- AI Emergence
- Posts
- AI Divide Comes Out - OpenAI’s SORA Leaked!!
AI Divide Comes Out - OpenAI’s SORA Leaked!!
Along with: Anthropic Makes Strides in Cloud and Open-Sources Data Integration Tool.
Analytics Vidhya (Curated by Kunal Jain)
November 28, 2024
n@
Hey there,
I believe that we will see a new divide emerge in the world in the coming days.
I refer to this divide as the “AI divide”. You might have seen/experienced this in some form already - you feel AI will replace jobs, and your friend feels AI is not replacing anyone. You think the use of creatives for training AI models is unethical, your friend is actually working on it! You get the drift!
This week a group called “Sora PR puppets” attacked/expressed their disapproval of how OpenAI is using their work. The news publishers have already been fighting a similar battle.
I think these battles are going to last long, with no clear winners, and will have different takes in different geographies and governments - so brace yourself for these in coming years!
For the time being - let’s look at the developments this week.
What would be the format? Every week, we will break the newsletter into the following sections:
The Input - All about recent developments in AI
The Tools - Interesting finds and launches
The Algorithm - Resources for learning
The Output - Our reflection
Table of Contents
A group known as “Sora PR Puppets” has made temporary access to OpenAI’s unreleased video generator, Sora, available on Hugging Face. This access enables users to create 10-second, 1080p videos from text prompts.
The group has accused OpenAI of exploiting artists for unpaid labor under the pretense of collaboration, labeling the program as “art-washing.”
They assert that artists are being utilized as unpaid bug testers, public relations tools, and sources of training data, with compensation that is significantly lower than the substantial benefits that OpenAI accrues.
The group has also criticized the program’s restrictions, such as the requirement for approval of all outputs, and accused OpenAI of prioritizing public relations over genuine artistic support.
The group demands fair compensation, increased transparency, and meaningful support for artists. They urge creators to explore open-source tools for greater creative freedom and advocate for ethical AI practices.
In response, OpenAI has stated that Sora is currently in a research and development phase, with participation being voluntary and focused on maintaining safety measures.
This is the first time when a group of people has made an active protest in this manner highlighting the ethical concerns we are likely to see in coming years. (source)
Anthropic has open-sourced the Model Context Protocol (MCP), a universal standard for connecting AI systems with diverse data sources like content repositories, business tools, and development environments.
MCP streamlines integration by replacing fragmented solutions with a unified protocol. Using its pre-built servers and SDKs, developers can create secure, two-way connections between data and AI applications.
By addressing scalability challenges posed by information silos and legacy systems, MCP fosters interoperability and adaptability, leveraging open-source collaboration to reduce reliance on proprietary solutions. (source)
DeepSeek launched R1-Lite-Preview, a reasoning-focused large language model (LLM), through its DeepSeek Chat platform.
This model excels in logical inference, mathematical reasoning, and problem-solving, achieving performance that rivals or surpasses OpenAI's o1-preview on benchmarks like AIME and MATH.
It uses a transparent "chain-of-thought" reasoning method, documenting each step of its decision-making process for user clarity. Although DeepSeek has not released the model’s full code or detailed training methodology, they plan to make open-source versions and APIs available soon. (source)
Anthropic is enhancing its Claude AI with customizable communication styles and a powerful partnership with AWS, pushing the boundaries of generative AI for enterprises.
Claude Gets a Personal Touch: Anthropic has introduced a customizable "styles" feature for its Claude AI assistant, enabling users to adjust the AI's communication style to formal, concise, explanatory, or even personalized modes by uploading custom samples.
Designed specifically to enhance enterprise usability, this feature ensures consistent and context-specific AI interactions. Early adopter GitLab praised the update, highlighting its success in streamlining workflows across various use cases. (source)
From Claude to Cloud: Anthropic recently secured an additional $4 billion from Amazon, bringing Amazon's total investment to $8 billion while remaining a minority investor.
The partnership establishes Amazon Web Services (AWS) as Anthropic’s primary cloud and AI training provider, leveraging AWS Trainium and Inferentia chips.
Whereas from Anthropic’s Claude AI chatbot, AWS customers gain exclusive benefits like fine-tuning Claude with proprietary data. (source)
OLMo 2 releases and advances fully open language models by offering 7B and 13B parameter models trained on up to 5 trillion tokens.
These models outperform some proprietary counterparts like Llama 3.1 and Qwen 2.5 on key benchmarks while maintaining complete transparency with weights, data, and training recipes.
By incorporating improvements in training stability, staged pretraining, and post-training fine-tuning, OLMo 2 sets new standards for open and reproducible AI.(source)
NVIDIA researchers developed Fugatto (Foundational Generative Audio Transformer Opus 1), a generative AI model that creates and transforms audio using text and audio prompts.
Fugatto composes music, modifies accents and emotions, and generates unique sounds for music, advertising, language learning, and gaming applications.
It excels at blending attributes, creating dynamic soundscapes, and producing entirely new, unheard sounds, offering unmatched versatility.(source)
Hugging Face developed SmolVLM, a series of small but powerful visual language models designed to be fast and efficient for a range of applications. It comes in three sizes—135M, 360M, and 1.7B parameters- providing developers with scalable options.
These models were trained on high-quality educational and synthetic datasets and fine-tuned to improve performance on tasks such as general knowledge questions, creative writing, and basic programming. They are optimized for local use and offer pre-trained and instructed models for easy integration into applications.
Currently, they are facing some challenges and may struggle with tasks that require complex or arithmetic reasoning.(source)
MIT researchers have developed Model-Based Transfer Learning (MBTL), to enhance the reliability and efficiency of reinforcement learning models in decision-making tasks like traffic control.
Instead of training AI on all tasks or each task separately, MBTL selects the most impactful tasks, enabling efficient learning and improved performance across all tasks. (source)
Tool: Sider
The Google Sider chatbot is a conversational AI tool or chatbot that integrates with Google’s suite of services as a complementary assistant. It may use natural language processing (NLP) to help with daily tasks such as organizing schedules, searching for information, or managing workflows through Google tools like Gmail, Calendar, Docs, and Sheets.
How to Access it
Click the link here.
Select "Add to Chrome."
Click on "Add Extension."
Sign in using your email ID.
You're all set! Use it seamlessly whenever you're browsing.
In the recent episode of Leading with Data, I had a great conversation with Didier Rodrigues Lopes, Founder and CEO of OpenBB. We discussed his experience as a sensor fusion engineer and how he feels that community-driven feedback and innovation are pivotal to building impactful open-source products like OpenBB.
This Free course ”GenerativeAI - A Way of Life” by Analytics Vidhya explores AI-driven text and image generation using tools like ChatGPT, Microsoft Copilot, and DALL·E3. It covers practical applications, ethical considerations, and strategies to harness generative AI for innovation across various industries.
Andrew Ng brings AIsuite to the rescue of the problem of juggling LLMs from multiple providers! This open-source Python package lets you effortlessly switch between language models like OpenAI's GPT-4, Claude, and Llama with just a single line of code, making AI integration as simple as changing a string.
This course “Reimagining GenAI: Common Mistakes and Best Practices for Success”, led by GenAI expert Shabazz Mohammed, explores the challenges of adopting generative AI and provides practical strategies to navigate them. It offers insights into real-world scenarios, overcoming implementation pitfalls, and ensuring scalable, ethical, and ROI-driven AI adoption.
What do you think about “the AI divide”? Would love to hear if you experienced it in some form. What was your reaction?
I am all ears.
How do you rate this issue of AI Emergence?Would love to hear your thoughts |
Reply