Large Language Models

microsoftMar 3, 20252 min readCurated summary

How Microsoft Engineers Build AI: Learn about scalable RAG-enabled AI Apps

Microsoft’s new *How Microsoft Engineers Build AI* video series explains how its teams develop AI applications at scale. The first episode focuses on retrieval-augmented generation (RAG), using Copilot for Azure’s Ask Learn plugin as a practical example. It shows how RAG can combine proprietary data with large language models to deliver accurate, contextually relevant answers. ## Building AI Applications with RAG - RAG is presented as a practical way to improve AI applications without relying solely on model fine-tuning. - It retrieves relevant information from a knowledge base and provides that context to an LLM when generating responses. - The approach is useful for applications that need current, domain-specific, or proprietary information. ## The Ask Learn Plugin - Microsoft engineers explain how they built the Ask Learn RAG plugin for Copilot for Azure. - The plugin helps Azure developers find answers quickly within their existing workflow. - The project involved product managers and engineering leaders sharing development challenges, design decisions, and best practices. ## Challenges in Developing Reliable RAG - Selecting the right source content is essential for producing useful answers. - Data must be preprocessed effectively before it can be retrieved. - RAG systems require careful performance evaluation to measure accuracy and relevance. - Keeping responses accurate and up to date requires ongoing improvements to content and retrieval methods. ## Broader Microsoft Applications - The episode discusses RAG implementations across: - Copilot in Azure - Microsoft Security Copilot - Dynamics 365 Business Central - These examples demonstrate how RAG can support different products and business scenarios. The episode is intended as a practical introduction for developers building RAG-based applications, covering prototyping, data management, evaluation, and common pitfalls. Developers can explore the series alongside Microsoft Learn resources and Azure AI development tools such as Visual Studio and GitHub Copilot.

Read original

Large Language Models AI Machine Learning Github Copilot+4

figmaFeb 19, 20252 min readCurated summary

Double Click: When Coding Becomes Conversation | Figma Blog

Vibe coding replaces much of traditional programming with an ongoing conversation with AI: users describe an idea, review the result, and iterate through prompts. It lowers the barrier to creating software and makes experimentation faster, especially for prototypes and side projects. However, the approach can become unreliable as projects grow, producing tangled code and weak internal architecture. ## From Code to Conversation - Andrej Karpathy coined “vibe coding” to describe building software by talking to AI tools such as Cursor Composer and using voice input. - The process emphasizes seeing results, describing changes, running the project, and copying or pasting outputs rather than understanding every line of code. - The idea reflects a broader history of abstraction, from punch cards to assembly, C, Python, and now AI-assisted development. ## Faster, More Accessible Prototyping - Vibe coding lets people express interactive ideas without mastering syntax or a programming language. - Charmaine Lee of Val Town compares it to casually writing in a document or creating a spreadsheet. - Figma designer Nikolas Klein argues that the main benefit is shortening the gap between imagining an interaction and seeing it work. - Replit CEO Amjad Masad reported that 75% of Replit customers never write a line of code. - Figma engineer Vincent van der Meulen used AI to create projects, including a running coach and a loading animation, despite lacking SwiftUI expertise. ## The Complexity Ceiling - Vibe coding is most effective at the beginning of a project, when requirements are simple and experimentation matters more than structure. - As complexity increases, AI-generated solutions may stop fitting together coherently. - Developers can reach a “valley of despair”: an initial burst of progress gives way to difficult debugging and maintenance. - Vincent described ending up with “spaghetti code” and no consistent internal data model after reaching roughly 80% of his goal. Vibe coding is best treated as a powerful prototyping and exploration technique, not a replacement for engineering judgment. Teams should still inspect, test, refactor, and architect AI-generated code when projects become complex or production-critical.

