Missed yesterdays AI news? Here's a video recap:

Today:

OpenAI GPT Store Launch 🔥 you are NOT prepared...

Today, let’s dives deep into the potential of OpenAI's GPT store and what it means for both creators and users.

In the ever-evolving landscape of technology and artificial intelligence, OpenAI is making waves with its latest venture, the GPT store. We'll explore the potential of custom GPTs, the rules for participation in the GPT store, and the immense market opportunities it presents.

WATCH THE VIDEO ON YOUTUBE

Microsoft, OpenAI hit with new lawsuit by authors over AI training

In Manhattan, authors Nicholas Basbanes and Nicholas Gage are suing OpenAI and Microsoft. They claim their books were used without permission to train AI models like ChatGPT. The authors, both ex-journalists, are steamed about their work fueling a big-money tech industry without seeing a dime. Their lawyer calls it "outrageous." 

Microsoft and OpenAI haven't commented yet. This lawsuit's part of a growing trend where writers are pushing back against tech giants using their work to smarten up AI without asking or paying up.

REUTERS

Microsoft Picks Dee Templeton as OpenAI Board Observer

Dee Templeton, a longtime Microsoft bigwig, just snagged a spot on OpenAI's board as a nonvoting observer. This move comes amid a shake-up in the boardroom after a short-lived boot of OpenAI's CEO. 

Templeton, who's been with Microsoft for over 25 years, handles tech and research partnerships and operations. She's already been to some board meetings, but this info comes from an insider who wanted to stay anonymous. Neither OpenAI nor Microsoft are spilling any beans on this.

BLOOMBERG

OpenAI’s news publisher deals reportedly top out at $5 million a year

OpenAI's making waves with news publishers, shelling out up to $5 mil a year to use their stories for AI smarts. This move gives a sneak peek into what AI big shots might pay for this kind of info. Apple's also in the game, eyeing a $50 mil deal with media folks for AI training data. Facebook and Google have made similar plays before. 

But now, there's a twist. Some companies are blocking AI crawlers like OpenAI's, and there's a buzz about copyright issues. Big names like The New York Times have even sued OpenAI. To sidestep these headaches, AI firms are buddying up with publishers, with bigwigs like Axel Springer and The Associated Press already on board. Plus, Google's cooking up an AI tool, Genesis, that could turn facts into news stories, showing AI's growing role in journalism.

THE VERGE

Nabla raises another $24 million for its AI assistant for doctors that automatically writes clinical notes

Nabla, a Paris-based startup, bagged $24 million in Series B funding for its AI assistant aimed at doctors. This tech whiz, backed by big names like Cathay Innovation, helps docs by writing clinical notes during consultations. 

Nabla, valued at $180 million, isn’t about replacing docs but easing their paperwork load. The brainchild of ex-Facebook AI chief Alexandre Lebrun, Nabla wowed folks in a demo, flawlessly handling medical chit-chat and whipping up reports in minutes. It's big in the U.S. and super cautious with patient data, using top-notch tech to keep things private. Now, Nabla's aiming higher, dreaming of an AI-driven healthcare future while staying grounded in today's reality.

TECHCRUNCH

PREVIEW AI to dominate this year's CES, even without OpenAI's Altman on hand

At this year's CES (Consumer Electronics Show) in Vegas, even without OpenAI's big boss Sam Altman, AI's gonna steal the show. Last year, AI went wild in tech, with companies pouring over $23 billion into it. Everyone's talking about mixing AI into their gear. From gadgets helping the visually impaired to high-tech car safety and facial recognition stuff, it's all about AI now. 

Big names like Germany's Bosch and Japan's NEC are showing off their AI tricks. Car companies are all about creating a personal touch with AI, making driving smoother and safer. There's buzz about AI in PCs and smartphones too, but folks aren't sure if they'd pay extra for it. 

REUTERS

🧠RESEARCH

TinyLlama: An Open-Source Small Language Model

TinyLlama, a new, smaller 1.1B model, rocks at learning language. It's trained on a trillion words, three times over, using some neat tech from Llama 2 and FlashAttention. Though it's not a heavyweight, it punches above its size, outdoing similar models. Plus, anyone can grab it from GitHub.

LLAMA PRO: Progressive LLaMA with Block Expansion

"LLAMA PRO" upgrades Large Language Models by adding new skills without losing old ones. It's a method where extra Transformer blocks get trained only on new stuff, making the model smarter without forgetting past knowledge. Focused on code and math, LLAMA PRO-8.3B stands out in general tasks, programming, and math, outperforming other models in its family. It's a big step in blending regular and programming languages, paving the way for smarter, more versatile language agents.

Understanding LLMs: A Comprehensive Overview from Training to Inference

This article dives into the growing use of Large Language Models (LLMs) like ChatGPT, focusing on cost-effective training and use. It covers how these models are built, fine-tuned, and improved, discussing topics like data handling, architecture, and optimization. Also, it looks at model compression and efficient computing for better performance, hinting at future advancements in the field.

