AI Weekly Digest #21: A New Wave of Open Source Models

LLaMA2, SDXL, OpenAI classifier and Gen2

AI Weekly Digest #21: A New Wave of Open Source Models

Hello, tech enthusiasts! This is Wassim Jouini and Welcome to my AI newsletter, where I bring you the latest advancements in Artificial Intelligence without the unnecessary hype.

You can find me on LinkedIn, Twitter and Medium! Let’s connect!

Now let's dive into this week's news and explore the practical applications of AI across various sectors.

Main Headlines

Here are the four main trends to keep in mind if you are working in AI today.

Early 2023, Meta’s LLaMA models leaked. This state of the art foundation model was made fully available to everyone! Hence it could be used or fine-tuned in any imaginable way. This unintentional model release led to a significant development in Open Source text generation models (read: Top News Shaping AI in 2023 - Part 1 LLMs).

Last week, Meta teamed up with Microsoft to release LLaMA v2!

Why is it a big deal?

  • This time the release is intentional and fully open source, including for commercial purposes!

  • It’s already available on major platforms such as Azure, HuggingFace and AWS.

  • Its performance seems comparable to ChatGPT on all standard tests without Fine-tuning!

  • We can bet that it will lead to a new wave of powerful and cheap Open Source models!

  • Personal Note: my early stage tests seem to indicate that it can be an alternative to ChatGPT for various NLP tasks including instruction based data-extraction or chatbots!
    This is also a big deal if Privacy is a concern and you want to rely on your in house model!
    On the funny side, also note that safety features in the model makes it… a bit dull sometimes. E.g., LLaMA2 would refuse explaining how to “kill a windows process” because … well you shouldn’t harm anyone or anything…

OpenAI sunsets its “AI text generated” detector. Why? “As of July 20, 2023, the AI classifier is no longer available due to its low rate of accuracy.” 

In April 2023, during an interview I already explained that detecting AI generated text is and will remain a hard problem!

In a nutshell, this a fundamental problem (mathematical speaking) and can be summarized as “If you manipulate the model to introduce a certain detectable bias: how can you detect that probability of (biased) sequence of words, if that sequence is altered?”. Well… with no additional hypothesis, you can’t.

This doesn’t mean that it’s impossible to detect in all scenarios ; it’s actually very reliable if you’re looking at a binary problem where it’s either a human written content or a fully AI written content. Yet, in the common case of an AI assisted text generation, the detector can’t be trusted anymore…

Interview Link: Here’s the link of the interview: check min 30:30.

Stability AI released this week its new Stable Diffusion flagship image generation model: SDXL1.0!

Why is it a big deal?

  • Performance: waaaay better than stable diffusion 1.5 or 2.1 ; it’s now comparable to MidJourney, providing a top notch image with a natural short prompt! You can simply test here for free: https://clipdrop.co/stable-diffusion

  • Model Access: this model remain open source (same as previous models released by Stability AI). This means that you can use it now to fine-tune it, or teach it new concepts via dreambooth or lora frameworks!

  • API Access: This is the best model out there available via APIs! This isn’t the case for Midjourney for instance, making automation with Midjourney difficult.

Examples of Images generated by author with SDXL.

Upload an image and let the AI animate it. This is what Gen2 offers today as one of its multiple Image and text based video generation!

The best way to explain it is probably to let you watch this short video with several examples of images animated by Gen2.

This is it for Today!

Until next time, this is Wassim Jouini, signing off. See you in the next edition!

Have a great Sunday and may AI always be on your side!