Consumer Apps report by a16z

  • 📊 AI startup Cerebras debuts 'world's fastest inference' service - with a twist

  • 😭 Why AI can’t spell ‘strawberry’

  • 🤖 AI Appears to Be Slowly Killing Itself

  • 🌏 How ‘Friendshoring’ Made Southeast Asia Pivotal to the AI Revolution

OpenAI is reportedly preparing to launch a new AI model named "Strawberry," which promises to advance beyond the capabilities of current AI systems. This model, previously known as Q-Star, has shown impressive performance, particularly in solving complex math problems and tasks like developing marketing strategies and solving word puzzles. Strawberry is expected to outperform previous models, such as GPT-4, on benchmarks, potentially marking a significant step toward artificial general intelligence (AGI). This development is noteworthy not only for its technical advancements but also for its connection to the brief period of internal turmoil at OpenAI last year, which led to the temporary ouster of CEO Sam Altman. The release of Strawberry, anticipated this fall, could position OpenAI significantly ahead of its competitors in the rapidly evolving AI landscape.

The latest edition of the Top 100 Gen AI Consumer Apps report by Andreessen Horowitz highlights the rapid growth and diversification in the AI consumer app landscape. The report reveals that creative tools continue to dominate, with 52% of the top web apps focused on content generation or editing across various modalities like image, video, music, and speech. The music generator Suno saw a significant rise in popularity, jumping from #36 to #5. On mobile, legacy tools like Meitu, SNOW, and Adobe Express have successfully pivoted to AI-first models, ranking high on the list. ChatGPT maintains its top position on both web and mobile platforms, while competitors like Perplexity and Claude are gaining ground. Notably, Bytedance is making a strong push with several AI-driven apps entering the rankings. The emergence of new categories, such as aesthetics and dating apps like LooksMax AI and Umax, also reflects the expanding influence of AI in consumer applications. The report underscores the increasing engagement and user retention within these platforms, signaling a growing integration of AI into everyday consumer experiences.

AI startup Cerebras has launched what it claims to be the "world's fastest" inference service, which is said to be 10 to 20 times faster than existing cloud AI inference services like those from Amazon and Microsoft. This new service, powered by Cerebras' proprietary CS-3 computers and its WSE-3 chip, offers a significant leap in speed and efficiency, allowing for more complex and real-time AI tasks that were previously impossible. Cerebras' approach shifts the business model from selling hardware to offering inference capacity as a service, promising substantial cost savings and performance improvements. This advancement not only accelerates AI processes but also opens the door to "agentic" AI, where systems can interact with multiple data sources and applications to deliver more accurate and sophisticated results.

Large language models (LLMs) like GPT-4o and Claude, despite their ability to process vast amounts of data and perform complex tasks, can still make surprisingly simple errors, such as miscounting the number of letters in a word like "strawberry." This issue highlights a fundamental limitation in how these models operate. LLMs are built on transformer architectures that break down text into tokens—often at the word or subword level—without understanding the individual characters that make up these tokens. This tokenization process, while effective for generating text and understanding context, leads to errors in tasks that require precise knowledge of letters or syllables. The problem is further complicated by the inherent challenges of tokenizing across different languages, especially those without clear word boundaries. Although improvements are being made, such as through new models and synthetic data generation, these issues underscore the current limitations of AI in tasks requiring fine-grained linguistic understanding.

The increasing prevalence of AI-generated content on the web is posing a significant risk to the future of AI itself. As generative AI models are trained on data scraped from the internet, the growing inclusion of AI-generated content in these datasets is leading to a degradation in the quality of these models, a phenomenon compared to inbreeding. This issue, termed "Model Autophagy Disorder" (MAD), results in a feedback loop where AI models trained on synthetic content produce increasingly nonsensical or homogenized outputs over time. This problem is exacerbated by the lack of regulations requiring AI-generated content to be labeled, making it difficult to prevent this self-consuming cycle. If left unchecked, this could lead to a decline in the effectiveness and reliability of AI technologies.

Southeast Asia, particularly Malaysia, has become a critical player in the global AI revolution due to a strategic shift known as "friendshoring," where companies diversify supply chains away from China to politically aligned and geographically closer nations. Malaysia's robust semiconductor ecosystem, centered in Penang and Kulim, is attracting massive investments from global tech giants like Intel, Nvidia, and Microsoft. This region's rise is driven by its ability to support advanced chip manufacturing and AI data centers, which are crucial for the development of AI technologies. The competition is fierce, with countries like India and Vietnam also vying for a piece of this lucrative market. However, Malaysia's well-established infrastructure, skilled workforce, and strategic location have made it a preferred destination. Despite this growth, the country faces challenges, including talent shortages and the need for sustainable energy sources to support the industry's expansion. As global tensions and the demand for resilient supply chains intensify, Malaysia is poised to play a pivotal role, although it must navigate the complex geopolitical landscape carefully.

🧑🏽‍💻 AI Jobs

🛠️ AI tools updates

Ray Dalio, the founder of Bridgewater Associates, is developing an AI-powered version of his widely read "Principles" coaching program. This digital tool is designed to engage users in personalized, high-quality conversations about various topics, including goal achievement, problem-solving, entrepreneurship, and personal development. Dalio emphasizes that this AI will avoid the common pitfalls of generative AI, such as hallucinations, and will provide guidance based on his decades of experience. The beta version of this AI tool is set to launch in the fall, with a waitlist already open for interested users. This initiative represents Dalio's effort to scale his mentorship and coaching, making it accessible to a broader audience.

💵 Venture Capital updates

Supio, a Seattle-based legal tech startup founded by former Microsoft engineers Kyle Lam and Jerry Zhou, has raised $25 million in a Series A funding round led by Sapphire Ventures. The company, established in 2021, aims to revolutionize the legal industry by using AI to help lawyers efficiently analyze and organize vast amounts of unstructured data, particularly in personal injury and mass tort cases. Supio's platform automates time-consuming tasks, allowing legal professionals to quickly identify crucial information through an AI-driven chatbot. The startup charges a subscription fee based on case volume and plans to double its headcount within the next year. Supio's technology has already proven effective in real-world cases, such as a recent lawsuit involving over 40,000 pages of medical records. This funding round brings Supio's total funding to $33 million, positioning it strongly within the growing legal tech sector.

🫡 Meme of the day

⭐️ Generative AI image of the day