At small scale, at compute match, with enough supervised data points, a flan-t5 baseline is very tough to beat. Of course t5 has its problems (outdated etc) but a modern one would be very OP.
LLMS
-
Google’s Ping-Pong Robot and Latest AI Breakthroughs
By
–
Top stories in AI today: -Google’s ping-pong-playing robot
-GPT-4 tackles top-secret tasks
-Create custom selfies with Meta AI
-AI speeds up schizophrenia cure
-6 new AI tools & 4 new AI jobs Read more: http://
therundown.ai/p/google-deepm
ind-ping-pong-robot
… -
UX suggestions for AI chat thread management and text copying
By
–
Some from my side – Collections require a default focus mode setting to be selected for all new threads (ex Writing).
– A copy button should have a way to copy only text without citations I understand the purpose of having them included but more than sure most of users are -
Discussion on LLM model versions in search queries
By
–
Ah true, it is a fine-tuned 3.5 if not outdated I saw 4o-mini used on non pro search queries for the pro account one day
-
Transformer Explainer: Interactive Visualization Tool for Text-Generative Models
By
–
Transformer Explainer
— AK (@_akhaliq) 9 août 2024
Interactive Learning of Text-Generative Models
discuss: https://t.co/YPvsiqgEl2
Transformers have revolutionized machine learning, yet their inner workings remain opaque to many. We present Transformer Explainer, an interactive visualization tool designed… pic.twitter.com/nu0Fs4QI2KTransformer Explainer Interactive Learning of Text-Generative Models discuss: https://
huggingface.co/papers/2408.04
619
… Transformers have revolutionized machine learning, yet their inner workings remain opaque to many. We present Transformer Explainer, an interactive visualization tool designed -
Img-Diff: Contrastive Data Synthesis for Multimodal LLMs
By
–
Img-Diff Contrastive Data Synthesis for Multimodal Large Language Models discuss: https://
huggingface.co/papers/2408.04
594
… High-performance Multimodal Large Language Models (MLLMs) rely heavily on data quality. This study introduces a novel dataset named Img-Diff, designed to enhance -
LLM-DetectAIve: Tool for Machine-Generated Text Detection
By
–
LLM-DetectAIve a Tool for Fine-Grained Machine-Generated Text Detection discuss: https://
huggingface.co/papers/2408.04
284
… The widespread accessibility of large language models (LLMs) to the general public has significantly amplified the dissemination of machine-generated texts (MGTs). -
Instruction Back-and-Forth Translation for Better LLM Alignment
By
–
Better Alignment with Instruction Back-and-Forth Translation discuss: https://
huggingface.co/papers/2408.04
614
… We propose a new method, instruction back-and-forth translation, to construct high-quality synthetic data grounded in world knowledge for aligning large language models (LLMs). Given -
Noam Shazeer Joins Google DeepMind for Gemini AGI Research
By
–
Thrilled to welcome @NoamShazeer to @GoogleDeepMind (& back to @Google
)! Noam co-created the transformers architecture that underpins the modern AI industry, and is one of the most inventive researchers I know. Really excited to be working with you on Gemini and AGI! -
Exponential compute demands force AI paradigm shift
By
–
At some point the math doesn't work… unless we flip the compute paradigm.
— Cerebras (@cerebras) 8 août 2024
GPT-4 requires 40,000x more compute than models from a few years ago. This exponentially increasing need for compute is the fundamental problem that AI is facing. Multiple football fields of silicon to… https://t.co/vDBA6PvztPAt some point the math doesn't work… unless we flip the compute paradigm. GPT-4 requires 40,000x more compute than models from a few years ago. This exponentially increasing need for compute is the fundamental problem that AI is facing. Multiple football fields of silicon to