SORA Archives - gettectonic.com
Scope of Generative AI

Exploring Generative AI

Like most employees at most companies, I wear a few different hats around Tectonic. Whether I’m building a data model, creating and scheduing an email campaign, standing up a platform generative AI is always at my fingertips. At my very core, I’m a marketer. Have been for so long I do it without eveven thinking. Or at least, everyuthing I do has a hat tip to its future marketing needs. Today I want to share some of the AI content generators I’ve been using, am looking to use, or just heard about. But before we rip into the insight, here’s a primer. Types of AI Content Generators ChatGPT, a powerful AI chatbot, drew significant attention upon its November 2022 release. While the GPT-3 language model behind it had existed for some time, ChatGPT made this technology accessible to nontechnical users, showcasing how AI can generate content. Over two years later, numerous AI content generators have emerged to cater to diverse use cases. This rapid development raises questions about the technology’s impact on work. Schools are grappling with fears of plagiarism, while others are embracing AI. Legal debates about copyright and digital media authenticity continue. President Joe Biden’s October 2023 executive order addressed AI’s risks and opportunities in areas like education, workforce, and consumer privacy, underscoring generative AI’s transformative potential. What is AI-Generated Content? AI-generated content, also known as generative AI, refers to algorithms that automatically create new content across digital media. These algorithms are trained on extensive datasets and require minimal user input to produce novel outputs. For instance, ChatGPT sets a standard for AI-generated content. Based on GPT-4o, it processes text, images, and audio, offering natural language and multimodal capabilities. Many other generative AI tools operate similarly, leveraging large language models (LLMs) and multimodal frameworks to create diverse outputs. What are the Different Types of AI-Generated Content? AI-generated content spans multiple media types: Despite their varied outputs, most generative AI systems are built on advanced LLMs like GPT-4 and Google Gemini. These multimodal models process and generate content across multiple formats, with enhanced capabilities evolving over time. How Generative AI is Used Generative AI applications span industries: These tools often combine outputs from various media for complex, multifaceted projects. AI Content Generators AI content generators exist across various media. Below are good examples organized by gen ai type: Written Content Generators Image Content Generators Music Content Generators Code Content Generators Other AI Content Generators These tools showcase how AI-powered content generation is revolutionizing industries, making content creation faster and more accessible. I do hope you will comment below on your favorites, other AI tools not showcased above, or anything else AI-related that is on your mind. Written by Tectonic’s Marketing Operations Director, Shannan Hearne. Like Related Posts Salesforce OEM AppExchange Expanding its reach beyond CRM, Salesforce.com has launched a new service called AppExchange OEM Edition, aimed at non-CRM service providers. Read more The Salesforce Story In Marc Benioff’s own words How did salesforce.com grow from a start up in a rented apartment into the world’s Read more Salesforce Jigsaw Salesforce.com, a prominent figure in cloud computing, has finalized a deal to acquire Jigsaw, a wiki-style business contact database, for Read more Health Cloud Brings Healthcare Transformation Following swiftly after last week’s successful launch of Financial Services Cloud, Salesforce has announced the second installment in its series Read more

