- unwind ai
- Posts
- OpenAI and Anthropic Might Merge 🤝
OpenAI and Anthropic Might Merge 🤝
PLUS: More Features in Runway GEN-2, Microsoft Releases Orca-2
Today’s top AI Highlights:
OpenAI Drama Doesn’t Seem to End
Runway Updates GEN-2 for More Control
Teaching Small Language Models How to Reason
Full Stack Transformer Language Models with RL
Automate the Entire Process of Video Creation in ChatGPT
& so much more!
Read time: 3 mins
Latest Developments 🌍
OpenAI is in a Deeper Hole ⛳️
Continuing on the developing drama of OpenAI, here’s what has happened after yesterday’s edition went out:
Employees at OpenAI have written to the Board that it "lacks competence, judgment, and care for our mission and employees" and have threatened to leave OpenAI to join the new Microsoft subsidiary that'll be run by Altman and Brockman.
Till now, 700+ out of 770 employees including Ilya Sustkever (Chief Scientist and a Board member who had initially) have signed the letter.
Consequent to the anticipation of the fall of OpenAI, customers using GPT API for their AI products have started looking for alternatives.
However, it has not been finlalized yet that Sam Altman and Grek Brockman will join Microsoft. They still want to return to OpenAI if the Board resigns. Microsoft is open to both the options of hiring the OpenAI employees or maintianing the strategic partnership with the company in other case. [Source: The Information]
OpenAI has approached the co-founder and CEO of Anthropic to discuss a potential merger of the two companies. [Source: The Information]
Along with Microsoft’s open position, Salesforce has also offered OpenAI employees who have tendered their resignation to join its Salesforce Einstein Trusted AI research team.
Express Your Vision with Greater Control 🌄
Runway’s GEN-2 has received further updates to give you more precise control and greater fidelity to your videos. The Motion Brush released earlier lets you animate specific parts of an image with a brush stroke. This has been complemented by updates in the Director Mode and an improved image model so your creation exactly matches your vision.
Key Highlights:
Gen-2 Style Presets: These presets let you generate content using curated styles without complicated prompting, covering a wide range of styles such as Abandoned, Anime, Cartoon, Cinematic, Digital Art, Impressionist Painting, and many more.
Director Mode Update: This update enhances the advanced camera controls in Director Mode, allowing for more granular control. You can adjust camera moves with greater precision using fractional numbers to control the direction and extent of movement.
Improved Image Model: Brings improved fidelity, greater consistency, and higher resolution image generations in Text to Image, Image to Image, and Image Variation modes.
Teaching Small Language Models How to Reason 👩🏫
While smaller language models are known for their efficiency and ease of deployment, they often lag behind the larger models like GPT-4 and PaLM in reasoning capabilities. Microsoft's latest method aims to bridge this gap. It focuses on enhancing the reasoning abilities of these smaller models, equipping them with advanced techniques to handle complex tasks more effectively.
Key Highlights:
The method significantly enhances the reasoning capabilities of smaller language models. It trains them in diverse reasoning techniques like step-by-step processing, recall-then-generate, and direct-answer strategies. It also teaches these models to identify and employ the most appropriate reasoning strategy for each specific task, enabling them to perform complex tasks more effectively.
It utilizes Explanation Tuning that enriches the learning process by training models on detailed reasoning signals. It helps smaller models understand and process tasks more deeply, particularly in challenging reasoning scenarios.
Microsoft’s new series of small models Orca-2 outperforms its counterparts across various benchmarks, showing its adaptability in handling reasoning capabilities, knowledge and language understanding, text completion, multi-turn conversations, and others. These methods make smaller models more versatile for a range of AI applications.
Train Transformer Language Models with RL 🤖
Hugging Face has introduced a new library named trl, specifically designed for training transformer language models with Reinforcement Learning, making language models more responsive and aligned with human preferences. Built upon the robust transformers library, trl offers seamless integration with pre-existing models.
Its highlight features include the SFTTrainer, RewardTrainer, and PPOTrainer, each catering to different stages of model training and fine-tuning. For instance, the SFTTrainer simplifies the process of adapting language models to specific datasets, while the RewardTrainer and PPOTrainer offer sophisticated mechanisms for refining models based on human feedback and reinforcement learning techniques.
Tools of the Trade ⚒️
Video Creation GPT: Automate the entire process of making videos. Just tell ChatGPT the purpose of the video and ChatGPT will handle everything, from ideating to generating fully editable videos.
Axcent: Collaborative and individual AI-enhanced content creation and brainstorming that offers direct blog publishing, AI assistance, team collaboration, ability to have multiple projects in one place and AI memory and context creation tools, all in one space.
Cypher: Create your AI clone that speaks just like you and hears your voice too. Just answer a few questions and your clone is ready.
Rtila: An automation software that enables users to create and monetize their own automation tools by turning repetitive tasks into standalone software applications, with capabilities like data mining, scrapping, and formatting.
😍 Enjoying so far, TWEET NOW to share with your friends!
Hot Takes 🔥
OpenAI will just become the 'AI Safety' arm of Microsoft when you boil it down over time. All the cautious folks are in one place to manage at a distance while Sam gets to play at the bleeding edge. ~ Theo
So it looks like this, was not that. I didn't think it was. Too early. But the thought that keeps running through my mind is that in 3 years the popcorn-eating will start for weird drama between Demis Hassabis and Google, and a month later everyone will be dead. ~ Eliezer Yudkowsky
Agi has been achieved internally, there’s no other explanation ~ Sully Omarr
Meme of the Day 🤡
That’s all for today!
See you tomorrow with more such AI-filled content. Don’t forget to subscribe and give your feedback below 👇
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!
Reply