My take: There's a prefixlm architecture and a prefixlm objective. – prefixlm arch is just a non casual decoder. – prefixlm objective is a standard causal lm training but with a non casual mask before a random split point (to form inputs/targets) during pretaining. The reason
RESEARCH
-
Improve LLM Application Accuracy with New Short Course
By
–
Learn a development pattern to systematically improve the accuracy and reliability of LLM applications in our new short course, Improving Accuracy of LLM Applications, built in partnership with @LaminiAI and @Meta, and taught by Lamini’s CEO @realSharonZhou, and Meta’s Senior… pic.twitter.com/b3YogwxPle
— Andrew Ng (@AndrewYNg) 14 août 2024Learn a development pattern to systematically improve the accuracy and reliability of LLM applications in our new short course, Improving Accuracy of LLM Applications, built in partnership with @LaminiAI and @Meta
, and taught by Lamini’s CEO @realSharonZhou
, and Meta’s Senior -
CSAIL Researchers Release AI Safety Research Database
By
–
Preprint: https://
tinyurl.com/t87ze8nk
CSAIL News: https://
tinyurl.com/mrxm2e5y
Database: https://
tinyurl.com/yhtjhb6m
Feedback: https://
tinyurl.com/mue7bmyj Researchers involved include @PeterSlattery1 @aksaeri
, @StephenLCasper
, @mnoetel
, @ProfNeilT
, @RistoUuk
, @soroushjp
, and @jmsdao
. -
AI Risk Taxonomy Framework Seven Domains Subdomains
By
–
Their Domain Taxonomy categorizes AI risks into 7 broad domains and 23 more specific subdomains. For example, "Misinformation" is one of the domains, while "False or misleading information" is one of its subdomains.
-
Risk Attribution in AI Systems: Intentional vs Unintentional
By
–
Four insights from this analysis (1/2): 51% of the risks extracted were attributed to AI systems, while 34% were attributed to humans. Slightly more risks were presented as being unintentional (37%) than intentional (35%). Six times more risks were presented as occurring
-
AI Risk Repository: 700+ Risks from 43 Frameworks
By
–
What are the risks from artificial intelligence? Presenting the first-ever AI Risk Repository: a comprehensive living database of 700+ risks extracted, w/quotes & page numbers, from 43(!) existing frameworks. Read & explore here: https://
tinyurl.com/554k67ej -
Taxonomic Framework for AI Risk Categorization: Entity, Intent, Timing
By
–
To organize their findings, the researchers adapt two existing frameworks into taxonomies for categorizing the identified risks. Their Causal Taxonomy categorizes risks based on three factors:
1. The Entity involved
2. The Intent behind the risk
3. The Timing of its occurrence -
CausalGym Wins Area Chair and Outstanding Paper Awards at ACL
By
–
causalgym won an area chair award and outstanding paper award at ACL 😁 thanks to my very cool advisors @ChrisGPotts and @jurafsky
-
Findings Model in NLP: A Reasonable Approach
By
–
I think this is the Findings model present in NLP. I think it's a pretty reasonable model.