Y lo que ha sacado OpenAI, pues bueno, se le parece demasiado… → Explorar el uso de SAE.
→ Para extraer features de un LLM.
→ Que podrían servir para condicionar el comportamiento del modelo. Ajam
GENERATIVE AI
-
SAE for LLM Feature Extraction and Behavior Conditioning
By
–
-
New AI model release available for exploration
By
–
FYI, not gpt-5 but still something you can play with, check inside
-
Sparse Autoencoders Scale Training Improves GPT-4 Interpretability
By
–
Improved methods for training sparse autoencoders at scale. Shows promise towards interpreting GPT-4's neural activity: https://t.co/zNbYnYr6lE
— Greg Brockman (@gdb) 6 juin 2024Improved methods for training sparse autoencoders at scale. Shows promise towards interpreting GPT-4's neural activity:
-
BabyAGI Maintenance: Finding Active Forks and Alternatives
By
–
fair question… haven't touched the babyagi repo in a long time… started working on a new version this year, but might make it a different repo. any suggestions on where to point ppl to who want to fork an actively supported version of babyagi (or similar project?)
-
Consistent Character Model Available on Replicate Platform
By
–
https://
replicate.com/fofr/consisten
t-character
… is the link! -
Consistent Character Now Available on Replicate
By
–
You can use Consistent Character right now on Replicate btw. Link below
-
Browserbase Added to AI Development Tools Shortlist
By
–
Browserbase is already on my shortlist of tools I plan on using 🙂
-
Code Generation for Agentic Behavior: Terminology and Approaches
By
–
We use code gen to do agentic behavior, what’s a good term for that do you think?
-
Common Challenges Data Professionals Face Today
By
–
I think this happens to all data professionals via @montecarlodata #DataEngineer #DataEngineering #AI #BigData #Analytics #DataScience #MachineLearning #DigitalTrasnformation #ArtificiaIntelligence #ChatGPT #Data #EmergingTech #DeepLearning #Robotics #GenerativeAI #GenAI
-
Token Efficiency Approaches in AI Model Optimization
By
–
I asked it to make it token efficient, but I didn’t provide details beyond that. Though for reasons pointed out by others, I don’t think is the right approach.