Mobile ALOHA

Stanford's Mobile ALOHA project introduces a bimanual, low-cost robot system for versatile tasks, like cooking or office work. It's a step up in robotics, moving beyond simple tasks to more complex, everyday actions. They've made a robot that learns by watching people, getting smarter at stuff like cooking shrimp or using elevators. It's a big deal for robots doing real-life tasks without breaking the bank.

🛠️TOP TOOLS

ScribeMD - offers an AI platform for doctors, automating medical notes and reducing paperwork. It features a digital scribe with 98% accuracy, HIPAA compliance, and EHR integration, saving doctors time and ensuring patient data security.

Rosebud AI -  is a platform for rapid game development, using AI to transform ideas into game assets, levels, and storylines. It offers tools for building immersive worlds, animating sprites, creating AI-driven NPCs, and customizing characters. 

Blenny AI - offers a Chrome extension that adds AI-powered vision capabilities to the web browser. Users can capture screenshots of web content, which Blenny can then summarize, translate, or analyze using custom AI agents. 

AdCreative.ai -  is a tool for generating high-conversion ad creatives using AI. It streamlines the design process, offering text and headline generation, competitor insights, and integration with ad platforms. 

Bland.ai - offers the fastest conversational AI with sub-second responses, mimicking human speed and quality. It features instant calls and aims to deliver consistent, high-quality interactions.

📲SOCIAL MEDIA

🗞️MORE NEWS

AI weather and climate prediction face real-world tests

AI weather and climate predictions, using tools like Google's GraphCast, face big tests soon. These AI models, trained on historical data, will be compared to traditional, physics-based methods during upcoming snowstorms and blizzards in North America and Europe. Skeptics worry about AI's novelty and reliance on past data, fearing it might miss new weather patterns. However, AI offers faster, cheaper forecasts. IBM's Hendrik Hamann believes AI will increasingly assist, but not replace, traditional methods. AXIOS

China unveils new artificial intelligence guidelines for scientists and bans use in funding applications

China's new AI guidelines forbid its use in funding applications by scientists and institutions. These rules aim to address AI's role in research, focusing on AI-generated content not being co-authored. The debate includes AI's credit in discoveries and its comparison to human knowledge pursuit. AI's misuse, like in ransomware cases, is a concern. The rules allow AI in research but require clear labeling of its use. Opinions differ among researchers, with some viewing AI as an irreversible trend in research, while others criticize its shallow outputs. The guidelines, reflecting international practices, also address authorship and intellectual property issues. China prioritizes AI regulation, emphasizing risk prevention and promoting "socialist values." SCMP

These AI-powered apps can hear the cause of a cough

Researchers developed an AI-powered smartphone app that identifies tuberculosis from cough sounds. Useful in low-income countries, this method can help catch TB early, reducing its spread. The app, trained on thousands of coughs from Kenya, represents a growing interest in using AI for health diagnostics, including detecting COVID-19 and influenza. While not a replacement for traditional methods, it's a step forward in healthcare technology. MIT

Meta's AI chief says he turned down a job as a Google director in 2002 partly because of low pay

Meta's AI boss, Yann LeCun, passed up a Google director job in 2002, citing low pay and other factors. He needed cash for his kids then and didn't want to move to California. LeCun, a big deal in AI, thought he'd have changed Google's research culture, making it more daring and open. This decision came up as AI competition heats up, especially with ChatGPT. BUSINESS INSIDER

A timeline of Sam Altman’s firing from OpenAI — and the fallout

In a shocker, Sam Altman got axed as OpenAI's CEO, followed by Greg Brockman and three top researchers quitting. This mess began late November when Altman briefly left, then returned as CEO with a new board. Then, Altman and Brockman flirted with Microsoft, sparking OpenAI's board to consider a merger with Anthropic. Ultimately, the board's flip-flopping and internal power plays culminated in Altman's firing and the company's uncertain future. TECHCRUNCH

Prompt engineering: Six strategies for getting better results

This guide provides strategies for improving results with large language models like GPT-4. 

1. Writing clear instructions: Be explicit in your requests to get better results.

2. Including details in queries: Provide context and specifics for more relevant answers.

3. Asking the model to adopt a persona: Specify the persona for the model's responses.

4. Using delimiters: Clearly separate different parts of input for better comprehension.

5. Specifying task steps: Break complex tasks into simpler instructions.

6. Providing examples: Offer sample inputs or behaviors for the model to follow.

7. Specifying output length: Request responses of a particular length.

8. Providing reference text: Instruct the model to use external knowledge or reference materials.

9. Splitting complex tasks: Divide intricate tasks into manageable subtasks.

10. Giving the model time to think: Encourage the model to reason before providing an answer.

11. Using external tools: Utilize additional tools or data sources to enhance model performance.

12. Testing changes systematically: Evaluate modifications with a diverse set of test cases.

13. Evaluating with reference to gold-standard answers: Assess model outputs against known correct answers.

These strategies can help maximize the effectiveness of GPT-4 for various tasks and applications.

READ THE ARTICLE ON OPENAI