Read More
Salesforce LlamaRank

Salesforce LlamaRank

Document ranking remains a critical challenge in information retrieval and natural language processing. Effective document retrieval and ranking are crucial for enhancing the performance of search engines, question-answering systems, and Retrieval-Augmented Generation (RAG) systems. Traditional ranking models often struggle to balance result precision with computational efficiency, especially when dealing with large datasets and diverse query types. This challenge underscores the growing need for advanced models that can provide accurate, contextually relevant results in real-time from continuous data streams and increasingly complex queries. Salesforce AI Research has introduced a cutting-edge reranker named LlamaRank, designed to significantly enhance document ranking and code search tasks across various datasets. Built on the Llama3-8B-Instruct architecture, LlamaRank integrates advanced linear and calibrated scoring mechanisms, achieving both speed and interpretability. The Salesforce AI Research team developed LlamaRank as a specialized tool for document relevancy ranking. Enhanced by iterative feedback from their dedicated RLHF data annotation team, LlamaRank outperforms many leading APIs in general document ranking and sets a new standard for code search performance. The model’s training data includes high-quality synthesized data from Llama3-70B and Llama3-405B, along with human-labeled annotations, covering a broad range of domains from topic-based search and document QA to code QA. In RAG systems, LlamaRank plays a crucial role. Initially, a query is processed using a less precise but cost-effective method, such as semantic search with embeddings, to generate a list of potential documents. The reranker then refines this list to identify the most relevant documents, ensuring that the language model is fine-tuned with only the most pertinent information, thereby improving accuracy and coherence in the output responses. LlamaRank’s architecture, based on Llama3-8B-Instruct, leverages a diverse training corpus of synthetic and human-labeled data. This extensive dataset enables LlamaRank to excel in various tasks, from general document retrieval to specialized code searches. The model underwent multiple feedback cycles from Salesforce’s data annotation team to achieve optimal accuracy and relevance in its scoring predictions. During inference, LlamaRank predicts token probabilities and calculates a numeric relevance score, facilitating efficient reranking. Demonstrated on several public datasets, LlamaRank has shown impressive performance. For instance, on the SQuAD dataset for question answering, LlamaRank achieved a hit rate of 99.3%. It posted a hit rate of 92.0% on the TriviaQA dataset. In code search benchmarks, LlamaRank recorded a hit rate of 81.8% on the Neural Code Search dataset and 98.6% on the TrailheadQA dataset. These results highlight LlamaRank’s versatility and efficiency across various document types and query scenarios. LlamaRank’s technical specifications further emphasize its advantages. Supporting up to 8,000 tokens per document, it significantly outperforms competitors like Cohere’s reranker. It delivers low-latency performance, ranking 64 documents in under 200 ms with a single H100 GPU, compared to approximately 3.13 seconds on Cohere’s serverless API. Additionally, LlamaRank features linear scoring calibration, offering clear and interpretable relevance scores. While LlamaRank’s size of 8 billion parameters contributes to its high performance, it is approaching the upper limits of reranking model size. Future research may focus on optimizing model size to balance quality and efficiency. Overall, LlamaRank from Salesforce AI Research marks a significant advancement in reranking technology, promising to greatly enhance RAG systems’ effectiveness across a wide range of applications. With its powerful performance, efficiency, and clear scoring, LlamaRank represents a major step forward in document retrieval and search accuracy. The community eagerly anticipates its broader adoption and further development. Like Related Posts Salesforce OEM AppExchange Expanding its reach beyond CRM, Salesforce.com has launched a new service called AppExchange OEM Edition, aimed at non-CRM service providers. Read more The Salesforce Story In Marc Benioff’s own words How did salesforce.com grow from a start up in a rented apartment into the world’s Read more Salesforce Jigsaw Salesforce.com, a prominent figure in cloud computing, has finalized a deal to acquire Jigsaw, a wiki-style business contact database, for Read more Health Cloud Brings Healthcare Transformation Following swiftly after last week’s successful launch of Financial Services Cloud, Salesforce has announced the second installment in its series Read more

Read More
GPT-o1 GPT5 Review

GPT-o1 GPT5 Review

OpenAI has released its latest model, GPT-5, also known as Project Strawberry or GPT-o1, positioning it as a significant advancement in AI with PhD-level reasoning capabilities. This new series, OpenAI-o1, is designed to enhance problem-solving in fields such as science, coding, and mathematics, and the initial results indicate that it lives up to the anticipation. Key Features of OpenAI-o1 Enhanced Reasoning Capabilities Safety and Alignment Targeted Applications Model Variants Access and Availability The o1 models are available to ChatGPT Plus and Team users, with broader access expected soon for ChatGPT Enterprise users. Developers can access the models through the API, although certain features like function calling are still in development. Free access to o1-mini is expected to be provided in the near future. Reinforcement Learning at the Core The o1 models utilize reinforcement learning to improve their reasoning abilities. This approach focuses on training the models to think more effectively, improving their performance with additional time spent on tasks. OpenAI continues to explore how to scale this approach, though details remain limited. Major Milestones The o1 model has achieved impressive results in several competitive benchmarks: Chain of Thought Reasoning OpenAI’s o1 models employ the “Chain of Thought” prompt engineering technique, which allows the model to think through problems step by step. This method helps the model approach complex problems in a structured way, similar to human reasoning. Key aspects include: While the o1 models show immense promise, there are still some limitations, which have been covered in detail elsewhere. However, based on early tests, the model is performing impressively, and users are hopeful that these capabilities are as robust as advertised, rather than overhyped like previous projects such as SORA or SearchGPT by OpenAI. Like Related Posts Salesforce OEM AppExchange Expanding its reach beyond CRM, Salesforce.com has launched a new service called AppExchange OEM Edition, aimed at non-CRM service providers. Read more Salesforce Jigsaw Salesforce.com, a prominent figure in cloud computing, has finalized a deal to acquire Jigsaw, a wiki-style business contact database, for Read more Health Cloud Brings Healthcare Transformation Following swiftly after last week’s successful launch of Financial Services Cloud, Salesforce has announced the second installment in its series Read more Top Ten Reasons Why Tectonic Loves the Cloud The Cloud is Good for Everyone – Why Tectonic loves the cloud You don’t need to worry about tracking licenses. Read more

