Apple's 3B LLM Outperforms GPT-4 💥
Apple's 3B LLM Outperforms GPT-4 💥
PLUS: Chat-to-edit images in DALL.E, Indeed's AI-powered Smart Sourcing, TikTok for news
Today’s top AI Highlights:
DALL.E-3 now lets you edit images within ChatGPT
Apple’s new AI models to make Siri smarter, outperforms GPT-4
Indeed’s AI-powered Smart Sourcing to make hiring faster
TikTok-for-news app Artifact to get the scale it never got
First autonomous API that goes from natural language to web actions
& so much more!
Read time: 3 mins
Latest Developments 🌍
Edit Images with DALL.E 🪄
DALL.E-3 now lets you edit generated images with simple text prompts within ChatGPT. It’s like the inpainting feature in Adobe’s Firefly where you can select an area of the image and put a prompt in natural language to make the desired edits. You can add, remove, or modify the objects in the selected area.
To edit the image: Click on the generated image and click on the Select icon. Adjust the size of the selection tool to paint the area where you want to make the edits. Add your text prompt in the Edit Selection box and that’s it!
The feature is being rolled out in both the website and ChatGPT mobile app.
Apple’s AI to See More Than What’s On-screen 👀
Apple has released ReALM (Reference Resolution As Language Modeling), a new method for improving how AI understands references made during conversations and to items displayed on a screen or operating in the background. Imagine asking your phone to “call the top pharmacy on the list” without specifying which one – ReALM aims to figure out exactly what you mean.
This is significantly going to improve the user experience and functionality of assistants like Siri for on-device assistance.
Key Highlights:
Method: ReALM translates the visual and contextual information on the screen into a text format that AI can interpret. This involves tagging screen elements to provide the AI with clear context about each item’s role and relevance.
Size and Efficiency: The ReALM models are significantly smaller in size than other vision language models, with the largest being ReALM-3B. Even the smallest model with 80 million parameters shows over 5% gains for on-screen references. The size allows the models to be deployed on small devices like smartphones.
Comparison: ReALM’s smallest model outperforms GPT 3.5 significantly and matches GPT-4 performance. The large models outperform GPT-4 across all domains, establishing a more potent solution.
Example: You ask your assistant “Show me the nearest grocery stores.” The phone displays a list of stores. But then you decide to go to the store where you have loyalty points but can’t remember its name. Without looking back at your screen, you say, “Navigate to the one where I have a loyalty card.”
The model here understands this context by analyzing both the current screen content (the list of grocery stores) and the conversational history (your request to see stores and mention of a loyalty card).
Faster Hiring, Better Job Profile, Improved Connections 💼
Indeed has just introduced an AI-powered suite called Smart Sourcing to simplify the hiring process for employers by matching their requirements with job seekers’ skills efficiently. By using a vast database of nearly 300 million workers, Smart Sourcing facilitates better connections between companies and potential employees to enhance the job search and recruitment experience for both parties.
Key Highlights:
AI-powered Matching: This is much different than keyword searching or matching just words. The company leverages its AI-powered matching engine to offer instant recommendations of ideal candidates based on the specific requirements of an open job.
Efficiency for Recruiters: Focusing on individuals actively looking for new opportunities, it significantly reduces the time hiring managers spend on sourcing candidates.
Better Profile: The updated Indeed Profile lets job seekers better display their unique work experiences, skills, and job preferences with over 40,000 types of skills listed.
AI Assistants: Job seekers can use features like Work Experience Writer, which uses AI to help craft and update work experience descriptions, and a Multi-Resume feature to store up to 5 versions of a resume in their profile. Hiring managers can also use custom AI-powered messages for communicating with candidates.
Yahoo to Buy Artifact for its Recommendation Tech 🗞️
Yahoo is acquiring Artifact news app by former Instagram co-founders Mike Krieger and Kevin Systrom. Artifact, a personalized news feed app that uses AI to curate articles and stories for its users, was launched last year. The TikTok-for-news app however was not able to gain many users and three months ago, it announced shutting down the platform. Since then, it reportedly gained interest from many potential acquirers.
The financial terms of this acquisition have not been disclosed but it focuses on Artifact’s technology rather than its team. Yahoo saw value in Artifact’s content taxonomy and recommendation systems, which were designed with significant care. The deal will expose Artifact’s AI to over 185 million monthly visitors to Yahoo News where it can be utilized at a scale.
Our opinion: Yahoo had been reportedly working on personalization and recommendations for its news platform and this deal gives exactly what Yahoo has been looking for but the ultimate success depends on the effective implementation and the ability to stand out in the market.
Let us know what you think in the comments below! 👇
😍 Enjoying so far, share it with your friends!
Tools of the Trade ⚒️
SWE Agent: Turn language models like GPT-4 into software engineers that fix bugs and issues in real GitHub repositories, achieving an SWE-bench score of 12.29% (Devin’s score is 13.86%). It uses an Agent-Computer Interface to facilitate interactions between the language model and the repository, improving the efficiency of repository-level coding agents.
RAGFlow: An opensource RAG engine based on deep document understanding. It offers a streamlined RAG workflow for businesses of any scale, combining LLM to provide truthful QA capabilities, backed by well-founded citations from various complex formatted data.
MultiON’s Agent API: Embed AI agents into devices and applications that autonomously perform tasks and workflows on the web. The API can be used across various industries, for example, it can be integrated into smart devices to act as a voice assistant for tasks like calling rides or making reservations.
Mention Moose: Promote your business on Reddit with AI, targeting conversations with tailored keywords to engage with your target audience and boost sales. Customize your settings for strategic placements and keyword targeting to meet your business objectives.
Hot Takes 🔥
Two weird things that are going to happen in marketing:
1) Marketing to AIs, as people increasingly ask AIs for advice, the goal is "persuading" AIs, not people, to prefer a solution
2) Selling AIs on tools. LLM agents decide what tools to use, how do you get them to pick yours? ~
Ethan MollickOpenAI is making GPT 3.5 instantly available - i.e., no login is required.
The big problem is that 3.5 isn't a very good model! Even open-source models beat it handily.
What OpenAI should be doing is to make GPT-4 available instantly, and after a person does a few queries, add a login-gate
Hook them before you can bill them :) ~
Bindu Reddy
Meme of the Day 🤡
That’s all for today! See you tomorrow with more such AI-filled content.
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!