Very proud to see the many ways that @GoogleResearch contributed to the many announcements at the recent Google IO event. PaLM 2 in dozens of uses, Imagen, Phenaki, Chirp, MusicLM, flood forecasting, Green Light, fair and inclusive ML work, and more!
MACHINE LEARNING
-
MIT ML Model Identifies Materials in Images for Robot Interaction
By
–
MIT researchers develop ML model that can identify all the pixels & highlight similar materials in an image, potentially helping robots interact with objects in the real world: http://
bit.ly/3ODmYNp -
4-bit Llama Fine-tuning with Alpaca Dataset Performance Analysis
By
–
I trained a 4bit on a Llama base with alpaca dataset today. Reasoning was worse compared to traditional alpaca.
-
4bit Training Reduces AI Reasoning Accuracy
By
–
4bit training leads to very inaccurate reasoning skills. Not sure why.
-
India Advances Fine-tuning and Machine Learning Capabilities
By
–
@GoI_MeitY @NeGD_GoI @nasscom @_DigitalIndia @jeevprabnivash @MIT @StevePolyak #finetuning #ArtificialIntelligence #MachineLearning #INDIAai
-
Improving AI accuracy in mimicking human judgment through data optimization
By
–
AI models meet human judgment. A study has predicted that increasing dataset visibility & fine-tuning normative data can positively impact the accuracy of an AI model in mimicking human judgements. Find out how AI models can be trained for this: https://
rb.gy/gxozg -
LLMs Transform NLP Landscape with Generative AI Capabilities
By
–
1/5 The recent evolution of large language models (LLMs) & generative AI is transforming the landscape of natural language processing (NLP). LLMs can handle a range of NLP use cases, making them an indispensable tool for tasks like generating original and coherent text.
-
Closing the capabilities gap between open and closed language models
By
–
Build better base models. “there exists a substantial capabilities gap between open and closed LMs that, with current methods, can only be bridged using an unwieldy amount of imitation data or by using more capable base LMs.”
-
Complex Tasks Cause Knowledge Forgetting During Execution
By
–
I think the interesting observation here is that in the process of completing a larger, more complex task, it seems to "forget" this knowledge
-
Word Length Memorization Works for Common Words
By
–
yeah I understand that and you're right it might fail on edge cases, but for most common words (you can test all the words it produced in the output I posted to see for yourself) it has indeed "memorized" the right answer for the length of the word