AI solves unsolved math problem

Welcome to Daily Zaps — Here’s what we got for you today: 🙌

DeepMind solves unsolved math problem

DeepMind recently announced that their AI system has potentially solved a complex mathematical challenge known as the "cap set problem." This problem involves arranging points in a specific manner where no three points form a straight line. The significant aspect of this development is that the AI's solution is entirely novel and was not part of its original training data, marking a potential first in AI research where the AI has independently made a unique scientific discovery.

This breakthrough is particularly noteworthy because it deviates from the usual application of AI, where the system operates within the bounds of its training data. In this case, the AI generated a solution that was previously unknown and not included in its training set. The DeepMind team has expressed some uncertainty about the exact mechanisms that led to this success, acknowledging the need for further research to understand and validate the AI-generated solution.

OpenAI gives board veto power over AI safety

OpenAI is enhancing its safety measures to address potential dangers associated with AI development. They've established a new "safety advisory group" that will provide recommendations to the leadership, with the board having veto power over decisions. This change follows a leadership restructure and growing discussions about AI risks.

OpenAI's updated "Preparedness Framework" aims to identify and manage "catastrophic" risks, defined as those causing significant economic damage or harm to individuals. The framework categorizes risks into cybersecurity, persuasion (like disinformation), model autonomy, and CBRN threats. Each model is assessed for these risks, and those with high or critical risks are either not deployed or further developed.

From Our Partners

AI Assistant that’s actually tailored to your business

Businesses are not looking to improve the average chatbot answer. What they really care about is the worst 10% of AI responses that confuse their users.

That's why Quickchat AI builds a platform with comprehensive customization tools, giving your business the power to design and refine your AI Assistant until every response is just perfect.

Customer Support, Sales, actual physical robots—you name it, they’ve got you covered.

Start building your AI Assistant for FREE!

OpenAI suspends ByteDance’s account

OpenAI has suspended ByteDance, the parent company of TikTok, from using its AI tools following reports that ByteDance was secretly using OpenAI's technology to develop a rival chatbot service. This use reportedly violates OpenAI's terms of service, which prohibit using their AI to develop competing models. ByteDance has allegedly been using OpenAI's GPT AI for both training and evaluating its own large-language AI model, "Project Seed."

Internal discussions at ByteDance suggested attempts to conceal their use of OpenAI's technology. ByteDance has launched its AI chatbot, "Duobao," in China, and claims to have replaced OpenAI-generated text in its dataset. The incident has reignited concerns about ByteDance's operations and potential misuse of AI technology.

How to get better answers from ChatGPT

The OpenAI released a prompt guide that offers six strategies for improving the quality of responses from AI models:

Write Clear Instructions: The accuracy of the AI's responses depends heavily on the clarity of the instructions provided. It's important to specify details, format, and the desired level of complexity in your query.

Provide Reference Text: Since AI models can generate incorrect information, especially on obscure topics, providing reference material can lead to more accurate and reliable responses.

Split Complex Tasks into Simpler Subtasks: Breaking down complex requests into smaller, more manageable tasks can reduce errors and improve the quality of the AI's outputs.

Give the Model Time to 'Think': Encouraging the model to reason through a problem before responding can enhance the accuracy of its answers. This can be done by asking the model to explain its thought process.

Use External Tools: To compensate for the AI's limitations, integrating other tools (like code execution engines) can be useful, especially for tasks involving calculations or specialized knowledge.

Test Changes Systematically: Regularly evaluating the AI's performance with a diverse set of examples ensures that improvements in prompts lead to consistently better results.