Meta's Massive AI Leak + Microsoft AI Event

Microsofts AI Event

Microsoft is holding a "Future of Work with AI" event on March 16th, where CEO Satya Nadella and Microsoft 365 lead Jared Spataro will be presenting on how AI will influence the workplace. Rumors suggest that AI-equipped versions of Microsoft 365 apps like Outlook and Word may be introduced, potentially improving search results and suggesting email replies, and making writing suggestions. Microsoft has already launched AI features for Dynamics 365 and Viva Sales, such as an AI "Copilot" that helps reply to customer chats and emails, summarize Teams meetings, and explore marketing data.

Microsoft has been integrating AI into services like Teams for a while and has been investing in OpenAI for years. The event will likely support Microsoft's strategy to establish itself as an AI leader, as other companies like Google are building their own AI models into their products.

Read More

Meta's Massive AI Leak

The LLaMA is free, and it spat on Zucks Face!

Last Week, Meta, the parent company of Facebook, announced its new language model, LLaMA, which is smaller and more efficient than its peers, such as OpenAI's GPT and Google's LaMDA.

Meta made LLaMA open source, allowing researchers to access and test the model for a lower cost than previous language models. However, access to the model was being granted on a case-by-case basis to prevent misuse.

Over the weekend, the entire LLaMA model was leaked by an anon on 4chan with a link to the BitTorrent mirror of LLaMA which eventually made it to GitHub, making it easy, safe, and available for anyone to download.

A Hacker News user was able to run and implement it on their PC and here’s their comment

“I'm running LLaMA-65B on a single A100 80GB with 8bit quantization. $1.5/hr on vast.ai

The output is at least as good as davinci.

I think some early results are using bad repetition penalty and/or temperature settings. I had to set both fairly high to get the best results. (Some people are also incorrectly comparing it to chatGPT/ChatGPT API which is not a good comparison. But that's a different problem.)

I've had it translate, write poems, tell jokes, banter, write executable code. It does it all-- and all on a single card.”

What does this mean? according to this user, Meta’s LLaMA is as good as ChatGPT’s davinci model, it has its quirks with repeating itself or giving wrong information (like ChatGPT does), to fix this the user just changed its temperature.

Meta has not yet disclosed its response to the leak, but it is unclear whether it will embrace its role as the developer of the most widely deployed AI in the world or attempt to control access to the model.

Here’s the Link to LLaMA

Copy/paste the prompt here: https://pastebin.com/XSagDMdC

Copy/paste the prompt here: https://pastebin.com/9WQw2hBv

Copy/paste the prompt here: https://pastebin.com/5rEtCTEj

Dennis Rodman in a wedding dress, full body shot, studio photography, volumetric lighting, realistic, 35mm, expressive, iconic, 4k --ar 2:3