Google has 25 years of coding history to train AI

Google's internal software systems have been capturing telemetry on literally everything their engineers do, for the last 25+ years...Which, stitched together, makes an awesome training dataset for a software engineering assistant. 😁

DIDACT (Dynamic Integrated Developer ACTivity) is a methodology for training machine learning models on the entire software development process, not just the end product. This approach allows the model to better understand and align with developers' workflows. Google's vast software development logs fuel DIDACT, providing a rich training dataset. Key features include:

• Multi-task model covering activities like editing and debugging.

• Development of three tools: Comment Resolution, Build Repair, and Tip Prediction, which received positive feedback from Google developers.

• Use of a mini programming language, "DevScript," for expressing complex actions.

• Capabilities include code clean-up, history-augmented code completion, and edit prediction.

In essence, DIDACT turns Google's software development activities into training demos for ML developer assistants, enhancing productivity and work quality for software engineers.

💥 Master ChatGPT & AI Hacks for FREE in Just 3 Hours! 💥

Join the 3-hour ChatGPT & AI Workshop for FREE (worth $49) by Growthschool to become the top 1% of AI users 🤯

Highly Rated at 9.5/10 with Over 400,000 Satisfied Students 🌟

👉 Claim Your FREE Spot (Limited to the First 100 Participants) 🎁

You should attend if you are a:

• ✅ Working Professional Seeking AI Proficiency

• ✅ Entrepreneur or Freelancer Eager to do more with your time

• ✅ Creative Mind, Designer, Coder, or Student

⏰ Hurry! Registrations Close in 24 Hours ⏰

👉Join the Workshop NOW (FREE for the First 100 Participants) 🚀

Sam Altman: If AGI goes wrong, oh boy, hiding in a bunker wouldn't spare anyone's life

Joanna Stern: What are those specific risk mitigations you're putting in?

Mira Murati (OpenAI CTO): #1 is rolling out the technology.

The conversation discussed the advancements and capabilities of AI models like GPT-4. Topics included the human-like nature of AI, its evolution, societal impacts, and the responsibility of handling such technologies. Data usage, especially from publishers, was emphasized, underscoring the need for data that respects user comfort. There was a consensus that while AI can be personalized, it shouldn't replace genuine human interaction. The potential of new AI-optimized computing platforms was mentioned, with NVIDIA recognized as a current market leader. The discussion also touched on AI and AGI's significant societal impacts, highlighting potential workforce disruptions and the fast pace of technological change. There's an emphasis on making AI accessible, anticipating job shifts, and ensuring humans remain engaged in purposeful work. Using tools like ChatGPT can help involve more people in the dialogue about AI's societal role.

Midjourney's new website has been launched in beta!

Other stuff

• How to blend images in Dalle3 (The link to this post was missing in yesterday’s issue)

• SF’s top AI hackers all came to a mansion to push the creative limits of art and music. These projects are next-level

• AI will never threaten humans, says top Meta scientist

• Tutorial: AI-Powered TikTok Videos: I Made a Bot Create Content for a Month!

• Impressive demo of Stable Diffusion and ControlNet

• This new data poisoning tool lets artists fight back against generative AI

• From doom to boom: AI is slowly re-energizing San Francisco

• Riley Reid talks about creating her virtual self with Clona.ai, and how compassion and consent play a role in AI's future.

Prompt of the Day 🍭: Apple Name Creation

Click on the image to load the full prompt in ChatGPT (requires Superpower ChatGPT)

Hope you enjoyed today's newsletter

Follow me on Twitter and Linkedin for more AI news and resources.

⚡️ Join over 200,000 people using the Superpower ChatGPT extension on Chrome and Firefox.