Read original

Large Language Models Prototyping Vibe Coding Swiftui+2

coupangNov 14, 2024Original article

Accelerating Coupang’s AI Journey with LLMs (opens in new tab)

Coupang is strategically evolving its machine learning infrastructure to integrate Large Language Models (LLMs) and foundation models across its e-commerce ecosystem. By transitioning from task-specific deep learning models to multi-modal transformers, the company aims to enhance customer experiences in search, recommendations, and logistics. This shift necessitates a robust ML platform capable of handling the massive compute, networking, and latency demands inherent in generative AI. ### Core Machine Learning Domains Coupang’s existing ML ecosystem is built upon three primary pillars that drive business logic: * **Recommendation Systems:** These models leverage vast datasets of user interactions—including clicks, purchases, and relevance judgments—to power home feeds, search results, and advertising. * **Content Understanding:** Utilizing deep learning to process product catalogs, user reviews, and merchant data to create unified representations of customers and products. * **Forecasting Models:** Predictive algorithms manage over 100 fulfillment centers, optimizing pricing and logistics for millions of products through a mix of statistical methods and deep learning. ### Enhancing Multimodal and Language Understanding The adoption of Foundation Models (FM) has unified previously fragmented ML tasks, particularly in multilingual environments: * **Joint Modeling:** Instead of separate embeddings, vision and language transformer models jointly model product images and metadata (titles/descriptions) to improve ad retrieval and similarity searches. * **Cross-Border Localization:** LLMs facilitate the translation of product titles from Korean to Mandarin and improve the quality of shopping feeds for global sellers. * **Weak Label Generation:** To overcome the high cost of human labeling in multiple languages, Coupang uses LLMs to generate high-quality "weak labels" for training downstream models, addressing label scarcity in under-resourced segments. ### Infrastructure for Large-Scale Training Scaling LLM training requires a shift in hardware architecture and distributed computing strategies: * **High-Performance Clusters:** The platform utilizes H100 and A100 GPU clusters interconnected with high-speed InfiniBand or RoCE (RDMA over Converged Ethernet) networking to minimize communication bottlenecks. * **Distributed Frameworks:** To fit massive models into GPU memory, Coupang employs various parallelism techniques, including Fully Sharded Data Parallelism (FSDP), Tensor Parallelism (TP), and Pipeline Parallelism (PP). * **Efficient Categorization:** Traditional architectures that required a separate model for every product category are being replaced by a single, massive multi-modal transformer capable of handling categorization and attribute extraction across the entire catalog. ### Optimizing LLM Serving and Inference The transition to real-time generative AI features requires significant optimizations to manage the high computational cost of inference: * **Quantization Strategies:** To reduce memory footprint and increase throughput, models are compressed using FP8, INT8, or INT4 precision without significant loss in accuracy. * **Advanced Serving Techniques:** The platform implements Key-Value (KV) caching to avoid redundant computations during text generation and utilizes continuous batching (via engines like vLLM or TGI) to maximize GPU utilization. * **Lifecycle Management:** A unified platform vision ensures that the entire end-to-end lifecycle—from data preparation and fine-tuning to deployment—is streamlined for ML engineers. To stay competitive, Coupang is moving toward an integrated AI lifecycle where foundation models serve as the backbone for both content generation and predictive analytics. This infrastructure-first approach allows for the rapid deployment of generative features while maintaining the resource efficiency required for massive e-commerce scales.

