Even we here at Prompt Engineering were taken off guard and playing catch up. Let's get straight into it.
Stanford Introduces Alpaca 7B Model
This week started with Stanford introducing the Alpaca 7B model, a lightweight alternative to GPT-3. The model is trained on 52,000 instructions, making it more accessible and easier to run on local computers. Alpaca 7B performed almost as well as GPT-3 and ChatGPT while requiring significantly less computing power, democratizing the use of large language models.
đź’ˇ
This democratization of large language models means that soon, anyone will be able to run them on their systems, bringing the power of AI to the masses.
The Alpaca Model is built on Meta's (Facebook) recently released LLamA model. This is exactly what we hoped for. Read More Here
Google Announces AI Functionality in Workspace Tools
On Tuesday, Google announced the integration of AI functionalities into their workspace tools, such as Google Docs and Gmail. This means we can expect GPT-3-like functionality in these platforms soon, with autocomplete features, brainstorming capabilities, and even writing full articles or resumes. Google's AI integration will eventually extend to Google Sheets, Slides, and Meet, potentially making many AI-based companies obsolete.
Google also announced the release of the Palm API to select developers, allowing them to build on top of the Palm model. Palm is a multimodal model that combines text prompts with image recognition, offering functionalities that many expected from GPT-4.
Anthropic, a company heavily invested in by Google, introduced Claude, their own version of a chatbot. Currently available only through API, Claude is used behind the scenes in tools like Poe from Cora and AI in Notion.
The biggest news of the week was the release of GPT-4 for use in ChatGPT. GPT-4 offers longer context in messages and the ability to analyze visual inputs, making it a multimodal model. The AI community was set ablaze by the higher quality and more accurate responses that GPT-4 provided.
Microsoft confirmed that for the last five weeks, Bing has been running on GPT-4, which they customized for search. This means that users have already experienced an early version of this powerful model.
Mid-Journey announced the launch of Mid-Journey Magazine and the release of Mid-Journey Version 5, which creates more realistic images and better recognizes natural language. The company is also working on an API to allow other products to use Mid-Journey for art generation.
Microsoft announced their 365 Co-pilot, which adds AI functionalities to tools like Word, Excel, PowerPoint, Outlook, and Microsoft Teams. This will create a fine-tuned chatbot for users, providing answers based on data from all Microsoft tools.
Baidu released Ernie, a ChatGPT rival, which received an underwhelming response from both the public and shareholders, causing the company's stock to slump by 10%.
Khan Academy has long been a pioneer in providing free, high-quality education to people all over the world. This week Sal Khan, founder of Khan academy, announced the integration of GPT-4 into their learning management system. The first large scale implementaiton we've seen.
This may have easily flew under most people's radar but this has massive implications for education moving forward. Read the full article here:
The GTC event, NVIDIA's conference on AI, will take place on Monday 20th Match, and feature presentations from companies like DeepMind, OpenAI, Meta, and Adobe. Emad, the founder of Stability.AI, hinted at big announcements regarding Stable Diffusion, possibly including the release of Stable Diffusion 3.0 and an overhaul of the online user interface.
The exciting developments in AI this week have set the stage for even more breakthroughs in the coming weeks and months. Subscribe to keep up with all the lates AI and prompt engineering news!
