Can AI help make important discoveries with practical value? Yes! "Algorithms such as sorting or hashing are used trillions of times each day…AlphaDev discovered sorting algorithms from scratch that outperformed previously known human benchmarks."
RESEARCH
-
Making LLMs Reliable: DSP Framework and MLflow Techniques
By
–
LLMs excel at text generation, but ensuring factual content is tricky. In the next talk, @matei_zaharia @databricks
, explores techniques to transform LLMs into reliable apps, drawing from academic research & industry tools. Stay tuned for insights on the DSP framework & MLflow -
Data-IQ: Systematic Data Stratification Tool for AI Applications
By
–
AI poster competition finalist Nabeel Seedat, @Cambridge_Uni
, will discuss Data-IQ, a powerful tool enabling systematic stratification of data subgroups. Join the talk and vote for the best poster https://
future.snorkel.ai -
Code-First Resources for Learning Deep Learning and Neural Networks
By
–
Thanks. Try code-first courses or pick up a book that emphasizes codes over math. There are many examples: Karpathy neural networks: zero to hero(new!!), one shared above by Sebastian Ras., practical deep learning for coders FAST dot AI, Chollet deep learning with python book.
-
Paprika: Procedure-Aware Pre-Training Model for Instructional Knowledge
By
–
AI poster competition finalist Honglu Zhou, @RutgersU
, will discuss Paprika, a procedure-aware pre-training model for instructional knowledge acquisition. Join the talk and vote for the best poster https://
future.snorkel.ai -
Google DeepMind Announces New AI System for Computing Efficiency
By
–
New from
@bellelin_
: Researchers at Google DeepMind announced on Wednesday a new AI system that could make computing more efficient and sustainable. @DeepMind
@googlecloud
#software #AI #artificialintelligence #CIO -
Meditation Break at Tech Conference: Reconnecting Mind During AI Discussions
By
–
It's time for a #MeditationBreak led by fellow Snorkeler, Karla Arteaga. Hit pause on the AI deep dives for a moment to reconnect with our inner calm. Even at a tech conf, our minds need space to breathe and process. See you back, rejuvenated & ready for more #datacentricAI
-
SQuId: AI Model for Speech Quality Assessment Across Languages
By
–
Learn how SQuId (Speech Quality Identification), a 600M parameter regression model that describes to what extent a piece of speech sounds natural, can be used to complement human ratings for the text-to-speech evaluation of many languages → https://
goo.gle/3J1kghg -
GitHub Copilot: Inside the LLMs Technology Behind It
By
–
Inside GitHub: Working with the LLMs behind GitHub Copilot | The GitHub Blog https://
bit.ly/3oi2PBO #AI #MachineLearning #DeepLearning #LLMs #DataScience -
DataComp-1B CLIP Model Outperforms OpenAI with Lower Compute Cost
By
–
Delve into DataComp with @lschmidt3 next at https://
future.snorkel.ai about innovating #ML by experimenting with new training sets. A new CLIP model, DataComp-1B, outperforms OpenAI’s by 3.7pp on ImageNet, and a 9x improvement in compute cost vs. LAION-5B.