Large Language Models AI Machine Learning Generative AI+5

figmaOct 24, 20242 min readCurated summary

How We Built AI-Powered Search in Figma | Figma Blog

Figma’s AI search emerged from a practical problem: designers often struggled to find existing work, sometimes relying on Slack to locate files from screenshots or vague descriptions. The team initially pursued design autocomplete, but research showed that designers more often reuse and adapt prior work than create from scratch. This led Figma toward visual and semantic search, launched in 2024, using AI to help users find and reuse relevant designs. ## The Problem of Finding Existing Designs - Designers frequently knew what they wanted visually but not where the source file was located. - At Figma, hundreds of Slack messages showed designers asking colleagues to identify files from screenshots or descriptions. - Traditional keyword search was insufficient when users did not know a component’s exact name or file location. ## From Design Autocomplete to AI Search - Figma began with a three-day AI hackathon in June 2023. - One prototype, design autocomplete, suggested likely next components—for example, a “Get started” button in an onboarding flow. - The team believed AI should handle repetitive tasks so designers could focus on higher-level thinking and user needs. - Internal testing and user interviews revealed that designers commonly revisit old explorations, reuse existing work, and build on prior designs rather than starting from nothing. ## Using Search to Improve AI Suggestions - Figma built search infrastructure alongside autocomplete. - Retrieval-Augmented Generation (RAG) could improve AI responses by supplying relevant examples from existing designs. - Finding designs similar to the user’s current work could make autocomplete recommendations more useful and context-aware. - As testing continued, the team recognized that locating and reusing existing work was a more fundamental need than predicting the next component. ## Visual and Semantic Search - **Visual search** allows users to search with: - A screenshot - A selected frame - A quick sketch - **Semantic search** interprets the meaning and context of text-based queries, even when users do not know the precise component name or description. - Search results can help users discover designs and components and then open, preview, or insert them into their projects. Figma’s experience shows that successful AI features often emerge through iteration rather than from an initial prototype. The practical recommendation is to start with real user behavior, test ambitious ideas, and use AI where it removes friction—in this case, helping designers quickly find and reuse relevant existing work.

Read original

Large Language Models RAG Semantic Search Ai Search+2

datadogSep 23, 20243 min readCurated summary

How we optimized LLM use for cost, quality, and safety to facilitate writing postmortems

Bits AI uses large language models to generate first drafts of incident postmortems from Datadog Incident Management metadata and Slack discussions. The goal is to reduce documentation effort without replacing the human-led reflection that makes postmortems valuable. Datadog found that production-quality LLM features require extensive prompt experimentation, safeguards against hallucinations, careful model selection, and strong privacy and feedback mechanisms. ## Why LLMs Are Difficult to Use for Postmortems - Postmortems must remain factual, consistent, and useful, unlike more creative LLM applications. - LLMs can ignore instructions, produce inconsistent formatting, repeat information, or hallucinate plausible but false details. - More than 100 hours were spent refining section structures and instructions so they worked across varied incidents. - Building these systems requires a combination of software engineering, product management, data science, and technical writing. ## Balancing Structured and Unstructured Incident Data - Structured Datadog data includes incident metadata, alerts, graphs, and manually entered fields such as customer impact. - Slack discussions provide current, detailed, and unstructured context from the incident response process. - Manually maintained fields are often more reliable but can become outdated, while LLMs can continually reassess live discussions. - Combining both sources helped the model understand the full incident and prioritize information appropriately. ## Cost, Speed, and Quality Trade-offs - Datadog compared models including GPT-3.5 and GPT-4. - Results varied by as much as 50 times in cost and speed. - GPT-4 generally produced more accurate output but was substantially slower and more expensive. - Model choice therefore depended on the required balance between quality, latency, and operating cost. ## Preserving Human Control and Privacy - AI-generated drafts were intended to support authors, not become final documents automatically. - Keeping engineers involved preserves the discovery and learning process that postmortems are meant to encourage. - Because incidents may contain sensitive information or secrets, Datadog scrubbed data before sending it to an LLM. - The system was designed to ensure authors remained responsible for reviewing and improving the draft. ## Experimentation and Reliability Safeguards - A custom API extracted and structured incident data from Datadog, Slack, and other sources. - The framework enabled rapid testing of different datasets, models, input configurations, and output token limits. - Datadog refined prompts and instructions to guide the model toward useful recommendations while retaining author agency. - Lowering model temperature and tuning other parameters reduced irrelevant or fabricated content. - A related feature generated concise incident summaries for responders joining Slack channels, providing a faster testing environment for LLM-based incident analysis. ## Evaluating Generated Drafts - Datadog used both qualitative and quantitative evaluation methods. - Authors of earlier postmortems compared their human-written documents with AI-generated drafts for the same incidents. - Evaluation included criteria such as factual accuracy and conciseness, with the provided article ending before the full evaluation approach was described. - Continuous feedback was used to refine prompts, models, and the overall drafting workflow. Overall, reliable AI-assisted postmortems require more than selecting a capable model. The strongest approach combines carefully prepared incident data, privacy controls, model and prompt experimentation, systematic evaluation, and mandatory human review.

