Amazon unveils more AI offerings

Plus an AI agent that can solve problems as fast as humans

Author

Nathaniel Whittemore
July 27, 2023

The AI Breakdown First Five - Thursday July 27, 2023

Today on the First Five:

  • 5. AdA Agent Can Solve Problems as Fast As Humans

  • 4. McKinsey Says Women Could Be Hit Harder by AI Job Loss

  • 3. Microsoft AI to Power Japanese Government

  • 2. New Paper Shows How to Attack LLMs

  • 1. Amazon’s Major AI Announcements

5. AdA Agent Can Solve Problems as Fast As Humans

Google DeepMind has just shared more information about AdA, an AI agent that they say is as adept at solving problems as humans. In a 3D virtual world it was able to test hypothesis, learn from demonstrations, recall previous learnings and quickly adapt to new challenges.

4. McKinsey Says Women Could Be Hit Harder by AI Job Loss

Consulting giant McKinsey has just released “Generative AI and the Future of Work in America.” In the report, they find that activities that currently account for 30% of work hours could be automated by 2030, with 12 million “occupational transitions” being required. Low-wage workers they say are 14x more likely to need to change jobs and women are 1.5x more likely to need to transition than men.

3. Microsoft AI to Power Japanese Government

With every country reacting to the rise of AI slightly differently, the Japanese government seems bent on harnessing its power. The Nikkei newspaper is reporting that the country’s government has signed a deal to use Microsoft and OpenAI technology on a trial bases for tasks such as analyzing government statistics. The Ministry of Economy, Trade and Industry (METI) is also working on a homegrown supercomputer.

2. New Paper Shows How to Attack LLMs

New research is calling into question just how safe LLMs can actually be made. A group of researchers claim to have developed an automated system to construct adversarial attacks on LLMs that lead to them obeying user commands even when those commands produce harmful content. The researchers argue that this behavior might never able to be fully prevented by LLM creators.

Researchers Poke Holes in Safety Controls of ChatGPT and Other Chatbots

A new report indicates that the guardrails for widely used chatbots can be thwarted, leading to an increasingly unpredictable environment for the technology.

www.nytimes.com/2023/07/27/technology/ai-chatgpt-safety-research.html

1. Amazon’s Big AI Announcements

Yesterday Amazon Web Services revealed a number of new efforts in the AI space. HealthScribe is a summarization tool for the medical field. Amazon’s data tool QuickSight is also getting an AI upgrade. Finally, Amazon added a number of models including ones from Cohere and the latest from Stability AI and Anthropic to its Bedrock development suite. Still, when it comes to community reaction on Twitter? Crickets.

BONUS: An epic sci-fi trailer, created entirely with AI

The Latest AI Breakdown Episodes

Thanks for reading! How did this edition hit you?

How was today's newsletter?

Login or Subscribe to participate in polls.

Signing off from the future’s past - NLW