Podcast Summary
Open source AI race: Mistral Large 2, an open source AI model with 123 billion parameters, outperforms Meta's Llama 3.1405b on leading benchmarks despite having fewer parameters, marking a significant advancement in the open source AI market.
The race for dominance in the open source AI market continues to heat up, with Mistral's latest model, Mistral Large 2, challenging Meta's Llama 3.1405b. Mistral Large 2, which has 123 billion parameters and fits on a single H100 node, is significantly more capable than its predecessor in code generation, mathematics, and reasoning. The model's training focused on minimizing its tendency to hallucinate or generate incorrect information, ensuring reliable and accurate outputs. Despite having only a third of the parameters of Llama 3.1405b, Mistral Large 2 performs equal or even superior to both Llama 3.1 and GPT4o across leading benchmarks. This rapid pace of innovation in open source AI is exciting for the developer community, as companies compete to provide the most capable models for various applications.
Foundation Model Competition: LLAMA 3.170b and Mistral 2 have shown impressive performance in foundation models, with LLAMA receiving praise for its superior results despite smaller size. However, Mistral 2's high cost and non-commercial license have raised concerns about its accessibility and purpose.
The competition in the world of foundation models has reached new heights, with LLAMA 3.170b and Mistral 2 showcasing impressive performance. LLAMA 3.170b, in particular, has been hailed as "insane" due to its superior results despite its smaller size. The open source community is abuzz with these developments, with some even suggesting that it surpasses the capabilities of GPT-4, Claude 3.5, and Sonic. However, the release of Mistral Large 2 was met with some criticism due to its high cost, non-commercial license, and limited accessibility, leading some to question its purpose. The release of Meta's commercial 405b model and Mistral's research-only 123000000000 parameter model within a day of each other added to the confusion. Some commentators have raised concerns about Mistral's direction, while others have pointed out that the commercial licenses and high costs may limit the models' impact and accessibility. Overall, the race to build the best foundation model continues, with each new release pushing the boundaries of what's possible.
AI Competition: OpenAI's free fine-tuning offer and Google's rumored math reasoning advances, along with the global launch of Chinese AI video generation model Cling and Bing's AI redesign, highlight an increasingly competitive and accessible AI industry
The AI world is seeing a surge of developments and announcements, with open source models like OpenAI's GPT-4 Mini putting pressure on commercial entities to innovate and make their offerings more accessible. OpenAI responded by offering free fine-tuning for the next two months, which many saw as a response to the growing influence of open source models. Additionally, there have been rumors of significant advances in math reasoning from Google's DeepMind. Chinese AI video generation model Cling also made a global launch, making its AI-generated content accessible to anyone with an email address. In the search engine realm, Bing unveiled an AI redesign, placing AI-generated answers front and center, while traditional search results were moved to the side. Microsoft's Bing is attempting to compete with Google by making bold moves in the search market. Overall, these developments indicate an exciting and competitive time in the AI industry, with a focus on making advanced technology more accessible to a wider audience.
Storytelling democratization, AI privacy: Colin Kaepernick's new AI startup, Lumi Story AI, aims to democratize storytelling by offering end-to-end tools for creation, publishing, merchandising, and monetization, while Venice AI prioritizes user privacy and free speech, providing a private, uncensored platform for text, image, and co-generation.
Colin Kaepernick's new AI startup, Lumi Story AI, is set to democratize storytelling by providing aspiring creators with end-to-end story creation tools, including physical and digital publishing and merchandising, allowing anyone to monetize their stories and level the playing field for diverse voices. Meanwhile, in the realm of AI privacy, Venice offers an alternative to the dominant AI companies that store and attach conversation histories to users' identities forever. Venice, a powerful AI app for text, image, and co-generation, prioritizes user privacy and free speech, providing a private, permissionless, and uncensored platform. These developments underscore the trend towards full product suites that integrate AI to empower individuals and protect privacy.