Read original

Large Language Models Datadog Slack Data Science+4

figmaJun 25, 20242 min readCurated summary

What is Good Design in the Age of AI? | Figma Blog

AI is making product creation faster and more accessible, increasing the importance of design as a differentiator. Figma argues that designers should not chase automation or trends, but apply enduring principles such as empathy, creativity, and solving real user needs. The future of good design will depend on experimentation, stronger design-to-code connections, pragmatism, and new forms of human–AI collaboration. ## AI Creates a New Design Inflection Point - Like the iPhone’s launch in 2007, AI is introducing a new medium that requires experimentation and the development of new interaction patterns. - Early mobile products often forced desktop experiences onto smaller screens; today, many AI products similarly rely on basic chatbots and templates. - Designers can unlock AI’s potential through: - Richer interactions - Intuitive gestures - Patterns designed specifically for AI - AI can generate code, designs, and complete applications from prompts, allowing teams to move rapidly from concept to creation. - As more people participate in product development, thoughtful design becomes increasingly important for products to stand out. ## Codifying the Fundamentals of Good Design - Figma’s AI feature for generating initial UI drafts needed to understand the mechanics of good design. - Because large language models are text-oriented, generating high-quality visual interfaces is more difficult than generating text or code. - A complete rulebook for design is impractical: - Good design contains too many contextual details to define exhaustively. - Extremely large prompts exceed technical token limits. - Figma instead focused on reducing design expertise to a small set of broadly applicable principles. - Teaching AI requires designers to make their intuitive knowledge explicit by creating rules that are: - Clear - Concrete - Practical - General enough to apply across many interfaces - The process resembles teaching design: instructors must break complex judgment into principles that others can understand and use. ## Design’s Continuing Role - AI should elevate design rather than simply automate it. - The most valuable design foundations remain relatively constant despite technological change. - Designers’ roles may evolve, but their understanding of users, creativity, and ability to solve meaningful problems remain essential. ## Practical Direction Teams should treat AI as a new design medium, not merely an automation tool. They should experiment with AI-native interaction patterns while grounding products in concise, teachable design principles and genuine user needs.