Read More
Small Language Models

Small Language Models

Large language models (LLMs) like OpenAI’s GPT-4 have gained acclaim for their versatility across various tasks, but they come with significant resource demands. In response, the AI industry is shifting focus towards smaller, task-specific models designed to be more efficient. Microsoft, alongside other tech giants, is investing in these smaller models. Science often involves breaking complex systems down into their simplest forms to understand their behavior. This reductionist approach is now being applied to AI, with the goal of creating smaller models tailored for specific functions. Sébastien Bubeck, Microsoft’s VP of generative AI, highlights this trend: “You have this miraculous object, but what exactly was needed for this miracle to happen; what are the basic ingredients that are necessary?” In recent years, the proliferation of LLMs like ChatGPT, Gemini, and Claude has been remarkable. However, smaller language models (SLMs) are gaining traction as a more resource-efficient alternative. Despite their smaller size, SLMs promise substantial benefits to businesses. Microsoft introduced Phi-1 in June last year, a smaller model aimed at aiding Python coding. This was followed by Phi-2 and Phi-3, which, though larger than Phi-1, are still much smaller than leading LLMs. For comparison, Phi-3-medium has 14 billion parameters, while GPT-4 is estimated to have 1.76 trillion parameters—about 125 times more. Microsoft touts the Phi-3 models as “the most capable and cost-effective small language models available.” Microsoft’s shift towards SLMs reflects a belief that the dominance of a few large models will give way to a more diverse ecosystem of smaller, specialized models. For instance, an SLM designed specifically for analyzing consumer behavior might be more effective for targeted advertising than a broad, general-purpose model trained on the entire internet. SLMs excel in their focused training on specific domains. “The whole fine-tuning process … is highly specialized for specific use-cases,” explains Silvio Savarese, Chief Scientist at Salesforce, another company advancing SLMs. To illustrate, using a specialized screwdriver for a home repair project is more practical than a multifunction tool that’s more expensive and less focused. This trend towards SLMs reflects a broader shift in the AI industry from hype to practical application. As Brian Yamada of VLM notes, “As we move into the operationalization phase of this AI era, small will be the new big.” Smaller, specialized models or combinations of models will address specific needs, saving time and resources. Some voices express concern over the dominance of a few large models, with figures like Jack Dorsey advocating for a diverse marketplace of algorithms. Philippe Krakowski of IPG also worries that relying on the same models might stifle creativity. SLMs offer the advantage of lower costs, both in development and operation. Microsoft’s Bubeck emphasizes that SLMs are “several orders of magnitude cheaper” than larger models. Typically, SLMs operate with around three to four billion parameters, making them feasible for deployment on devices like smartphones. However, smaller models come with trade-offs. Fewer parameters mean reduced capabilities. “You have to find the right balance between the intelligence that you need versus the cost,” Bubeck acknowledges. Salesforce’s Savarese views SLMs as a step towards a new form of AI, characterized by “agents” capable of performing specific tasks and executing plans autonomously. This vision of AI agents goes beyond today’s chatbots, which can generate travel itineraries but not take action on your behalf. Salesforce recently introduced a 1 billion-parameter SLM that reportedly outperforms some LLMs on targeted tasks. Salesforce CEO Mark Benioff celebrated this advancement, proclaiming, “On-device agentic AI is here!” Like Related Posts Salesforce OEM AppExchange Expanding its reach beyond CRM, Salesforce.com has launched a new service called AppExchange OEM Edition, aimed at non-CRM service providers. Read more The Salesforce Story In Marc Benioff’s own words How did salesforce.com grow from a start up in a rented apartment into the world’s Read more Salesforce Jigsaw Salesforce.com, a prominent figure in cloud computing, has finalized a deal to acquire Jigsaw, a wiki-style business contact database, for Read more Health Cloud Brings Healthcare Transformation Following swiftly after last week’s successful launch of Financial Services Cloud, Salesforce has announced the second installment in its series Read more

