Apple's Vision of the Future 👁️

Today’s top AI Highlights:

1. A New Way to Discover Places in Maps Using Generative AI

2. SliceGPT: Compress Large Language Models by Deleting Rows and Columns

3. Learn How to Make Professional Quality Videos Within ChatGPT

4. Coolest Apps on Apple Vision Pro

& so much more!

Read time: 3 mins

Latest Developments 🌍

Generate Images Within Bard 🪄

In the latest update from Google, a suite of new and improved generative AI tools have been introduced, aimed at enhancing creative expression across images, music, and text along with updates to Bard and Google Maps.

1. ImageFX: ImageFX, powered by Imagen 2 which was updated in December, is a new image-generation tool that allows you to create images from simple text prompts, featuring an interface with "expressive chips" for exploring adjacent concepts and ideas rapidly. All the outputs will be watermarked with Google DeepMind’s SynthID.

1. MusicFX: You can now create tunes up to 70 seconds long on MusicFX, with new capabilities for higher-quality audio and faster music generation, based on improvements to the MusicLM model. Over 10 million tracks have been created since its launch.

1. TextFX: Developed in partnership with Lupe Fiasco, TextFX is a generative AI tool for creative writing, recently updated for improved usability and navigation, encouraging exploration in text and language creativity.

2. Bard Updates with Gemini Pro Global Access: Bard's capabilities have been expanded to include Gemini Pro in over 40 languages and 230 countries, along with ImageFX in most countries. The double-check feature for corroborating Bard’s responses with web content, is also available in more than 40 languages.

3. Generative AI in Maps for Place Discovery: Google Maps will now give a new experience of discovering places. Just tell your requirements, no matter how broad or specific, and it’ll analyze Maps’ rich information about businesses and places to give suggestions. You can also ask follow-up questions on the same chat.

Reduce Upto 25% Model Parameters Without Performance Decline 🤩

Facing the challenge of deploying LLMs due to their substantial computational and memory requirements, the AI research community has been in search of efficient solutions to mitigate these constraints. Enter SliceGPT, a newly developed post-training sparsification scheme that offers a significant reduction in the resources needed to run these complex models. By enabling the removal of up to 25% of a model's parameters while retaining as much as 99% of its zero-shot task performance, SliceGPT presents a viable approach to making the deployment of LLMs both more accessible and efficient.

Key Highlights:

1. SliceGPT achieves remarkable computational savings by reducing the total compute for inference on models like Llama 2-70B to just 64% on 24GB consumer GPUs and 66% on 40GB A100 GPUs. This efficiency gain is achieved without necessitating additional code optimization, thereby streamlining the process of deploying large-scale AI models across a broader range of hardware.

2. Despite the significant reduction in model size, SliceGPT impressively maintains high levels of performance. For instance, after compressing the Llama 2-70B, OPT 66B, and Phi-2 models by 25%, the scheme preserves 99%, 99%, and 90% of the original models' zero-shot task performance, respectively.

3. Unlike traditional sparsification techniques that often require complex recovery fine-tuning or introduce additional computational overhead, SliceGPT simplifies the process by removing entire rows or columns from weight matrices. This approach is grounded in the principle of computational invariance, leveraging orthogonal matrix transformations to minimize the impact on a model's predictive capabilities.

Create Videos Within ChatGPT

Creating videos has never been easier and faster! Something that would take hours earlier can now be done in a few minutes.

VEED’s VideoGPT lets you create full videos within ChatGPT itself and it’s insanely simple. You just have to tell VideoGPT what you want to make and in no time the GPT will make studio-quality editable video with images, music, script, and everything which you had to manually work on! You can even add your own avatar or clone your voice and use it within the video.

Here’s what I did with VideoGPT:

With a few edits on VEED’s website, here is the result:

Apple Vision Pro Special 👓

1. Swap face in real-time with DeepFake Live

1. Learn to play piano with Flowing Tiles

1. Witness the future of immersive memories with Luma AI

1. Watch the F1 races unlike ever before

1. Zillow's futuristic way to tour homes for sale

More such apps in this compilation! Check it out and share it.

Hot Takes 🔥

1. ChatGPT and Bard are both perfect examples of how engineering tweaks can compensate for the limitations of the underlying language model. There's a ton of technical work that needs to be performed to fit an AI to its intended market. This reality is often overlooked. ~ Carlos E. Perez

2. I think chain of thought prompting is still one of the coolest things LLMs can do. It is more impressive the better the model - that is a bigger model trained for longer ~ anton

Meme of the Day 🤡

Apple Vision Pro users right now

That’s all for today!

See you tomorrow with more such AI-filled content. Don’t forget to subscribe and give your feedback below 👇

Real-time AI Updates 🚨

⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!!

PS: I curate this AI newsletter every day for FREE, your support is what keeps me going. If you find value in what you read, share it with your friends by clicking the share button below!