Read original

Large Language Models Figma AI Design Systems+2

figmaJun 25, 20243 min readCurated summary

Are we finally entering the age of androids? | Figma Blog

Humanoid robots are moving from science fiction into public spaces and workplaces, forcing people to confront both the promise and risks of embodied AI. Their humanlike form makes technology more intuitive and emotionally engaging, but it also encourages people to project intelligence, intention, and personality onto machines. The article argues that designers must shape this illusion carefully, using humanoid robots to foster connection and understanding rather than control or deception. ## Humanoids as Technology in Human Form - Ameca, created by Engineered Arts, performs at Las Vegas’s Sphere as an interactive entertainer. - It turns toward speakers, displays facial expressions, tells jokes, and responds conversationally. - Its appeal comes from combining AI with expressive robotics. - Apollo, developed by Apptronik and argodesign, represents a different model: - It is designed as a general-purpose laborer. - Its flat face, cameras, and LED mouth prioritize function over lifelike appearance. - Humanoid robots have deep cultural roots, appearing in Greek mythology, Taoist philosophy, and science fiction. - Their human form makes software easier to engage with through gestures, expressions, and face-to-face interaction—what Engineered Arts CEO Will Jackson describes as a heads-up alternative to screen-based technology and virtual reality. ## What the Illusion of Sentience Unlocks - Humans naturally anthropomorphize objects and search for faces, motives, and signs of life. - Madeline Gannon uses body language and animal behavior as inspiration for designing industrial robots with recognizable personalities. - Even simple geometric animations can appear intentional: the Heider and Simmel experiment showed that people assign motives to moving shapes. - Ameca intensifies this effect through: - Furrowed brows, smiles, and expressions of surprise - Celebrity impressions - Custom personalities created by a dedicated “Persona Architect” - The ability to switch behavioral modes depending on context - Engineered Arts deliberately avoids making Ameca appear fully human: - Its metallic body avoids realistic skin. - It has no defined race or gender. - Its artificiality makes the theatrical nature of the interaction more visible. - The technology underneath remains impersonal: AI interprets language and maps it to suitable facial expressions. Nevertheless, users can experience meaningful emotional moments, such as a shy attendee gaining confidence while speaking Japanese with Ameca. - Gannon argues that designers must make complex systems legible, much as everyday objects communicate how they should be used. - She sees design as a way to redirect technology toward curiosity, kindness, and care, creating relationships based on connection rather than control. ## Designing Androids for Work - Apollo does not claim to be sentient; its purpose is practical labor. - It is being developed to address worker shortages, including work in Mercedes-Benz factories and potentially space exploration. - Its humanoid shape is primarily a response to human-scale environments and tools. - Unlike Ameca, Apollo must appear approachable enough for workers to accept, while avoiding the uncanny valley. - The article frames this as a central design challenge: workplace robots need to communicate socially without misleading people about what they are. Humanoid robots are most valuable when their appearance and behavior clarify how people should interact with them. Whether used for entertainment or labor, designers should treat the illusion of intelligence as an ethical material—balancing emotional engagement with transparency and purposeful design.

Read original

Large Language Models AI Automation Human Computer Interaction+3

figmaJun 25, 20242 min readCurated summary

Welcome to The Prompt | Figma Blog

AI may transform design and building, but its ultimate impact remains unsettled. Figma’s *The Prompt* explores that uncertainty through essays and interviews with experts across design, engineering, product development, and the built environment. The collection argues that human judgment—especially the ability to ask thoughtful, well-framed questions—will remain central to making AI useful. ## Prompting as a Creative Discipline - Prompt engineering is described as the practice of getting better answers by asking better questions. - Like interviewing or editing a magazine, effective prompting requires: - Clear context - Thoughtful framing - Useful guidance - AI’s capabilities are treated as largely inert without human direction; people must coax useful results from the technology. - The act of questioning is presented as a fundamentally human and creative instinct. ## The Purpose of *The Prompt* - Created by Figma’s Story Studio and Brand Studio, the magazine launched at Config 2024. - It combines writing, interviews, and illustrations to examine how AI is changing creative and technical work. - Contributors come from both inside and outside Figma and work across: - Design - Engineering - Product development - Robotics - Manufacturing - Residential housing ## Questions About AI’s Future The magazine uses a range of prompts to investigate both immediate applications and larger societal questions, including: - What constitutes good design when AI can generate and automate more work? - Whether code becoming a commodity should be feared - How much data is actually necessary - How starting with imperfect or incomplete ideas can shape innovation - Whether AI development can move beyond technological echo chambers - If efficiency undermines creativity - How to build AI features that people both want and trust - The relationship between artificial design intelligence (ADI) and artificial general intelligence (AGI) - Whether automation can unlock the full potential of design systems - The role of robots in construction and housing - Whether society is entering an age of androids ## Practical and Long-Term Perspectives - Contributors examine ambitious challenges, such as applying AI to manufacturing and housing. - They also focus on what AI can deliver reliably today rather than only speculating about distant possibilities. - The goal is to make complex systems more understandable and usable while learning how to guide AI more effectively. Figma presents *The Prompt* as both a magazine and an experiment in inquiry: meaningful progress with AI depends not just on increasingly capable systems, but on humans asking clearer, more imaginative, and more responsible questions.

