Aisulu Omar

Models like OpenAI's CLIP combine text and images for search, but they have limitations, particularly with longer texts and complex relationships. This article talks about JINA's new model that improves on CLIP by integrating both text and image embeddings in one model, making it faster and more efficient for multimodal search tasks. Experiments and Findings: → Search with Both Text and Images: Combining both embeddings leads to better and more diverse search results, improving relevance and variety in results. → Image-Driven Diversification: By using images for diversification, Jina-CLIP v1 can present a broader range of results, offering users more varied options. Results Comparison: → Using both text and image embeddings (via averaging) consistently outperformed searches based solely on text or images, providing more precise and diverse retrieval across both broad and fine-grained categories. https://lnkd.in/guNRQpBU

15

Here's why OpenAI o1 is a big deal. Up until now, the only way to make a model smarter was through training. You had to train on more data with more compute. These training runs are expensive. ChatGPT 4 cost about $100 million to train. Future models are expected to cost $1bn to train, then $10bn. You get where this is going. Once the model is trained, it's pretty much as smart as it's going to get. What they've done with OpenAI o1 is develop a technique where it can get smarter at run time (inference) just by pondering the problem longer. How much smarter it gets scales with how long it thinks. This is a breakthrough.

21

8 Comments

Intuitive Example, Principles, Code Explanation and Insights about CRAG Take your LLM performance to the next level. #llm #llmops #chatgpt4 #mlops #ai #datascience #ml #gemini tps://https://lnkd.in/dDkZiZ5m

13

🚀 Today we’re thrilled to be launching Semantic Search with Behavioral Re-Ranking 🚀 Shaped’s text retrieval models work hand-in-hand with behavior-driven re-rankers to give your users industry-leading search personalization in less than one sprint. Imagine you're an e-commerce site, and a user searches for "linen pants." A traditional semantic search might return a generic assortment of styles for both men and women. But what if that user previously clicked on the "menswear" category or purchased men's items? Suddenly, it’s obvious: the results should prioritize "men's linen pants," even if the user didn't explicitly specify it. This is where Shaped shines—using behavior-driven insights to deliver relevant, personalized search results, all based on the unique context of each session. Keen to learn more? Check out our announcement in the comments below 👇👇

32

2 Comments

📊 Optimizing RAG Applications with TruLens: A Quick Guide Recently explored TruLens for evaluating and fine-tuning Large Language Models (LLMs) in Retrieval Augmented Generation (RAG) applications. Here's what I learned: Key Evaluation Metrics: ✅ Groundedness: Ensures responses are factually supported by context ✅ Answer Relevance: Measures if responses directly address queries ✅ Context Relevance: Evaluates quality of retrieved information The Process: Set up automated evaluation pipelines using TruLens Track performance across different configurations Use metrics to optimize: Chunking strategies Retrieval parameters Context window sizes Embedding models Benefits: Data-driven optimization Automated quality assessment Systematic performance tracking Clear metrics for improvement Result: More accurate, relevant, and reliable RAG applications with measurable improvements in response quality. #AI #MachineLearning #LLM #RAG #DataScience #TruLens https://lnkd.in/gHQnjREi

3

Anthropic's prompt generation feature is good to have for generating baseline prompt to start with, and do iterative improvements as per corresponding prompts version evaluation performance. However, getting a good baseline prompt with this feature also relies upon how good of an ask or problem statement provided , like if problem provided is too generic but expecting it to work for some specific narrow scenarios, then that won't be an ideal scenario for this feature's usage.

8

If you like Python, use bstrap. It uses bootstrapping, which is basically the same idea as in this post. Simplifies my life a lot for any kind of data-driven claims. pip install bstrap https://lnkd.in/gwEqDZxJ

9

In this paper, I present a study of how biases can influence an experiment. The same design & data can provide very different results. Flawless decision making requires the right amount of skepticism. https://lnkd.in/gFNY9PxG

35

5 Comments

As I keep on saying : Prompt Engineering has never been a thing. It's wrong for 2 reasons : it makes you think it's deterministic (one great prompt= one great output, whereas LLMs are inherently stochastic machines, exactly like your Most Significant Other), and; it tricks you into thinking you need to be an engineer to get it right. Guess what: IT engineers have the most difficult time of their life with LLMs! And now we have a third, mike-dropping reason: smart bots can now engineer your most perfect prompt for you: just ask away, and add on!