Read More

Train Your Own SORA Model

Unveiling the Vision Transformer: A Leap in Video Generation The closest open-source model to SORA is Latte, which uses the same Vision Transformer architecture. So, what makes the Vision Transformer so outstanding, and how does it differ from previous methods? You can Train Your Own SORA Model. Latte hasn’t open-sourced its text-to-video training code. We’ve replicated this code from the paper and made it available for anyone to use in training their own SORA alternative model. Let’s discuss how effective our training was. From 3D U-Net to Vision Transformer Image generation has advanced significantly, with the U-Net model structure being the most commonly used: If you’re confused about the network structures, remember the key principle of deep learning: “Just Add More Layers!” Vision Transformer: A Game Changer In 3D U-Net, the transformer can only function within the U-Net, limiting its view. The Vision Transformer, however, enables transformers to globally manage video generation. Training Your Open-Source SORA Alternative with Latte Latte uses the video slicing sequence and Vision Transformer method discussed. While Latte hasn’t open-sourced its text-to-video model training code, we’ve replicated it here: GitHub Repo. Training involves three steps: For more details, see the GitHub repo. They’ve also made improvements to the training process: Model Performance The official Latte video shows impressive performance, especially in handling significant motion. However, our own tests indicate that while Latte performs well, it isn’t the top-performing model. Other open-source models have shown better performance. We will continue to share information on models with better performance, so stay tuned to Tectonic’s Insights. Hardware Requirements Due to its large scale, training Latte requires an A100 or H100 with 80GB of memory. Like Related Posts Salesforce OEM AppExchange Expanding its reach beyond CRM, Salesforce.com has launched a new service called AppExchange OEM Edition, aimed at non-CRM service providers. Read more The Salesforce Story In Marc Benioff’s own words How did salesforce.com grow from a start up in a rented apartment into the world’s Read more Salesforce Jigsaw Salesforce.com, a prominent figure in cloud computing, has finalized a deal to acquire Jigsaw, a wiki-style business contact database, for Read more Health Cloud Brings Healthcare Transformation Following swiftly after last week’s successful launch of Financial Services Cloud, Salesforce has announced the second installment in its series Read more

Read More
How Good is Our Data

How Good is Our Data?

