Memory Leak — #27
VC Astasia Myers’ perspectives on machine learning, cloud infrastructure, developer tools, open source, and security. Sign up here.
🚀 Products
Speakeasy provides production-quality developer surfaces for your API that delight users and make integration easy. Speakeasy announced they are moving from beta to general availability of their managed native-language SDKs and Terraform providers. Speakeasy manages the entire workflow of SDK and Terraform provider creation: from spec validation/enrichment, through code creation, and package publishing.
Why does this matter? We are honored to announce that Quiet Capital led Speakeasy’s $3.3M preseed round. We are also excited to be joined by GV who led the $7.6M seed. With Speakeasy, the process of SDK development goes from weeks to minutes, saving developers significant time. With Speakeasy’s Terraform Provider Generation, API producers for the first time can easily create and maintain Terraform providers from an API spec directly.
MotionGPT: Human Motion as a Foreign Language
Human motion displays a semantic coupling akin to human language, often perceived as a form of body language. By fusing language data with large-scale motion models, motion-language pre-training that can enhance the performance of motion-related tasks becomes feasible. Driven by this insight, Biao Jiang and team propose MotionGPT, a unified, versatile, and user-friendly motion-language model to handle multiple motion-relevant tasks.
Why does this matter? Extensive experiments demonstrate that MotionGPT achieves state-of-the-art performances on multiple motion tasks including text-driven motion generation, motion captioning, motion prediction, and motion in-between. MotionGPT is a step forward in being able to perform motion synthesis and motion captioning. It underscores that nearly everything will be encoded by an AI model.
Tabnine Chat is an enterprise-grade, code-centric chat application that allows developers to interact with Tabnine’s AI models in a flexible free-form way, using natural language. It is available in beta.
Why does this matter? Tabnine was an early player in the AI-enabled developer assistance space. The category is very hot with Github Copilot, Sourcegraph Cody, among others vying to be the one stop shop for developers to understand code, search code, generate new code, extend code, refactor code, create tests, and documentation generation.
📰 Content
Towards Measuring the Representation of Subjective Global Opinions in Language Models
Anthrophic developed a quantitative framework to understand how similar the responses of models and different participants across the world are in order to help guide the development of inclusive AI that serves all people worldwide. They first built a dataset, GlobalOpinionQA, composed of questions and answers from cross-national surveys designed to capture diverse opinions on global issues. Next, they derived a metric that captures the similarity between LLM responses and people’s responses, conditioned on country. With their framework, they ran three experiments on an LLM trained to be helpful, honest, and harmless with Constitutional AI. When Anthropic translated GlobalOpinionQA questions to a target language, the model’s responses did not necessarily become more similar to the opinions of speakers of those languages.
Why does this matter? The results of Anthropic’s work suggest that model answers are generally similar to average users from the U.S., Canada, and Europe. When a model was asked to respond from the perspective of another country, the answers could be more country-specific. The work underscores the need for models to embody a diversity of cultural viewpoints, not just certain perspectives. This will a consideration for adopters.
Databricks Sees and Raises Snowflake, With Gen AI, LLMops, More
Databricks made a number of AI announcements at its annual Data and AI Summit (DAIS) in San Francisco. This week Snowflake Summit also occurred, and the cloud data warehouse provider showcased their own set of AI functionality.
Why does this matter? With the Databricks announcements of Lakehouse AI and Lakehouse IQ and Snowflake announcing Snowpark ML, it is clearer than ever before that the data stack and ML stack are converging. Additionally, Databricks Marketplace is now GA and will include Lakehouse Apps, which are like Snowflake Native Apps, available on Snowflake Marketplace. These vendors are on a collision course.
U.S. Considers Tightening Restrictions on AI Chip Exports
Officials in the United States are mulling tighter controls on an export regulation created to minimize the availability of artificial intelligence (AI) chips to China. According to sources close to officials, the stricter regulations would include clamping down on the level of computing power in chips that are exported. The sources say an update to the rules could come by late July.
Why does this matter? Washington has ramped up efforts to cut China off from key technologies that can support its military, which includes semiconductors. The concerns rise over the power of the AI technology in the hands of U.S. rivals. Moreover, the dependence on Taiwan to produce the vast majority of the most complex chips has become a major focus for U.S. policymakers. For decades semiconductors have been a key component of national power.
💼 Jobs
⭐️Claypot — Founding Engineer (Infra)
⭐️Chroma — Member of Technical Staff
⭐️Speakeasy — Founding Engineer
