Daily Digest: Google sneaks to the top

PLUS: GitHub's move to woo AI engineers

August 02, 2024

Want to get in front of 100k AI enthusiasts? Work with us here

Hello folks, next week we have two amazing workshops lined up.

Here’s what we have today;

PICKS

  1. New tutorials:

  2. Google climbs to the top of AI leaderboards. A new experimental version of Gemini 1.5 Pro now ranks #1 on the LMsys leaderboard with a huge lead over GPT-4o. We don’t have scores on traditional benchmarks yet, but the initial impressions from Twitterverse are positive. It’s available on Google’s AI studio and Gemini API to test.

  3. GitHub is adding an AI playground to its platform. GitHub Models comes with open models like Llama and Mistral and closed ones like GPT-4o and 4o mini from OpenAI. The limited public beta is open for signups now.🍿Our Summary (also below)

  4. Enhance RAG response quality with better data preparation. Unstructured data is notoriously hard to wrangle, but AI systems are hungry for it. Parse, de-identify, chunk, and embed clean, normalized data from any file format for your vector store with Tonic Textual for better quality RAG systems. (sponsor)

TOP TOOLS

  • Wordware (YC S24) gives non-technical domain experts an IDE + toolkit to build AI apps and workflows with natural language. Try it out with promo code LAUNCH (I’m an investor).

  • Toby - Live speech translation on any video call.

  • Martin - Your AI butler with calendar, email, search and more.

  • TorchChat - Run PyTorch LLMs locally on servers, desktop and mobile.

  • Trudo AI - Python-based workflow builder with AI copilot.

  • Eduwiz - Write magical paperwork in a few seconds for free.

  • LangGraph Studio by LangChain - Visualize, interact with, and debug complex agent-based applications.

  • Workflows by LlamaIndex - Event-driven approach to building multi-agent applications.

  • Clarity - Use AI to supercharge founder-led sales.

NEWS
QUICK BITES

GitHub's giving its 100 million+ users a shiny new AI toy to play with. It’s adding an AI playground to its platform. Think of it as your personal AI sandbox, right where you already store your code.

What's going on here?

GitHub's rolling out "GitHub Models," letting developers test-drive various AI models directly on their platform.

What does this mean?

At the core of GitHub Models is a nifty playground to test prompts and fiddle with parameters, all for free.

If you dig what you see, you can seamlessly move to coding with these models in Codespaces or VS Code. And if you're ready for the big leagues, there's a clear path to deploy via Azure too. GitHub promises your prompts and outputs won't be shared or used to train the models.

You can tinker with big-name models like GPT-4o, Llama 3.1, and Mistral Large 2 right on GitHub in the limited beta. More language, vision, and other models are in the pipeline as they gear up for general availability.

Why should I care?

GitHub wants to keep "AI engineers" (software engineers using AI APIs) on their platform. Many are shifting to tools like Replit and Cursor, but GitHub wants to offer native AI access to make their Copilot offering more appealing.

Testing multiple models in one place is a big ask from AI engineers, and entire startups are being built to tackle this. GitHub's current lineup doesn't quite cut it for me though—I'd love to see top models from Anthropic, Google, and OpenAI side-by-side. The current option feels like single-party elections 🫣.

Ben’s Bites Insights

We have 2 databases that are updated daily which you can access by sharing Ben’s Bites using the link below;

  • All 10k+ links we’ve covered, easily filterable (1 referral)

  • 6k+ AI company funding rounds from Jan 2022, including investors, amounts, stage etc (3 referrals)

Reply

or to participate.