9

I have read several posts about LLMs picking not-so-random numbers when asked to pick a random number. This study is comprehensive, and confirms the observation. https://lnkd.in/gU-4KGkD Of course, this was not exactly the right question to ask. LLMs' picks of random numbers are biased towards the training data, and that adversely affects the randomness. The more important question; what are the questions that we should avoid relying on generic LLMs to answer correctly?

4

New research shows that, indeed, o1 models are better at planning than even the best LLMs—but we're just getting started. o1 models crush LLMs on short planning tasks with low complexity, but still struggle with longer, more complex, and more unclear tasks—and with knowing when a task is unsolvable. Still, this seems like significant progress. We also have a new term to consider adding to our AI lexicon, "large reasoning models" or "LRMs." Not sure if that's been widely used already. Paper is here: https://lnkd.in/dzgsdkt7 Credit to people on X for circulating this, as it's making the rounds there.

2

Chatbots were step 1. AI agents are step 10. And they’re already delivering jaw-dropping results across 46+ companies. Let me break it down. Here's what makes them game-changers: 1. Autonomy: They design their own workflows 2. Adaptability: They learn from each interaction 3. Integration: They work seamlessly with existing systems 4. Problem-solving: They tackle complex, multi-step tasks Real-world impact: • 50% increase in first-contact resolution • 30% reduction in operational costs • 90% customer satisfaction improvement The kicker? This is just the beginning. How do you see AI agents transforming your industry? Follow Ghazi for more.

29

23 Comments

If you are trying to make AI killer drones, you don't need to think much about AI Reasoning as long as your drone gets onto the enemy side. AI is like nuclear science, you can use it for good and bad things. If you try to find the place for LLMs in the history of media, it is similar to the Gutenberg Revolution. When the printing press became popular, it led to a wide scale proliferation of information on print media. Written words have been sacred to us. When it was difficult to produce written information as in the old ages, the written texts were revered many orders of magnitude more than an average high sounding AI slop of today. The value of the written word is still very precious to us, though. Someone's handwritten letter is a thing you don't want to lose, for a long time... a text? There are so many... some books written by great authors are going to stand the test of time for many years, and will be revered by many people. But a blog article written by AI? Who cares.

1

Navigating data ingestion complexities can be hard, but it's crucial for maintaining an up-to-date RAG. Our free workshop is a practical way to learn how to build scalable data ingestion pipelines using open-source tools. 👇🏼 Akela Drissner-Schmid will cover essential steps of data ingestion for RAG applications: data collection, preprocessing, and timely updates. This is an excellent opportunity to understand how these elements integrate to form a pipeline. We'll use dlt to: ​1. Extract data from REST APIs ​​2. Load and vectorize into LanceDB ​3. Keep your data up to date with incremental loading We meet on July 8 (Monday) at 4:30 PM CET. Join us using this link: https://lu.ma/cnpdoc5n

