Posts in category: Resources

🔬 The Oncology Counterfactual Therapeutics Oracle (OCTO) represents a groundbreaking approach in personalized oncology, merging AI and biology to discover new cancer treatments.

🌐 What sets OCTO apart?

• Multimodal Data Integration: OCTO analyzes diverse patient data, including proteins, genes, and tumor sequences.

• Hypothetical Simulations: It answers critical questions like “What if we increase gene X expression?”

• Innovative Learning: OCTO detects cell structures independently and excels in zero-shot learning.

🚀 Practical Impact:

• Predicts new drug targets and tests them in vivo.

• Enhances our understanding of tumor-immune interactions.

🌟 As we advance, AI models like OCTO will play a crucial role in developing personalised, effective cancer therapies.

About OCTO.

Meta will withhold its next multimodal LLaMA AI model — and future ones — from customers in the European Union due to regulatory uncertainties. This decision sets the stage for a confrontation between Meta and EU regulators and highlights a trend of U.S. tech giants withholding products from European markets.

💬 “We will release a multimodal Llama model over the coming months, but not in the EU due to the unpredictable nature of the European regulatory environment,” Meta stated. This move impacts European companies and prevents non-EU companies from offering products in Europe that utilize these models.

🔍 Meta’s issue isn’t with the still-being-finalized AI Act, but rather with how it can train models using data from European customers while complying with GDPR — the EU’s existing data protection law. Meta announced in May its intention to use publicly available posts from Facebook and Instagram to train future models, offering EU users a means to opt out. Despite briefing EU regulators months in advance and receiving minimal feedback, Meta was ordered to pause the training in June, leading to numerous questions from data privacy regulators.

🇬🇧 Interestingly, Meta does not face the same level of regulatory uncertainty in the U.K., which has nearly identical GDPR laws, and plans to launch its new model there.

🌍 The broader picture reveals escalating tensions between U.S.-based tech companies and European regulators, with tech firms arguing that stringent regulations harm both consumers and the competitiveness of European companies.

🔑 A Meta representative highlighted the importance of training on European data to ensure products accurately reflect regional terminology and culture, noting that competitors like Google and OpenAI are already doing so.

Meta will not launch multimodal Llama AI model in EU

OpenAI announced the launch of GPT-4o mini, OpenAI most cost-efficient small model yet! 💡

🔹 Cost-Effective Intelligence: Priced at just 15 cents per million input tokens and 60 cents per million output tokens, GPT-4o mini is an order of magnitude more affordable than previous models and over 60% cheaper than GPT-3.5 Turbo! 💸

🔹 Versatile and Powerful: With a context window of 128K tokens and support for up to 16K output tokens per request, GPT-4o mini excels in tasks like chaining multiple API calls, handling large volumes of context, and providing real-time text responses. It’s perfect for customer support chatbots, code analysis, and more! 🔍💬

🔹 Superior Performance: GPT-4o mini outperforms other small models on key benchmarks:

• 82.0% on MMLU (textual intelligence and reasoning)

• 87.0% on MGSM (math reasoning)

• 87.2% on HumanEval (coding performance)

• 59.4% on MMMU (multimodal reasoning)

🔹 Safety First: Built-in safety measures ensure reliable and safe responses, thanks to advanced techniques like reinforcement learning with human feedback (RLHF). 🛡️

🔹 Accessibility: Available now in the Assistants API, Chat Completions API, and Batch API. Free, Plus, and Team users can access GPT-4o mini today, with Enterprise users joining next week.

Why GPT-4o Mini Over GPT-4? 🆚

GPT-4o mini is designed to make AI more accessible and affordable without compromising on performance. It’s perfect for applications requiring low cost and latency, enabling developers to build scalable AI solutions efficiently. While GPT-4 offers more advanced capabilities, GPT-4o mini provides a high-performance alternative at a fraction of the cost, making it ideal for a broader range of use cases. 🌐

Get ready to unlock new possibilities with GPT-4o mini and join us in making AI accessible to all! 🌟

GPT-4o mini: advancing cost-efficient intelligence

Ensuring that AI-generated text is understandable is vital for their effective use, especially in complex tasks like math problem-solving. Recent research has shown that optimizing AI for just correctness can make the text harder to understand, leading to more errors in human evaluation.

🔎 Prover-Verifier Games involve a “prover” generating a solution and a “verifier” checking its accuracy. This approach not only ensures correctness but also makes the text easier for both humans and other AI systems to verify.

🔧 By training strong models to produce solutions that weaker models can easily verify, we enhance the legibility and trustworthiness of AI outputs. This method, which focuses on clear and verifiable justifications, promises to improve AI’s reliability in critical applications.

Areas for Improvement in Business Applications:

1. Customer Support: AI-generated responses can be clearer and easier for customers to understand, enhancing user satisfaction.

2. Documentation: Simplifying technical documentation, making it more accessible for users and support teams.

3. Decision-Making: Providing clear and verifiable insights for business strategies, ensuring stakeholders understand and trust AI recommendations.

4. Compliance and Reporting: Generating transparent and understandable compliance reports, aiding in regulatory adherence.

