5/ Zero-Shot Prompting Works Best Unlike other LLMs, deepseek-R1 typically performs better without examples. Instead of showing it a few answers, just use the techniques above for best results.
LLMS
-
Improving LLM Reasoning Through Multi-Option Comparison
By
–
4/ Contrast Multiple Approaches
— God of Prompt (@godofprompt) 7 février 2025
Want the best solution? Have deepseek-R1 compare multiple options.
Example:
“List 10 ways to achieve X. Then determine the most effective approach.”
This increases its reasoning time and improves decision-making. pic.twitter.com/v8fpgGaizk4/ Contrast Multiple Approaches Want the best solution? Have deepseek-R1 compare multiple options. Example:
“List 10 ways to achieve X. Then determine the most effective approach.” This increases its reasoning time and improves decision-making. -
Improving AI Response Accuracy via Self-Reflection Prompting
By
–
3/ Encourage Self-Reflection
— God of Prompt (@godofprompt) 7 février 2025
The model is great at checking its own work—if you ask it to.
Prompt it with:
“Identify potential flaws in your approach and suggest solutions.”
You’ll get a more refined and accurate response. pic.twitter.com/WrHq6DtAA93/ Encourage Self-Reflection The model is great at checking its own work—if you ask it to. Prompt it with:
“Identify potential flaws in your approach and suggest solutions.” You’ll get a more refined and accurate response. -
Optimizing Prompting Strategies for Deepseek-R1 Reasoning
By
–
2/ Encourage Step-wise Reasoning
— God of Prompt (@godofprompt) 7 février 2025
Deepseek-R1 performs significantly better when it works step by step.
Use phrases like:
“Break down the problem step by step and solve it.”
This forces deeper thinking and better results. pic.twitter.com/kzihKoeRYI2/ Encourage Step-wise Reasoning
Deepseek-R1 performs significantly better when it works step by step. Use phrases like:
“Break down the problem step by step and solve it.” This forces deeper thinking and better results. -
Improving DeepSeek-R1 Output Quality via Role Specification
By
–
1/ Role Specification
Always assign a relevant role to deepseek-R1. Example:
“You are a senior software engineer specializing in AI infrastructure.” This sets context and improves output quality. -
Optimizing Prompting Techniques for DeepSeek-R1 Performance
By
–
Many people struggle to get the best responses from deepseek-R1. They assume it works like ChatGPT. But with the right prompting techniques, you can unlock its full potential. Here’s how to improve accuracy and encourage deep thinking:
-
GenAI Models Released Daily Yet CIOs Report Unmet ROI
By
–
New GenAI foundation models are released every two and a half days, yet nearly half of CIOs report that AI hasn't met ROI expectations. #ArtificialIntelligence #AI #RAG #AgenticAI #AIagentInnovation #AIAgents #LLMs #LLM #GenerativeAI #GenAI #technology #TechRevolution #tech
-
LIMO: Less Data Improves Reasoning Performance
By
–
LIMO: Less is More for Reasoning With only 817 examples, it outperforms previous models, proving less data can enhance reasoning. https://
arxiv.org/pdf/2502.03387 https://
github.com/GAIR-NLP/LIMO https://
huggingface.co/datasets/GAIR/
LIMO
… -
Language Priors Steering Thoughts in AI Models
By
–
Cute idea, reminds me of “let’s think step by step” trick. Both lean on the language prior to steer the thoughts.
-
Chains of Thought for o3-mini AI Model
By
–
chains of thought for o3-mini! (we try to organize the raw CoT to make it more readable, and optionally to translate languages, but we try to keep it quite faithful to the raw one) great work from @mia_glaese
, @joannejang
, @akshaynathan_ , and their teams!