Good morning. It’s Wednesday, April 3rd.

Did you know: On this day in 1973, the first cell phone call in public was made by Martin Cooper.

In today’s email:

• OpenAI Opens Up ChatGPT for Access Without an Account

• Apple Researchers Develop AI that can 'See' and Understand Screen Context

• Chatbot Outperformed Physicians in Clinical Reasoning Study

• 10 New AI Tools

• Latest AI Research Papers

• AI Creates Comics

You read. We listen. Let us know what you think of the new format of the newsletter, a slightly abbreviated version with more links, by replying to this email.

OpenAI Opens Up ChatGPT for Access Without an Account

OpenAI simplified access to ChatGPT's basic functions by removing the signup requirement. This allows anyone to interact with the AI without registration.

The update aligns with OpenAI's goals of broader AI access, particularly to students. Advanced features require a ChatGPT Plus subscription. While accountless users miss functionalities like chat history saving, OpenAI prioritizes privacy with opt-out options for conversation data used in model training. Read more.

Apple Researchers Develop AI that can 'See' and Understand Screen Context

Apple researchers introduced ReALM (Reference Resolution As Language Modeling), an AI system designed to understand ambiguous references within conversations and on-screen elements.

ReALM tackles this challenge by converting it into a language modeling problem, leveraging the power of large language models. This approach, coupled with the reconstruction of screen layouts based on identified elements, generates a textual representation that aids in comprehending visual context. Compared to existing methods and even the advanced GPT-4 model, ReALM demonstrates significant improvements, particularly in handling on-screen references.

Apple likely has plans to allow ReALM to “read” context on your screen, and be able to perform AI functions by analyzing the displayed data, possibly moving one step closer to fully automated agents.

However, the system relies on automated screen parsing, which has limitations. Future advancements require integrating computer vision and multi-modal techniques for handling complex visual references. Read more.

Chatbot Outperformed Physicians in Clinical Reasoning in Head-to-Head Study

A new study suggests that ChatGPT-4 may be able to outperform doctors in clinical reasoning tasks. Researchers at Beth Israel Deaconess Medical Center (BIDMC) used a standardized test (r-IDEA score) to assess how ChatGPT-4 compared to 21 attending physicians and 18 residents in analyzing medical data.

While the AI achieved higher scores overall, it also exhibited more instances of incorrect reasoning compared to residents. This underscores the potential of AI to augment human clinical reasoning rather than replace it entirely. Read more.

🖇️ Etcetera

> Gen-AI Search Engine Perplexity Has a Plan to Sell Ads

> Billie Eilish and Katy Perry among 200 artists protesting AI's "assault on human creativity"

> Yahoo is buying Artifact, the AI news app from the Instagram co-founders

> Exclusive: NYT to soon offer most articles via automated voice  

> Microsoft’s AI advantage is so critical it can’t be quantified, analyst says

> OpenAI's DALL-E 3 gets a text-based image editor right inside ChatGPT

> US, Britain announce partnership on AI safety, testing

> Expansion to Japan: OpenAI opens Tokyo office

> Artificial intelligence will help pick Nobel Prize for Physics winners

> Anthropic researchers detail how 'many-shot jailbreaking' can manipulate AI responses

Dealwise is an AI M&A advisor for startups. Facilitates acquisitions for $1-10M revenue firms, leveraging AI, hybrid search, and confidential processes. Cutting-edge features include hybrid search.

RAFA AI is an AI-driven investment platform. It monitors markets, portfolios, offers insights, risk analysis, alerts on options activities. Features include personalized news, performance tracking, strategy testing.

Creatie is an AI-driven design platform. Features include AI image enhancer, Magicon for 3D assets, automatic style guide, design auditor, and effortless migration.

Hoory AI is an AI-powered customer support assistant. Features include knowledge-based and rule-based AI assistants, omni-channel inbox, multilingual support, automation, macros, and advanced reporting.

DryMerge is a workflow automation tool that streamlines tasks like sales notifications, meeting summaries, and customer service escalations. Integrates with Slack, Gmail, Notion.

Qwen 1.5 MoE is a small yet powerful mixture-of-expert (MoE) model by Alibaba with 2.7 billion parameters, matching the performance of larger 7B models like Mistral 7B and Qwen1.5-7B. Efficient and cost-effective, it offers impressive performance, making it a notable addition to the AGI landscape.

AIxBlock offers an end-to-end AI platform harnessing decentralized computing resources through blockchain. Builders save up to 90% on compute costs, while suppliers earn more.

Embed AI’s White Label AI Chatbot enhances customer engagement through brand customization, advanced AI, multi-platform integration, analytics dashboard, 24/7 support, and easy setup for scalable business solutions.

Linfo.ai, an advanced AI-powered summarization and mind mapping extension, simplifies digesting lengthy content, making it valuable for research.

Map This is a specialized tool converting PDFs into visual mind maps, facilitating information organization. It aids in data extraction, cognitive enhancement, and streamlined communication of complex ideas.

📄 MambaMixer: Efficient Selective State Space Models with Dual Token and Channel Selection

📄 Getting it Right: Improving Spatial Consistency in Text-to-Image Models ​​

📄 FlexiDreamer: Single Image-to-3D Generation with FlexiCubes

📄 MaGRITTe: Manipulative and Generative 3D Realization from Image, Topview and Text  

📄 Direct Preference Optimization of Video Large Multimodal Models from Language Model Reward 

📄 3D Congealing: 3D-Aware Image Alignment in the Wild

📄 Octopus v2: On-device language model for super agent

📄 Bigger is not Always Better: Scaling Properties of Latent Diffusion Models  

📄 Are large language models superhuman chemists?

📄 LLaVA-Gemma: Accelerating Multimodal Foundation Models with a Compact Language Model

AI Creates Comics

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, apply here.