Memory Leak — #16
VC Astasia Myers’ perspectives on machine learning, cloud infrastructure, developer tools, open source, and security. Sign up here.
🚀 Products
Fixie is a cloud-hosted Platform-as-a-Service that enables individuals to build and integrate smart agents that leverage the power of LLMs to solve problems based on natural language. In Fixie, each Agent is a standalone service that combines an LLM with a little bit of code — which can be implemented in any programming language — that understands how to connect to an external system, like a database or an API.
Why does this matter? Foundational models including LLMs have dramatically changed how individuals can build ML applications. It significantly lowers the bar to get started and development can be minutes instead of days. There’s a new stack of tooling and set of methodologies arising from prompt engineering to embeddings and nearest neighbor search to foundational model orchestration (FOMO). I discussed FOMO as a core layer in the stack in a recent primer. Fixie is a FOMO offering and hits on our point that solutions should have multi-modal support.
Code-generating platform Magic challenges GitHub’s Copilot with $23M in VC backing
Magic is an AI-driven tool designed to help software engineers write, review, debug and plan code changes. The tool, not yet generally available, can “communicate” in natural language and collaborate with users on code changes. The CEO claims Magic operates like a pair programmer that’s able to understand and continuously learn more about the context of both coding projects and developers. Magic is taking steps to prevent copyrighted code from showing up in the tool’s suggestions and citing the source of suggested code where possible.
Why does this matter? Copilot has been used by over 1.2 million people, and GitHub is aggressively positioning it as an enterprise-scale tool, recently launching a corporate-focused plan called Copilot for Business. The category of ML-enabled coding products is heating up with other offerings including Tabnine, Mutable, BigCode, Grit.io, and Sourcegraph Cody, which announced a private beta two weeks ago. There is a split in the market between teams that are leveraging hosted foundational models (e.g. Cody uses Anthropic’s Claude) and those that are building their own ML models like Magic. Training AI is very expensive. Kite founder Adam Smith estimated that it could cost over $100 million to build a “production-quality” tool capable of synthesizing code reliably.
MindsDB NLP Inside Your Database
MindsDB is an open source ML platform for developers that provides a powerful NLP engine that can use pre-trained NLP models to extract insights from text-based data. MindsDB’s NLP engine for pre-trained models is powered by Hugging Face, a leading open-source NLP library that provides a wide range of pre-trained NLP models that can be used to extract insights from your text-based data and OpenAI’s GPT-3 models.
Why does this matter? Historically, companies have called upon their data teams to handle the time-consuming task of rigging up connections between their business data (stored in databases) and machine learning tech. MindsDB wants to inject the AI functionality directly into the database so that developers can do the work themselves. We are starting to see developers become ML practitioners and believe that developers building ML applications will outnumber traditional data scientists shortly.
📰 Content
Transform’s specialty is semantic tools to build data science insights from data troves that can be used for business intelligence. This tech will now be folded into dbt’s Semantic Layer product, which already let engineers use simple phrases like revenue, customer count and churn rate to define the metrics they were looking to answer through the data.
Why does this matter? Transform’s technical team is super strong so this is a great win for all parties. Metrics were a hot area of investment and at one point I knew of six startups focused on building a metrics catalog. Most of the have pivoted or shutdown. Interestingly Cube.dev, which started as an embedded analytics startup, evolved its product line to include a semantic layer. It looks like a semantic layer product fits best within a larger product portfolio.
Google Shows off New AI Search Features, but a ChatGPT Rival Is Still Weeks Away
Google demoed its latest advances in AI search at a live event in Paris on Wednesday — but the features pale in comparison to Microsoft’s announcement of the “new Bing” powered by upgraded ChatGPT.
Why does this matter? Google has been unrivaled in search for over a decade. ChatGPT’s meteoric rise in popularity has reportedly prompted Google’s management to declare a “code red” situation for its core product: online search. Advances in generative AI pose a threat to Google’s dominance in search. Microsoft and Baidu Ernie in addition to startups like You.com, Perplexity, and Neeva are emerging to compete for one of the largest technology markets in the world.
SRE vs. DevOps vs. Platform Engineering
The article covers a broad level comparison between SRE, Platform Engineering and DevOps. One can say that SRE and platform engineering teams are successors of the typical and traditional operation teams.
Why does this matter? Platform engineering is an emerging role that monitors the entire lifecycle of a software development process, right from the source to production.
💼 Jobs
⭐️Claypot — Founding Engineer (Infra)
⭐️ Grit — Design Engineer
⭐️ Omni — Senior Data Visualization Engineer
