Posts

Are you sure you asked the question right?

Image
More of my AI co-creations LLMs don’t have a truth gauge. They say things that sound correct even when they’re completely wrong . 

Project IDX

Image
Paraphrased from the blog article , "How we built Project IDX: A high-level overview": Project IDX is an experimental browser-based development experience that Google has designed to make the end-to-end multi-platform app development workflow faster and more frictionless. IDX was built on Code OSS , the Visual Studio Code product that Microsoft has open-sourced under the standard MIT license. Every IDX workspace comes with a full Debian-based VM and provides full access to the terminal so you can install your favorite tools. VMs automatically hibernate and activate as needed. IDX workspaces are built on top of Google's Cloud Workstations offering which enable provisioning and scaling of VMs. System software such as tools for running services, scheduling jobs (such as Supervisor), and other custom written jobs such as preview server are packaged within custom Docker containers. With IDX, you can get going in a Flutter environment with just one click instead of wasting hou...

This Week I Learned - Week #8 2024

Image
This Week I Learned -  * Intuit's own Gen AI operating system orchestrates the large language model (LLM) activity across the entire company. GenOS is a full-blown operating system to support the responsible development of GenAI-powered features across its technology platform. *  Good data creates good models.  Evaluating models based on what they contain and what answers they provide is critical.  Anytime we look to implement GenAI features, we have to balance the size of the model with the costs of deploying and querying it. The criteria for an LLM in production revolve around cost, speed, and accuracy. Response times decrease roughly in line with a model’s size (measured by number of parameters). *  Model drift—where an LLM becomes less accurate over time as concepts shift in the real world—will affect the accuracy of results. The sweet spot for updates is doing it in a way that won’t cost too much and limit duplication of efforts from one vers...