5. Training and Onboarding: Creating legible and easy-to-verify training materials, improving learning outcomes for employees.

Implementing Prover-Verifier Games can significantly boost the effectiveness and reliability of AI in these areas, leading to more transparent and trustworthy business processes.

🌟 Key Takeaways:

1. Improved Legibility: Making AI outputs easier for humans and weaker models to verify.

2. Trust and Safety: Enhancing the reliability and transparency of AI in real-world applications.

3. Future Alignment: Reducing reliance on human oversight, crucial for aligning superintelligent AI with human values.

Prover-Verifier Games improve legibility of LLM outputs

Andrej Karpathy, former Tesla AI Director and OpenAI co-founder, has just announced his new venture: Eureka Labs! 🌟

Eureka Labs aims to revolutionize education by leveraging Generative AI to create a new kind of AI-native school. With an AI teaching assistant, students will be guided through course materials crafted by real teachers, making high-quality education accessible to all. 📚✨

Karpathy’s vision is to make it easier for anyone to learn anything, anytime, anywhere. His first course, LLM101n, will teach students how to train their own AI, blending digital and physical learning experiences. 🌐🤖

Karpathy’s extensive background in AI and education, from his time at Tesla to his YouTube tutorials and Stanford courses, makes this new venture one to watch! 👀

Stay tuned as Eureka Labs paves the way for a future where AI amplifies human potential. What would you like to learn next? 🤔

Andrej Karpathy unveils Eureka Labs, an AI education start-up

In the dynamic world of GenAI applications, choosing the right tool for your business needs can make all the difference. Here’s a comparative look at ChatGPT-4o and Microsoft Copilot, two leading solutions, across various business functions:

Operational Efficiencies 🏢

ChatGPT-4o: Excels in automating repetitive tasks and streamlining workflows with customizable capabilities.

Microsoft Copilot: Integrated seamlessly into Microsoft 365, enhancing productivity within familiar environments.

Document Processing and Annotation 📄

ChatGPT-4o: Offers advanced NLP capabilities for detailed document analysis and annotation.

Microsoft Copilot: Provides efficient document editing and summarization directly within Office applications.

Writing Emails 📧

ChatGPT-4o: Generates context-aware, personalized emails tailored to specific business needs.

Microsoft Copilot: Helps draft and refine emails with suggestions to improve clarity and tone.

Preparing Presentation Decks 📊

ChatGPT-4o: Creates comprehensive and visually appealing presentation content based on input data.

Microsoft Copilot: Assists in building and enhancing presentations within PowerPoint, leveraging its deep integration.

Analysing Business Data Sets 📈

ChatGPT-4o: Utilizes powerful AI to extract insights and trends from complex data sets for strategic decision-making.

Microsoft Copilot: Integrates with Excel to provide data analysis and visualization directly in your spreadsheets.

Tracking Appointments 📅

ChatGPT-4o: Manages schedules and reminders efficiently with customizable notification settings.

Microsoft Copilot: Syncs with Outlook to streamline calendar management and scheduling tasks.

Writing Proposals 📑

ChatGPT-4o: Crafts detailed, persuasive proposals with a focus on specific business objectives.

Microsoft Copilot: Aids in structuring and refining proposals, ensuring they meet professional standards.

Responding to RFI/RFP During Bidding 📝

ChatGPT-4o: Generates comprehensive and competitive responses tailored to specific RFI/RFP requirements.

Microsoft Copilot: Provides templates and guidance to create structured and compelling RFI/RFP responses.

Integration of Advanced AI

Microsoft Copilot integrates advanced AI technologies from OpenAI, including foundational models like GPT-4, Supervised Fine-Tuning (SFT), and Reinforcement Learning from Human Feedback (RLHF). These enhancements, provided by OpenAI, are tailored by Microsoft to fit seamlessly into their ecosystem, ensuring users benefit from state-of-the-art AI technology.

Both ChatGPT-4o and Microsoft Copilot bring unique strengths to the table. The choice ultimately depends on your specific business needs and existing ecosystem. For a more flexible, customizable AI solution, ChatGPT-4o is your go-to. For seamless integration within Microsoft 365, Copilot stands out.

If there is no need for integration into Microsoft Office, ChatGPT-4o tends to be more versatile and accurate for bid writing. Here’s why:

Versatility

ChatGPT-4o:

• Customization: ChatGPT-4o can be fine-tuned to specific bid-writing requirements, making it highly adaptable to different industries and bid formats.

• Flexibility: It supports a wide range of languages and can be used across various platforms and applications, not being tied to a specific ecosystem.

• Advanced NLP Capabilities: With advanced natural language processing, ChatGPT-4o can handle complex bid requirements, ensuring that all critical details are covered accurately.

Microsoft Copilot:

• Integration-Dependent: While powerful within the Microsoft ecosystem, its versatility is somewhat limited outside it.

• Streamlined for Office: Copilot is designed to enhance productivity within Microsoft Office applications, which might limit its adaptability for bid writing outside this environment.

Accuracy

ChatGPT-4o:

• Advanced AI Models: Built on the latest GPT-4 model, it offers high accuracy in generating context-aware and detailed content.