118

1 Comment

5 landmark papers in NLP that significantly moved the needle These are 5 of my favorite papers that brought a step function change in natural language understanding. Anyone looking to learn NLP should read these, or at least some articles about them: • 𝗔 𝘂𝗻𝗶𝗳𝗶𝗲𝗱 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 𝗳𝗼𝗿 𝗻𝗮𝘁𝘂𝗿𝗮𝗹 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗽𝗿𝗼𝗰𝗲𝘀𝘀𝗶𝗻𝗴: 𝗗𝗲𝗲𝗽 𝗻𝗲𝘂𝗿𝗮𝗹 𝗻𝗲𝘁𝘄𝗼𝗿𝗸𝘀 𝘄𝗶𝘁𝗵 𝗺𝘂𝗹𝘁𝗶𝘁𝗮𝘀𝗸 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 (https://lnkd.in/gQfjDhsb ) : This paper was the first to show how a single neural network architecture could learn task-specific word embeddings and apply them to multiple NLP tasks, such as part-of-speech tagging, chunking, and named entity recognition. Before this paper, word embeddings existed mostly as a concept but weren’t widely applicable. • 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝗘𝘀𝘁𝗶𝗺𝗮𝘁𝗶𝗼𝗻 𝗼𝗳 𝗪𝗼𝗿𝗱 𝗥𝗲𝗽𝗿𝗲𝘀𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻𝘀 𝗶𝗻 𝗩𝗲𝗰𝘁𝗼𝗿 𝗦𝗽𝗮𝗰𝗲 (https://lnkd.in/gKcs6uyh) : This paper introduced Word2Vec model, which efficiently learns word embeddings by utilizing continuous bag-of-words (CBOW) and skip-gram architectures to capture semantic relationships between words. Word2Vec was the first major model to be scalable and widely used in industrial applications. It sparked a wave of X2Vec models (e.g., Doc2Vec, Tweet2Vec, Graph2vec), where various types of data were transformed into vector representations, further expanding the application of embedding techniques. • 𝗔𝘁𝘁𝗲𝗻𝘁𝗶𝗼𝗻 𝗶𝘀 𝗮𝗹𝗹 𝘆𝗼𝘂 𝗻𝗲𝗲𝗱 (https://lnkd.in/gNwi_Jhu) : Who could forget this paper? :) It introduced the Transformer model, which eliminated the need for recurrence by relying entirely on self-attention mechanisms to model relationships between tokens in a sequence. This dramatically improved performance and efficiency in NLP tasks and laid the foundation for LLM models. • 𝗕𝗘𝗥𝗧: 𝗣𝗿𝗲-𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝗼𝗳 𝗗𝗲𝗲𝗽 𝗕𝗶𝗱𝗶𝗿𝗲𝗰𝘁𝗶𝗼𝗻𝗮𝗹 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗲𝗿𝘀 𝗳𝗼𝗿 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 (https://lnkd.in/gd5HKpW4) : This introduced BERT, the first LLM model based on bidirectional Transformers, pre-trained on vast amounts of text and can be fine-tuned for various NLP tasks setting a new standard for transfer learning. • 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹𝘀 𝗮𝗿𝗲 𝗙𝗲𝘄-𝗦𝗵𝗼𝘁 𝗟𝗲𝗮𝗿𝗻𝗲𝗿𝘀 (https://lnkd.in/gwVU4Xi6) : This paper introduced GPT-3, the largest language model at the time with 175B parameters, capable of performing tasks with minimal fine-tuning or task-specific data. It demonstrated the power of few-shot learning and essentially began the shift away from the need for task-specific trained models. Of course, the list is much longer, and there are many other milestone papers in NLP. Which ones are your favorites? I'd love to hear your thoughts! #MachineLearning #NaturalLanguageProcessing #LLM

111

26 Comments

Food is more than just nourishment; it's deeply intertwined with our health, lifestyle, and culture. In the age of AI, eCommerce search is evolving from a tool into a source of inspiration. It's no longer just about showing you what you're looking for—it also offers delightful surprises you didn't know you wanted. Often, we don't know exactly what to add to our cart. Maybe we have a vague idea, a theme, or an impulse—craving traditional Asian cuisine, trying making sushi at home but needing guidance on the ingredients, planning a fun football night, aiming for a healthier lifestyle with gluten-free meals, or simply wondering, "What's for dinner?" Thanks to LLMs and their vast knowledge of brands, cuisines, recipes, nutrition, cultures, and more, we're making these experiences possible. It takes a dedicated team to seamlessly blend the world knowledge of LLMs with Instacart's insight into our users and product catalog. This involves careful engineering, thoughtful product design, and rigorous A/B testing with real-world traffic. It's been an exciting journey collaborating with such a talented team to supercharge eCommerce search with LLMs.

6

🚨 Big News from Meta Connect! 🚨 Meta just dropped its Llama 3.2 models on Watsonx, including the first multimodal 11B and 90B Vision models! Key Highlights: - Image-in, text-out support for tasks like document understanding and image captioning. - High-res image support (up to 1120x1120) for classification, OCR, and more. - Lightweight Llama 3.2 models (1B & 3B) run on most devices with low latency and enhanced privacy. - New multimodal safety model: Llama Guard. Meta continues leading with efficient, open-source AI models! Groq now support Llama 3 models for even faster AI Performance #AI #Meta #Llama32 #TechInnovation #MultimodalAI #Watsonx #genai

26