It's been an absolutely insane past 24 hours of AI developments (OpenAI, Google Gemini, Microsoft, Slack, X, Meta, Langchain, Stable Diffusion) Will go more in-depth on everything you need to know/my thoughts/highlights in tomorrow’s newsletter:
LLMS
-
Efficient Fine-tuning with Adapters for LLMs
By
–
Want to learn about efficient #finetuning techniques, why #adapters are the future of #LLMs, and how to get started? Save your spot for our upcoming ML Real Talk for an informative and interactive conversation on adapter-based fine-tuning. https://
pbase.ai/3I2x1Y0 -
Smaug-72B Tops Hugging Face Leaderboard as Best Open-Source Model
By
–
Two weeks ago we released Smaug-72B – which topped the Hugging Face LLM leaderboard and it’s the first model with an average score of 80, making it the world’s best open-source foundation model. We applied several techniques on a fine tune derived from a Qwen-72B for this model.
-
Large Context Windows Improve AI Writing Style Mimicry
By
–
Good point. The massive context window allows you to provide a ton of more context, which should help the model mimic writing style better
-
Google announces Gemini 1.5, a 1 million token context multimodal model
By
–
Google just dropped a 1 million token context multimodal model, Gemini 1.5 pic.twitter.com/YITwQK7brs
— AI Breakfast (@AiBreakfast) 15 février 2024Google just dropped a 1 million token context multimodal model, Gemini 1.5
-
LangSmith GA Launch Sequoia Series A Funding New Brand
By
–
Today, we’re thrilled to announce – The general availability of LangSmith (no more waitlist!)
– Our Series A fundraise led by @sequoia – Our beautiful new homepage and brand We've worked hard over the past few months to add requested features and ensure LangSmith can -
Gemini 1.5 Pro Outperforms Ultra in Latest Benchmarks
By
–
Here's how Gemini 1.5 Pro compares to Gemini Pro 1.0 and Gemini Ultra 1.0. These are huge improvements. It's shocking to see this new model beat out Gemini Ultra (the paid version that was launched just last week) in so many benchmarks.
-
Gemini Pro Processes 700K Words and Code in Seconds
By
–
Here's the model in action in Google AI Studio.
— Rowan Cheung (@rowancheung) 15 février 2024
You can upload multiple pdf files that can tally up to ~700,000 words, and Gemini Pro will process the information and give you an output in seconds.
This also works for uploading entire codebases, 1 hour of video, 11 hours of… pic.twitter.com/VPn1Jd8BR9Here's the model in action in Google AI Studio. You can upload multiple pdf files that can tally up to ~700,000 words, and Gemini Pro will process the information and give you an output in seconds. This also works for uploading entire codebases, 1 hour of video, 11 hours of
-
Million Token Context Windows Transform RAG Deployments
By
–
You can shove in entire codebases! 1M context windows are game-changing especially for RAG-based deployments. You can just put a ton more data into the context.
-
AI Model Demonstrates Long-Context Understanding with Apollo 11
By
–
Here’s a fun demo of long-context understanding. First, we asked the model to find 3 amusing moments in the 402-page pdf transcription of the iconic Apollo 11 mission. Then we uploaded a simple drawing of a boot and it identified the moment we had in mind: Neil’s one small step! pic.twitter.com/mx8id3cqSi
— Demis Hassabis (@demishassabis) 15 février 2024Here’s a fun demo of long-context understanding. First, we asked the model to find 3 amusing moments in the 402-page pdf transcription of the iconic Apollo 11 mission. Then we uploaded a simple drawing of a boot and it identified the moment we had in mind: Neil’s one small step!