• Continuous Learning: Utilises reinforcement learning from human feedback (RLHF) to improve over time based on user interactions, ensuring responses are precise and relevant.

Microsoft Copilot:

• Office-Optimised: Its accuracy is optimised for tasks within the Office suite, such as document editing, email drafting, and presentation creation, but it may not perform as well outside these contexts.

Summary

Without the need for integration into Microsoft Office, ChatGPT-4o stands out as the more versatile and accurate option for bid writing. Its customisation options, advanced NLP capabilities, and flexibility make it a superior choice for crafting detailed and competitive bids.

Feel free to share your experiences with these tools in the comments! 👇

🚀 We introduce Florence-2, a cutting-edge vision foundation model with a unified, prompt-based representation for diverse computer vision and vision-language tasks. Unlike existing models, Florence-2 handles various tasks using simple text instructions, covering captioning, object detection, grounding, and segmentation. It relies on FLD-5B, a dataset with 5.4 billion visual annotations on 126 million images, created through automated annotation and model refinement. Florence-2 employs a sequence-to-sequence structure for training, achieving remarkable zero-shot and fine-tuning capabilities. Extensive evaluations confirm its strong performance across numerous tasks.

Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks

🚀 Problem: Large Language Models (LLMs) have static knowledge, making updates costly and time-consuming. Retrieval-augmented generation (RAG) helps, but irrelevant info can degrade performance.

🔧 Solution: Retrieval Augmented Iterative Self-Feedback (RA-ISF) refines RAG by breaking tasks into subtasks.

It uses:

1. Task Decomposition: Splits tasks into subtasks.

2. Knowledge Retrieval: Fetches relevant info for each subtask.

3. Response Generation: Integrates info to generate accurate answers.

🌟 What’s Next: RA-ISF reduces hallucinations and boosts performance, enhancing LLMs for complex tasks. As it evolves, expect more powerful, knowledge-enhanced LLMs.

RA-ISF: Learning to Answer and Understand from Retrieval Augmentation via Iterative Self-Feedback

In the rapidly evolving AI landscape, relying on external APIs and open-source models presents significant challenges.

Villains:

1. External APIs: These pose data confidentiality risks and the constant threat of access being revoked, as seen in recent restrictions towards China. Additionally, monopolistic tendencies can lead to unpredictable and steep price increases. 🔒🚫💸

2. Open-Source Models: While they mitigate some risks, their deployment is costly due to significant hardware requirements. This creates a profitability challenge for businesses trying to avoid external API dependencies. 💻💰

Solutions:

1. Mitigating Reliance on External APIs:

• By researching and identifying the best applicable LLM models and potentially developing our own foundation model, we can ensure data privacy and control. This approach also protects against sudden API price hikes and service discontinuations. 🔍🔐📈

2. Cost-Efficiency with Open Source Models:

• Implementing a strategy involving initial Supervised Fine Tuning (SFT) to enhance model performance, followed by rigorous optimization, allows us to balance high accuracy with cost-efficiency. This dual approach ensures profitability while maintaining operational autonomy. ⚙️💡📉

3. Optimizing Business Processes:

• Integrating GenAI models seamlessly into existing business workflows is crucial. This involves developing algorithms that enhance current processes, making AI solutions highly practical and efficient. For instance, in healthcare, an optimized GenAI model can significantly improve patient appointment scheduling, ensuring accuracy and efficiency while reducing administrative burdens. 🏥📅🤖

Call to Action:

By focusing on data privacy, cost-efficiency, and seamless integration into existing workflows, we can drive tangible improvements and foster sustainable growth. Let’s work together to navigate these challenges and unlock the full potential of AI.

Abstract:

Recent advancements in large language models (LLMs) have significantly impacted various domains. However, the delay in knowledge updates and the presence of hallucination issues limit their effectiveness in knowledge-intensive tasks. Retrieval augmented generation (RAG) offers a solution by incorporating external information. Traditional RAG methods primarily rely on similarity to connect queries with documents, following a simple retrieve-then-read process. This work introduces MetRag, a framework that enhances RAG by integrating multi-layered thoughts, moving beyond mere similarity-based approaches. MetRag employs a utility model supervised by an LLM to generate utility-oriented thoughts and combines them with similarity-oriented thoughts for improved performance. It also uses LLMs as task-adaptive summarizers to condense retrieved documents, fostering compactness-oriented thought. This multi-layered approach culminates in a knowledge-augmented generation process, proving superior in extensive experiments on knowledge-intensive tasks.

Key Contributions:

• Utility-Oriented Thought: Incorporates a small-scale utility model for better relevance.

• Compactness-Oriented Thought: Utilizes LLMs to summarize large sets of retrieved documents.

• Knowledge-Augmented Generation: Combines multiple thought layers for enhanced output.

Significance: MetRag demonstrates improved performance in knowledge-intensive tasks by addressing the limitations of traditional RAG methods through a multi-layered thought approach.

Applications: This framework can be applied to various domains requiring up-to-date and accurate knowledge, enhancing the reliability and efficiency of LLMs in real-world tasks.

Similarity is Not All You Need: Endowing Retrieval Augmented Generation with Multi Layered Thoughts