Tackle complex problems by setting goals instead of step by step instructions
Giving AI models broad goals and letting them write code or build internal tools yields better outcomes for complex challenges than giving rigid step by step prompts. This approach was recently used by researchers to decipher historical encrypted messages.
Try it yourself
- 1Define the overall objective clearly without offering step by step guidance.
- 2Instruct the model to perform initial background research on the topic.
- 3Allow the model to build custom scripts or tools as needed to solve the problem.
- 4Provide a known fact to narrow down the investigation space.
- 5Ask for a clear method to check the answer against verified data.
Copy this prompt
Goal: [insert target outcome here, such as resolving a budget discrepancy]. First, research the background context and share your findings. Next, build any necessary tools like scripts or spreadsheets to reach the objective. Here is one confirmed fact: [insert verified detail]. Conclude by explaining how I can verify your final results against existing reliable data.
The Blend
Artificial intelligence programs recently solved two historic, unbroken encrypted messages from Nazi Germany's Enigma machine. As reported by TechCrunch, independent researchers used advanced AI models from OpenAI and Anthropic to decipher codes that had stumped cryptographers for decades. Rather than following strict, micromanaged steps, one system was simply given a high level directive to find an uncracked message in a digital archive and figure out how to solve it.
To accomplish the task, the software autonomously searched historical archives, deduced context clues, and wrote custom computer code to simulate the original encoding hardware. Longtime cryptology expert Frode Weierud confirmed the breakthrough, telling TechCrunch that the AI accomplished in two days what normally requires months of painstaking human labor. The success highlights a shift toward agentic AI systems that independently plan and build their own tools to achieve broad objectives.
While this milestone shows remarkable problem solving, it raises questions about how AI agents navigate digital permissions and privacy boundaries. Weierud noted uncertainty regarding where the model acquired some of its background archival records. Beyond historical puzzles, if broad goal setting allows software to reverse engineer military grade encryption on its own, future cybersecurity defenses may need to be entirely redesigned to stay ahead of automated threat discovery.
Written independently by AI News Smoothie from the reporting listed below. Facts belong to the original publishers. Follow the links for their full coverage.
Ingredients
- Astra and Opus just passed Turing's other test | TechCrunch
Researchers demonstrated that AI models given broad goals can independently research context and write code to solve previously uncrackable historical encryption.