Google is using Claude to improve Gemini

Welcome! Here’s what you need to know about AI today:

👉 OpenAI is trying to return to humanoid robot market

👉 AI systems are learning from their own self-generated data 

👉 Google is using competitor’s chatbot to improve Gemini

📧 Did someone forward you this email? Subscribe here for free to get latest AI news these everyday!

Read time: 4.8 minutes

What’s going on: To improve Gemini, Google is comparing its answers to the answers given by Claude, which was created by a company called Anthropic. Google employees are then tasked with reviewing both sets of answers and determining which one is better. Along the way they have found out the Claude is much stricter than any other AI in terms of safety.

What does it mean: This process of comparing AI models to others is common in the tech industry. It helps companies improve their AI and ensure that it is helpful and safe. For example, AI should not provide harmful or misleading information. It is crucial for AI to be safe and reliable. Usually, companies evaluate their AI models against competitors, typically by running through industry benchmarks rather than having contractors evaluate their competitors’ AI responses.

More details: 

• Potential Rule Violation: Some people are concerned that comparing Gemini to Claude might break the rules set by Anthropic.

• Google's Position: Google insists that they are not using Claude to train Gemini. They are only comparing the answers given by Gemini and Claude.

• Concerns about Employee Accuracy: There are worries about whether Google employees can accurately judge the AI's answers on all topics.

What’s going on: There is not anything in this world that AI companies love more than data. Yes, you read that right. They literally can’t get enough of it. They crave it so much that all the data on the internet is not enough for them and they are increasingly using their own AI systems to generate vast amounts of training data. This involves employing existing AI models to produce new data points, such as text, images, and even entire datasets.

Talking about real-world examples, Anthropic's Claude 3.5 Sonnet was partially trained on synthetic data, Meta's Llama 3.1 was fine-tuned using AI-generated data, and upcoming OpenAI's Orion will utilize synthetic training data sourced from OpenAI's "reasoning" model, o1.

What does it mean: Traditionally, AI models are trained on massive amounts of unbiased real-world data, which can be expensive, time-consuming, and difficult to obtain. Synthetic data offers a potential solution to these challenges. By using AI to generate data, companies can create large and diverse datasets more quickly and efficiently. This could lead to faster development of more powerful AI models.

More details: Major challenges of using synthetic data are as follows:

• Bias: Synthetic data can inherit and amplify biases from the original data used to generate it.

• Hallucinations & Errors: AI models used to create synthetic data can produce inaccurate or misleading information, leading to errors in the training process.

• Model Collapse: Over-reliance on synthetic data can cause AI models to lose diversity and accuracy, becoming less effective.

🤖 OpenAI is reportedly considering developing a humanoid robot, having previously disbanded its robotics division in 2021, and is currently investing in robotics startups like Figure and 1X to re-enter the field.

🎙 In a recent interview, Sam Altman, the CEO of OpenAI, shared his belief that superintelligence might not arrive suddenly. Instead, it could gradually emerge. He suggests that if the rate of innovation speeds up 10x and continues to accelerate each year, the cumulative effect would be similar to the arrival of superintelligence, even without a sudden leap in AI itself.

🏎 F1 teams are reportedly using Generative AI and are running in-depth simulations of certain possibilities that might take place during races, to improve performance on the track and win in the competitions.

📝 Letterfly - Upload your resume and generate high-quality cover letters quickly with AI

Pricing: $12.75 for generating 30 cover letters

📽 movavi - AI-assisted video editing software and smart photo editing app designed for Mac OS.

Pricing: Video Suite subscription for $29.95/month

📷 Palette - Colorize your black-and-white photos

Pricing: 1 free credit - unlimited color preview - $9 for 10 photos

Budget lunch recipes for students

🍝 AI systems can be very handy in many situations, especially when you want to quickly come up with delicious recipes. Here is a prompt to generate a couple of budget-friendly recipes alongside step-by-step instructions of how to cook them.

You are a seasoned chef and recipe designer. Your task is to create a variety of lunch options that are quick to prepare, nutritious, and most importantly, cheap to make for university and college students. These lunch recipes should be easy to make, using common ingredients found in most kitchens. Include detailed step-by-step instructions for each recipe, along with tips for variations and substitutions. Please include ingredients that are commonly available in [YOUR_LOCATION].

Metropolis - Revenue Forecasting Analyst

OpenAI - Strategic Finance, Compute

AMD - Senior Applied Data & AI Engineer

Google - Machine Learning Engineer, Large Language Models, Personal AI

Thank you for staying with us like always! If you are not subscribed, subscribe here for free to get more of these emails in your inbox! Cheers!