Interesting GPT-4V can correctly identify the make and model of a wristwatch but can’t read the time it’s displaying.
LLMS
-
WikiVec2Text: Wikipedia Embedding to Text Model Released
By
–
MF-FOOM/wikivec2text: Simple embedding -> text model trained on a small subset of Wikipedia sentences. https://
bit.ly/3Zl4YuD #AI #MachineLearning #DeepLearning #LLMs #DataScience -
Llava Image Chat Support Added to Llama2.ai
By
–
https://t.co/3ZZemXeQjs pic.twitter.com/bAYkUniXwu
— Replicate (@replicate) 12 octobre 2023I just added Llava support to http://
llama2.ai — you can now chat with your images! It's fully open-source and pretty amazing. -
AI Agents and Foundation Models Shape Future of Artificial Intelligence
By
–
A comprehensive report on state of #ai Happy to see our Voyager work featured in it. The future is not just foundation models like LLMs and diffusion models, but agents interacting with them in systematic ways to solve complex open-ended tasks. @yukez @DrJimFan @guanzhi_wang
-
Four criteria for evaluating new prompting techniques adoption
By
–
For any new prompting technique (e.g., tree-of-thought, least-to-most, graph-of-thoughts prompting), I consider four things to decide if it will become widely adopted: 1. How easy is it to implement
2. How much compute it use
3. How many tasks does it improve
4. How much does it -
GPT-4 Vision AMA: Multimodal AI Capabilities Discussion
By
–
Hell YEAH! I'm in! AmA. #GPT4V #GPT4-Vision
-
Building Transformers from Scratch: Concept and Implementation
By
–
Transformer: Concept and code from scratch https://
bit.ly/45YXobu #AI #MachineLearning #DeepLearning #LLMs #DataScience -
Mitigating Hallucinations and Biases in Large Language Models
By
–
7 tips to Mitigate Hallucinations and Biases in Large Language Models Here is the first of 8 videos from our Training & Fine-Tuning LLMs for Production course with @towards_AI
, @activeloop and the @intel Disruptor initiative! Learn more in the video: https://
youtu.be/CCFkfouJFE8 -
Med42 LLM Passes USMLE with 72% Grade
By
–
(1/2) We are pleased to announce that our strategic partner M42 has developed Med42, an LLM fine-tuned on Llama-70B that passes the USMLE with a 72% grade. Learn more here:
-
Llama-70B Fine-Tuned for Medical AI Outperforms GPT-3.5
By
–
(2/2) We worked with Core42, a G42 company, to preprocess the medical dataset and fine-tune Llama-70B in 1 weekend. No code changes were required, and no infrastructure considerations were needed. This model performs above GPT-3.5 despite being smaller and trained on less data