Generative AI promises to significantly reshape how you manage your customer relationships, but it requires data that is accurate, updated, accessible, and complete. Why is this important? You may do something differently this quarter than you did last quarter, based on the latest data. But if your data is outdated or incorrect, that’s what the AI will use.  Generative AI focuses on creating new and original content, chat responses, designs, synthetic content or even deepfakes. It’s particularly valuable in creative fields and for novel problem-solving, as it can autonomously generate many types of new outputs. Generative Artificial Intelligence models often present inaccurate information as though it were correct. This is often caused by limited information in the system, biases in training data, and issues with the algorithm. These are commonly called ‘hallucinations‘ and they present a huge problem. When training your models for generative AI, you should first ensure high information excellence from top to bottom. To get your information house in order, remove duplicates, outliers, errors, and other things that can negatively affect how you make decisions. Then connect your data sources — marketing, sales, service, commerce – into a single record, updated in real time, so the AI can make the best recommendations.   McKinsey recently wrote, “Companies that have not yet found ways to harmonize and provide ready access to their information will be unable to unlock much of generative AI’s potentially transformative power.” Why is data important in generative AI? Aside from the cost factor, poor information quality can introduce unnecessary and harmful noise into the generative AI systems and models, leading to misleading answers, nonsensical output, or overall lower efficacy. What is high-quality data for AI? High-quality information is essential for AI systems to deliver meaningful results. Data quality possesses several key attributes: Accuracy: High-quality information is free from errors and inaccuracies. Inaccurate information can mislead AI models and produce unreliable outputs. Is AI 100 percent accurate? Because AI will still rely on your data for decision making and accuracy depends on the quality of your information. AI machines must be well-programmed to make sure the machine is making decisions based on the correct, available information. Also, privacy and security of the data are paramount. AI machines need to access information that is encrypted and secure. Understand that Generative AI is most effective at creating new data based on existing patterns and examples, with a focus on text and image data. Generative AI is most suitable for generating new data based on existing patterns and examples. It doesn’t actually think for itself. Yet. Known Limitations Of Generative AI Large language models (LLMs) are prone to “hallucinations” – generating fictitious information, presented as factual or accurate. This can include citations, publications, biographical information, and other information commonly used in research and academic papers. Like1 Related Posts Salesforce OEM AppExchange Expanding its reach beyond CRM, Salesforce.com has launched a new service called AppExchange OEM Edition, aimed at non-CRM service providers. Read more The Salesforce Story In Marc Benioff’s own words How did salesforce.com grow from a start up in a rented apartment into the world’s Read more Salesforce Jigsaw Salesforce.com, a prominent figure in cloud computing, has finalized a deal to acquire Jigsaw, a wiki-style business contact database, for Read more Health Cloud Brings Healthcare Transformation Following swiftly after last week’s successful launch of Financial Services Cloud, Salesforce has announced the second installment in its series Read more

Read More
Salesforce Document Generation

Generative AI

Artificial Intelligence in Focus Generative Artificial Intelligence is a type of artificial intelligence technology that can produce various types of content, including text, imagery, audio and synthetic data. What is the difference between generative AI and general AI? Traditional AI focuses on analyzing historical data and making future numeric predictions, while generative AI allows computers to produce brand-new outputs that are often indistinguishable from human-generated content. Recently, there has been a surge in discussions about artificial intelligence (AI), and the spotlight on this technology seems more intense than ever. Despite AI not being a novel concept, as many businesses and institutions have incorporated it in various capacities over the years, the heightened interest can be attributed to a specific AI-powered chatbot called ChatGPT. ChatGPT stands out by being able to respond to plain-language questions or requests in a manner that closely resembles human-written responses. Its public release allowed people to engage in conversations with a computer, creating a surprising, eerie, and evocative experience that captured widespread attention. This ability of an AI to engage in natural, human-like conversations represents a notable departure from previous AI capabilities. The Artificial Intelligence Fundamentals badge on the Salesforce Trailhead delves into the various specific tasks that AI models are trained to execute, highlighting the remarkable potential of generative AI, particularly in its ability to create diverse forms of text, images, and sounds, leading to transformative impacts both in and outside the workplace. Let’s explore the tasks that generative AI models are trained to perform, the underlying technology, and how businesses are specializing within the generative AI ecosystem. It also delves into concerns that businesses may harbor regarding generative Artificial Intelligence. Exploring the Capabilities of Language Models While generative AI may appear as a recent phenomenon, researchers have been developing and training generative AI models for decades. Some notable instances made headlines, such as Nvidia unveiling an AI model in 2018 capable of generating photorealistic images of human faces. These instances marked the gradual entry of generative AI into public awareness. While some researchers focused on AI’s capabilities generating specific types of images, others concentrated on language-related AI. This involved training AI models to perform various tasks related to interpreting text, a field known as natural language processing (NLP). Large language models (LLMs), trained on extensive datasets of real-world text, emerged as a key component of NLP, capturing intricate language rules that humans take years to learn. Summarization, translation, error correction, question answering, guided image generation, and text-to-speech are among the impressive tasks accomplished by LLMs. They provide a tool that significantly enhances language-related tasks in real-world scenarios. Predictive Nature of Generative AI Despite the remarkable predictions generated by generative AI in the form of text, images, and sounds, it’s crucial to clarify that these outputs represent a form of prediction rather than a manifestation of “thinking” by the computer. Generative Artificial Intelligence doesn’t possess opinions, intentions, or desires; it excels at predicting sequences of words based on patterns learned during training. Understanding this predictive nature is key. The AI’s ability to predict responses aligns with expectations rather than reflecting any inherent understanding or preference. Recognizing the predictive character of generative AI underscores its role as a powerful tool, bridging gaps in language-related tasks for both professional and recreational purposes. Like Related Posts Salesforce OEM AppExchange Expanding its reach beyond CRM, Salesforce.com has launched a new service called AppExchange OEM Edition, aimed at non-CRM service providers. Read more The Salesforce Story In Marc Benioff’s own words How did salesforce.com grow from a start up in a rented apartment into the world’s Read more Salesforce Jigsaw Salesforce.com, a prominent figure in cloud computing, has finalized a deal to acquire Jigsaw, a wiki-style business contact database, for Read more Health Cloud Brings Healthcare Transformation Following swiftly after last week’s successful launch of Financial Services Cloud, Salesforce has announced the second installment in its series Read more