AI investment: Despite investor concerns over profitability and cost-effectiveness, AI industry sees significant investment, with OpenAI potentially spending $5B this year, while Superintelligent offers discounted access to AI tools and applications
The AI industry is experiencing significant investment, with OpenAI potentially spending $5 billion this year according to a report from The Information. This investment comes as some investors question the profitability of conversational AI and the need for these companies to increase prices or reduce costs. The Information's analysis is based on internal financial data and sources within the industry. Superintelligent, a platform aimed at helping individuals and teams maximize their use of AI, is offering a discount for those looking to learn more about AI tools and applications. With the code "year50," new users can get 50% off the already reduced annual fee, providing access for under $100 for a full year. The offer expires soon, so those interested should sign up at bsuper.ai. The broader context of this discussion is the potential for an AI bubble in the financial sector, with concerns about profitability and cost-effectiveness.
OpenAI costs: OpenAI, the AI research lab, is projected to have $8.5B operating costs this year, including $4B for server rentals, $3B for training, and $1.5B for workforce. ChatGPT generates $80M/month but potential losses could reach $45B if costs outpace revenue.
OpenAI, the leading AI research lab, is projected to have operating costs of around $8.5 billion this year, primarily due to the massive expenses of running and training its advanced AI models, like ChatGPT. These costs include an estimated $4 billion for server rentals, $3 billion for training, and $1.5 billion for its growing workforce. The revenue story is less certain, with ChatGPT currently generating around $80 million per month, but potential losses could reach as high as $45 billion if costs outpace revenue. OpenAI's Sam Altman has acknowledged the company's capital-intensive nature.
AI industry costs and competition: OpenAI, despite high costs, outperforms competitors with better revenue profile and lower server rentals. Anthropic grows faster but faces higher computing costs. Google's AI investments and revenue growth are under scrutiny.
OpenAI, despite its high operating costs, is outperforming some of its competitors in the AI industry. The company, which is backed by Microsoft, has a better revenue profile and pays less for server rentals than rivals like Anthropic. Anthropic, which is growing faster than OpenAI, is projected to reach around $406 million in net revenue this year, but its computing costs alone are estimated to be $2.5 billion. OpenAI, on the other hand, is exploring ways to reduce costs and increase revenue, including the development of a search engine and a computer using agent. The company is also expected to release GPT-5 before the end of the year. Meanwhile, The Wall Street Journal reported that Google's recent financial results showed revenue growth in line with expectations, but increased spending on AI infrastructure continued to be a concern. The article suggests that Google's advertising business faces tough growth comparisons, making the ROI from AI investments a critical factor for the tech giant. Overall, the discussion highlights the high costs and growing competition in the AI industry, and the importance of finding ways to generate revenue and reduce expenses to remain competitive.
AI bubble: Despite transformative potential, investors are becoming more cautious due to lack of immediate revenue benefits, leading to concerns of an AI bubble.
Despite the ongoing hype and investment in artificial intelligence (AI) by tech giants like Alphabet (Google) and Meta (Facebook), there are growing concerns among analysts about a potential bubble in the sector. Google's CEO, Sundar Pichai, acknowledged the transformative potential of AI but also emphasized the importance of long-term investment, with Alphabet's large cash reserves providing a safety net. However, recent market volatility, including significant losses for tech stocks like Nvidia, Alphabet, Microsoft, Apple, and Tesla, has raised concerns among investors about the lack of immediate revenue benefits from these AI investments. While this may not mark the end of belief in AI, it does signal a shift towards a more cautious approach from investors, who are now focusing more on returns in this space. The ongoing debate around the potential bubble in AI is a reminder of the importance of balancing innovation and investment with financial prudence.
Wall Street and AI pricing: Wall Street's involvement in AI market accelerated due to its unique dynamics, leading to discussions about market and valuation bubbles, but it's essential to separate these conversations from the actual value and utility of AI.
The involvement of Wall Street in the AI market is unusual due to the early presence of Big Tech. Typically, new technologies like generative AI would be developed and refined in the private sector for a decade or more before going public. However, the unique dynamics of AI have accelerated this process. Wall Street is grappling with how to price these new technologies, which has led to discussions about market and valuation bubbles. It's essential to separate these conversations from the actual value and utility of AI. Despite the challenges, it's an exciting time for AI, and we'll continue to explore its developments in future briefs.