Hugging Face
Search models, datasets, users...
Ctrl K
Models
Datasets
Spaces
Buckets
Docs
Enterprise
Pricing
Website
Tasks
HuggingChat
Collections
Languages
Organizations
Community
Blog
Posts
Daily Papers
Hardware
Learn
Discord
Forum
GitHub
Solutions
Team & Enterprise
Hugging Face PRO
Enterprise Support
Inference Providers
Inference Endpoints
Storage Buckets
Log In
Sign Up
Brian Crouch
Briancrouch
1
1
34
Follow
0 followers
ยท
23 following
https://www.avalara.com/us/en/learn.html
Briancrouch
BrianAva
AI & ML interests
Generative coding for API connectors at scale using permission-based integration to complete codebase
Recent Activity
reacted
to
dejanai
's
post
with ๐
3 days ago
Unlocking hidden skills inside Google's model https://dejan.ai/blog/latent-skills/ AI models hold skills that no prompt can unlock. A NeurIPS 2025 study trained small numbers of randomly chosen weights in Llama and Qwen models of four sizes, across ten tasks: 10 to 100 trained weights closed up to half the gap between the original model and a fully fine-tuned one. We repeated the test on Google's smallest Gemma model, Gemma 3 270M. The best of 20 prompt wordings got it to sort five numbers correctly 1% of the time. After we trained 1,000 of its 100 million weights (0.001%), it sorted correctly 97% of the time. The same method switched on reversing word lists (84%) and clean JSON output (99% with 30 weights), and lifted long addition, while multiplication stayed near 30% even with 100,000 trained weights.
reacted
to
dejanai
's
post
with ๐ฅ
3 days ago
Unlocking hidden skills inside Google's model https://dejan.ai/blog/latent-skills/ AI models hold skills that no prompt can unlock. A NeurIPS 2025 study trained small numbers of randomly chosen weights in Llama and Qwen models of four sizes, across ten tasks: 10 to 100 trained weights closed up to half the gap between the original model and a fully fine-tuned one. We repeated the test on Google's smallest Gemma model, Gemma 3 270M. The best of 20 prompt wordings got it to sort five numbers correctly 1% of the time. After we trained 1,000 of its 100 million weights (0.001%), it sorted correctly 97% of the time. The same method switched on reversing word lists (84%) and clean JSON output (99% with 30 weights), and lifted long addition, while multiplication stayed near 30% even with 100,000 trained weights.
liked
a model
17 days ago
dejanai/LinkBERT
View all activity
Organizations
buckets
2
Sort:ย Recently updated
Briancrouch/fanout-diffusion-bucket
3 GB
Briancrouch/WikiProfile-bucket
7.51 MB
spaces
3
Sort:ย Recently updated
Running
OrchestrateAI Demo Platform โจ
๐ณ
Automate complex workflows with AI
Running
DeepSite Project
๐ณ
Create a static web page by editing HTML
Running
Ai Marker Explorer Decoding Gpt 5 S Secrets
๐
models
1
Briancrouch/DEJAN-LM
11.6M
โข
Updated
Feb 12
โข
4
datasets
0
None public yet