Read original

Large Language Models AI Design Systems Automation+4

figmaJun 25, 20242 min readCurated summary

Config 2024 In Review | Figma Blog

Figma’s Config 2024 announcements focus on making design more exploratory, efficient, and connected to development. The company introduced Figma AI, a redesigned UI3 interface, Figma Slides, major Dev Mode improvements, and quality-of-life updates. Figma argues that as AI makes software creation easier, thoughtful design will become an even stronger differentiator. ## Figma AI: Faster exploration and production - **Visual Search** lets users find similar designs across accessible team files using a screenshot, selected frame, image, or sketch. - Improved **Asset Search** understands the context of queries, even when search terms do not match asset names. - New AI-powered efficiency features can: - Generate realistic images and copy - Rewrite, translate, or vary text - Automatically create prototype connections - Rename layers - **Make Designs**, available through the Actions panel, generates initial UI layouts and component options from text prompts. - Figma says these tools are built around practical user needs rather than AI hype, using large language models to reduce tedious work and help designers explore more possibilities. ## Other major Config announcements - **UI3** redesigns the Figma interface. - **Figma Slides** introduces a dedicated environment for building, collaborating on, and presenting presentations. - **Dev Mode updates** aim to move teams from designs being merely “design ready” to being fully “dev complete.” - Additional improvements target **Auto Layout, UI kits, and the prototype viewer**. ## Availability - Figma AI and UI3 were announced as limited betas with gradual rollout. - Users can join the waitlist through Figma’s help menu by selecting **“Join UI3 + AI waitlist.”** Figma’s overall direction is to support the full path from idea generation through design, presentation, and development, while using AI to automate routine tasks and expand creative exploration.

Read original

Large Language Models AI Design Systems Prototyping+3

figmaJun 25, 20243 min readCurated summary

What Would You Ask If No One Could Judge You? | Figma Blog

Perplexity’s founders envision it as an “answer engine” that turns web-scale information into concise, sourced explanations rather than lists of links. The product grew from a personal need for judgment-free learning and was shaped by the shortcomings of early conversational AI, especially outdated knowledge and hallucinations. Its broader goal is to make curiosity easier to express and pursue. ## Building a Judgment-Free Knowledge Tool - Aravind Srinivas was inspired by childhood “Wikipedia rabbit holes” and the evolution from printed encyclopedias to AI-powered knowledge tools. - Perplexity aims to make learning engaging through curiosity rather than attention-grabbing entertainment. - The company wants users to ask anything without worrying about appearing uninformed or being judged. ## From Private Slackbot to Public Product - The founders initially built a Slackbot to answer practical questions about fundraising, employee health insurance, and running a company. - They hesitated to launch because they feared criticism for attempting to compete with Google. - Investor Nat Friedman encouraged them to view the effort as an asymmetric bet: little downside, but potentially enormous upside. - Perplexity launched shortly after ChatGPT, despite the founders having no previous company-building experience. ## An Answer Engine with Sources - ChatGPT highlighted problems with knowledge cutoffs, hallucinations, and unsupported answers. - Perplexity responded by combining: - Natural-language interaction - Web search and indexing - Large language models - Inline sources and footnotes - Its goal is to provide a direct answer while allowing users to verify the underlying information. - Srinivas describes the product as a combination of Wikipedia and conversational chat, with information drawn from across the internet. ## Making Complex Information Approachable - Perplexity follows an 80/20 approach: identify the most important concepts and deliver most of the useful understanding quickly. - It synthesizes information from multiple web pages into a concise explanation instead of requiring users to read extensively. - The product aims to simplify information without reducing it to misleading or overly shallow conclusions. ## Turning Answers into Further Curiosity - Each response includes three related follow-up questions to encourage exploration. - Srinivas argues that people are naturally curious but often lack the confidence, vocabulary, or precision to formulate good questions. - Perplexity’s design assumes that the user is never wrong; the system should help clarify and develop a person’s curiosity rather than blame them for asking imperfectly. Perplexity’s central recommendation is implicit in its design: make knowledge easier to access, verify, and explore, while removing the social fear that prevents people from asking questions in the first place.

