데브허브 | DEVHUB | DeepSeek V3.1: Upgraded to be MORE POWERFUL?
DeepSeek 3.1 Open Source Release: Complete Review & Performance Test
This video shows how DeepSeek version 3.1 has been released as a completely open-source model available on Hugging Face. The new model significantly outperforms other models in reasoning tasks, particularly in front-end coding capabilities.
Model Details
700GB model size
Released under MIT license
Major improvements in mathematics, coding, and front-end design
Available on Hugging Face, Hyperbolic, and other platforms
Can be tested on DeepSeek's official website
Performance Testing Results
This video demonstrates various tests of the model's capabilities:
Reasoning Tests
Successfully generated 10 sentences ending with "apple" ✓
Failed to count words correctly in its own response ✗
Correctly counted the number of 'R's in "strawberry" (with an extra R added) ✓
Failed the trolley problem test due to misunderstanding dead vs. living people ✗
Python Coding Challenges
Passed a difficult challenge on calculating shuffles ✓
Passed an expert-level bitwise logical negation challenge ✓
Passed the expert-level Josie's first permutation challenge ✓
Conclusion
The presenter was impressed with the model overall, especially considering this is a base model without dedicated reasoning capabilities. Adding reasoning would likely make this model even more powerful.
📱 Follow us on social media
🔔 Subscribe for more AI model reviews and tutorials
💬 Share your experience with DeepSeek 3.1 in the comments below!
#DeepSeek #AIModels #OpenSourceAI #MachineLearning #CodingAI #AITesting #TechReview
Timestamp:
0:00 - Introduction to DeepSeek 3.1 release
0:19 - Model details and availability
0:46 - Testing reasoning capabilities
1:32 - Testing word counting abilities
1:55 - Trolley problem reasoning test
2:29 - Python coding challenge: shuffles
2:52 - Python expert challenge: bitwise negation
3:05 - Python expert challenge: permutation
3:17 - Conclusion