Memory Leak — #34

VC Astasia Myers’ perspectives on AI, cloud infrastructure, developer tools, open source, and security. Sign up here.

🚀 Products

Inflection-2.5: Meet the World’s Best Personal AI

Inflection adds IQ to Pi’s EQ. Inflection-2.5 leverages their upgraded in-house model that is competitive with all the world’s leading LLMs like GPT-4 and Gemini. It couples raw capability with their signature personality and empathetic fine-tuning. Inflection-2.5 approaches GPT-4’s performance, but it used only 40% of the amount of compute for training. They made particular strides in areas of IQ like coding and mathematics. Pi now also incorporates world-class real-time web search capabilities to ensure users get high-quality breaking news and up-to-date information.

Why does this matter? It is clear foundation models need to be full stack with an application layer so they can collect data for Reinforcement Learning from Human Feedback (RLHF). We similarly saw Mistral release Le Chat on February 26th. Looking back on the Phi release last May, it did not seem to achieve the level of interest originally anticipated. Inflection shared user metrics for the first time, saying it has 1 million daily active users and 6 million monthly active users. OpenAI announced last November that it had 100 million weekly active users. The Inflection team continues to innovate. The advancement that stood out the most to us is adding real-time web search capabilities similar to Perplexity, which this week was rumored to have raised a new round at $1B+ valuation.

Microsoft Introduces Copilot AI Chatbot for Finance Workers in Excel and Outlook

Microsoft is launching a Copilot for Finance, which it said will be able to perform a handful of common role-specific actions in Excel and Outlook. The Copilot for finance will initially run a variance analysis, reconcile data in Excel and speed up the collections process in Outlook. According to a Microsoft finance lead, a couple of thousand people on a financial planning and analysis team each spend one or two hours doing reconciliation each week, and with the new Copilot, that takes more like 10 or 20 minutes per week.

Why does this matter? Microsoft already has a Copilot for general-purpose industrial use in Office applications, and it has released Copilots designed for sales and customer-service workers. The reconciliation process is monotonous by nature, but vital to the financial health of a company. By enabling financial professionals to complete this task quicker, they can close the books faster and provide time for more engaging work that can contribute more to the company.

Modal Introduces WebSockets

Modal now supports WebSocket connections. WebSocket is a communication protocol for real-time, bidirectional transfer of data between a client and server. Unlike HTTP, in which connections are opened and closed per request/response, WebSocket establishes a persistent connection between client and server. This is advantageous for applications that require low latency and real-time updates.

Why does this matter? One of Modal’s primary benefits is automatic scaling based on the volume of inputs your functions are receiving. Modal supporting WebSockets make it super easy for AI engineers to build applications that can handle variable request volumes. Modal can auto-scale containers for both your WebSocket server and model so services can handle multiple users at once. This new functionality also enables Modal to support more real-time use cases.

📰 Content

Startups Are Still Buying Fewer Startups

For venture-backed startups, it’s historically rather common to wind up selling to another startup. Lately, however, the pace of these deals has been slowing down. In all of 2023, just over 650 funded startups globally were acquired by other funded startups, the lowest total in three years according to Crunchbase.

Why does this matter? The M&A market continues to be hard across the board. M&A dropped 31% for VC-backed startups in 2023. In 2023, only 1,738 venture-backed startups were acquired globally, the lowest amount since 2015, when 1,687 startups were bought, per Crunchbase data. Founders should be prepared for long acquisition processes 6–12 months and consider managing cash burn accordingly. It is also important to perform analysis to determine if running an M&A process even makes sense based on the potential exit value and current cash position. Often it is more efficient for very early stage teams to shutdown and return capital. It is a hard decision but can save time, money, and headaches.

Why Facebook Doesn’t Use Git

The Graphite team highlights that Facebook uses Mercurial for their code version control and incrementally added custom tooling on top. Facebook migrated off Git because of their codebase size, performance, collaboration experience, and the ability to have engineers contribute to Mercurial patches.

Why does this matter? Github and Gitlab dominate the source code repository market and are based on Git. This piece highlights the value of alternative approaches and particularly how technology choices can directly impact collaboration.

Mistral and Snowflake Partnership

Mistral AI’s newest and most powerful model, Mistral Large, is available in the Snowflake Data Cloud for customers to securely harness generative AI with their enterprise data Snowflake Ventures partners with Mistral AI to expand generative AI capabilities and empower more developers to seamlessly tap into the power of leading large language models Snowflake Cortex LLM Functions, now in public preview, enables users to quickly, easily, and securely build generative AI apps

Why does this matter? Last week, Sridhar Ramaswamy (ex-Neeva) became the new CEO of Snowflake. Sridhar has a strong technical background in AI from Google. Snowflake continues to push forward in AI with the Mistral partnership and Cortex product, a new fully managed service that makes AI simple and secure for users to drive business value. Using Cortex’s specialized functions, any user with SQL skills can leverage smaller LLMs to cost-effectively address specific tasks such as sentiment analysis, translation, and summarization in seconds. This trend of small models for specific tasks paired with larger models for task planning or complex tasks continues to stand out. Mistral’s partnership with Snowflake follows their new partnership with Microsoft.

💼 Jobs

⭐️DragonflyDB — React Tech Lead — Dragonfly Cloud

⭐️Chroma — Member of Technical Staff

⭐️Speakeasy — Product Engineer

Views expressed in posts and other content linked on this website or posted to social media and other platforms are my own and are not the views of Felicis Ventures Management Company, LLC. The posts do not constitute investment, legal, tax, or other advice and do not constitute an offer to invest in any security.