Exciting to see Greylock announce their new fund dedicated to early stage investing. We are privileged to have them as partners since day one, and grateful for their support on Snorkel's journey!
@snorkelai
-
Foundation AI Models Training: Labeling and LLM Development Insights
By
–
Catch up on the latest podcasts this weekend featuring conversations with @snorkelai CEO @ajratner
: @thecloudcastnet: Training Foundation AI Models, where Alex discusses the intricacies of labeling & training of LLMs, the power of foundational models, and the art of -
Simple Prompting Techniques Achieve 49 F1 Score Baseline
By
–
2/ We started with simple prompting — tell the model what you're looking for and how to format its responses. This catches a lot of the low hanging fruit (~49 F1), but still needs refinement.
-
Foundation Models: Data-Centric Techniques for Business Quality
By
–
5/ Foundation models are great, but they're just a foundation. For business-critical applications where quality matters, use the full slate of data-centric model improvement techniques!
-
Data-Centric Prompt Engineering: Iterative Model Error Refinement
By
–
3/ We then began data-centric prompt engineering, iteratively identifying model error modes and addressing them in the prompt where possible, either with added detail in the prompt or adding in-context examples (~68 F1).
-
DistilBERT Fine-Tuning Achieves 84 F1 With Programmatic Labeling
By
–
4/ For maximum quality, we finished with fine-tuning, using Snorkel Flow to programmatically label data in areas where the model was still underperforming. Our best model (~84 F1) is fine-tuned DistilBERT, a model 10,000x smaller than GPT-4.
-
Fine-tuning and Prompt Engineering Outperform GPT-4 Baseline
By
–
1/ Prompt engineering LLMs is valuable, but used alone leaves many points on the table. See how we combine prompt engineering and fine-tuning to create a specialized model that outperforms prompted GPT-4 by 15 to 34 points on a real-world extraction task! https://
snkl.ai/pro -
AI Founders and Innovators Meetup in Redwood City
By
–
We're excited to see everyone at our Redwood City, CA office this afternoon for the AI Founders and Innovators Meetup in partnership with @msft4startups
! We have a few spots left if you'd like to join, RSVP: https://
partiful.com/e/vLaaFcXDoUFO
CY6vCpWr
… -
Weak Supervision for Open-Source LLM Instruction Tuning Data Curation
By
–
We used weak supervision to programmatically curate instruction tuning data for open-source LLMs like Llama 2 and RedPajama, enabling more granular error analysis and higher quality—without an army of manual annotators. Links to data and models on the blog!
-
SF Bay Area Startup Journey Panel and Roof Party July 26
By
–
If you’re in the SF Bay Area, we hope you can join our meetup and roof party on July 26 at our office in Redwood City featuring a panel on the startup journey with @ajratner @jerryjliu0 @lqiao and @jungofthewon
. It's free, but space is limited. RSVP