Read More
tectonic logo

AI Large Language Models

What Exactly Constitutes a Large Language Model? Picture having an exceptionally intelligent digital assistant that extensively combs through text, encompassing books, articles, websites, and various written content up to the year 2021. Yet, unlike a library that houses entire books, this digital assistant processes patterns from the textual data it undergoes. This digital assistant, akin to a large language model (LLM), represents an advanced computer model tailored to comprehend and generate text with humanlike qualities. Its training involves exposure to vast amounts of text data, allowing it to discern patterns, language structures, and relationships between words and sentences. How Do These Large Language Models Operate? Fundamentally, large language models, exemplified by GPT-3, undertake predictions on a token-by-token basis, sequentially building a coherent sequence. Given a request, they strive to predict the subsequent token, utilizing their acquired knowledge of patterns during training. These models showcase remarkable pattern recognition, generating contextually relevant content across diverse topics. The “large” aspect of these models refers to their extensive size and complexity, necessitating substantial computational resources like powerful servers equipped with multiple processors and ample memory. This capability enables the model to manage and process vast datasets, enhancing its proficiency in comprehending and generating high-quality text. While the sizes of LLMs may vary, they typically house billions of parameters—variables learned during the training process, embodying the knowledge extracted from the data. The greater the number of parameters, the more adept the model becomes at capturing intricate patterns. For instance, GPT-3 boasts around 175 billion parameters, marking a significant advancement in language processing capabilities, while GPT-4 is purported to exceed 1 trillion parameters. While these numerical feats are impressive, the challenges associated with these mammoth models include resource-intensive training, environmental implications, potential biases, and more. Large language models serve as virtual assistants with profound knowledge, aiding in a spectrum of language-related tasks. They contribute to writing, offer information, provide creative suggestions, and engage in conversations, aiming to make human-technology interactions more natural. However, users should be cognizant of their limitations and regard them as tools rather than infallible sources of truth. What Constitutes the Training of Large Language Models? Training a large language model is analogous to instructing a robot in comprehending and utilizing human language. The process involves: Fine-Tuning: A Closer Look Fine-tuning involves further training a pre-trained model on a more specific and compact dataset than the original. It is akin to training a robot proficient in various cuisines to specialize in Italian dishes using a dedicated cookbook. The significance of fine-tuning lies in: Versioning and Progression Large language models evolve through versions, with changes in size, training data, or parameters. Each iteration aims to address weaknesses, handle a broader task spectrum, or minimize biases and errors. The progression is simplified as follows: In essence, large language model versions emulate successive editions of a book series, each release striving for refinement, expansiveness, and captivating capabilities. Like Related Posts Salesforce OEM AppExchange Expanding its reach beyond CRM, Salesforce.com has launched a new service called AppExchange OEM Edition, aimed at non-CRM service providers. Read more The Salesforce Story In Marc Benioff’s own words How did salesforce.com grow from a start up in a rented apartment into the world’s Read more Salesforce Jigsaw Salesforce.com, a prominent figure in cloud computing, has finalized a deal to acquire Jigsaw, a wiki-style business contact database, for Read more Health Cloud Brings Healthcare Transformation Following swiftly after last week’s successful launch of Financial Services Cloud, Salesforce has announced the second installment in its series Read more

Read More
gettectonic.com