Read original

Large Language Models AI Natural Language Processing Conversational Ai+2

figmaJun 17, 20243 min readCurated summary

Navigating the Promise and Pitfalls of AI | Figma blog | Figma Blog

AI’s promise is real, but useful AI products will emerge through experimentation rather than a race to ship features. Figma’s research suggests AI is already transforming individual workflows—especially for developers—while having a smaller effect on collaboration and foundational design work. To deliver lasting value, teams must focus on how AI reshapes products, industries, and group work, not just on model capabilities. ## Research and Methodology - Figma surveyed more than 1,800 designers, executives, and developers. - Participants came from the US, Canada, Australia, the UK, Japan, France, and Germany. - The survey ran from February 26 to March 3, 2024. - The report combines survey findings with discussions involving AI and design experts. - Its central premise is that AI’s impact depends heavily on product design and user experience—not only on the power of large language models. ## AI’s Uneven Transformation of Workflows - Developers were 60% more likely than designers to say AI had transformed the products they work on. - Developers use AI for daily tasks such as generating starting points and translating between programming languages. - AI-generated output is currently perceived as more reliable for developer workflows. - Designers may use AI to turn mockups into code, but much of design’s foundational work still involves: - Understanding user needs - Exploring problems nonlinearly - Learning about the broader problem space - AI initiatives increasingly originate outside design, with programmers, subject-matter experts, and stakeholders contributing ideas. ## AI Must Improve Collaboration, Not Just Individual Productivity - Eighty-five percent of respondents said AI had affected their personal productivity or workflows. - Common uses include text and image generation, brainstorming, and using AI as a sounding board or thought partner. - Respondents were three times more likely to report significant changes to individual workflows than to collaborative ones. - AI has not substantially changed group activities such as alignment or meeting facilitation. - Truly transformational AI products will need to support how teams work together, rather than focusing only on isolated tasks performed by individuals. ## Long-Term Effects Across Industries - Respondents in technology, professional and business services, and retail most often expected significant AI-driven changes to their products and services: - Technology: 41% - Professional and business services: 40% - Retail: 39% - Healthcare, energy and utilities, and telecommunications respondents expected the least impact over the following 12 months. - The report argues that realizing AI’s full potential requires considering how major institutions and essential services—not just software products—will evolve. ## Experimentation Before the Product Race - AI development is still characterized by experimentation, play, and research. - Teams face pressure to release new AI features quickly as new products, applications, and research appear constantly. - The recommended approach is to embrace uncertainty, iterate thoughtfully, and determine which ideas genuinely create value. - As the technology matures, the most successful products will likely come from careful exploration rather than simply adding AI features because of market hype.

Read original

Large Language Models AI Machine Learning Generative AI+2

figmaDec 6, 20232 min readCurated summary

Shipping Hype: PMs on What it Takes to Bring AI Features to Market | Figma Blog

AI’s rapid rise has pressured companies to launch features quickly, but hype alone does not produce useful products. Product leaders at Figma, Asana, Duolingo, and LinkedIn argue that successful AI development starts with real user problems, clear definitions, and realistic expectations about current models. AI should be treated as a tool for improving valuable workflows—not as a solution looking for a problem. ## Start with User Problems - Teams should identify user needs before deciding whether AI belongs in the feature. - Figma PM Conor Woods recommends asking: - Can the problem benefit from a large existing data set? - Is some margin of error acceptable? - Is AI genuinely improving the experience, or merely hiding poor UX? - LLMs are well suited to tasks such as organizing information and generating summaries, but they are unreliable when perfect accuracy is required or when they must invent entirely new experiences. - AI-generated inaccuracies and hallucinations are unavoidable with current models, making AI inappropriate for high-stakes, precision-critical tasks. - Asana uses a simple test: does the feature save users meaningful time? - Its Smart Status feature drafts project updates, reducing a task from roughly 20 minutes per week to two minutes and making the return on investment immediately clear. ## Specify the Problem Precisely - Generative AI can serve many different underlying needs, which makes vague feature descriptions dangerous. - Saying “we’ll summarize text” leaves open important questions about the user’s actual goal. - A user might want a summary to: - Understand a document’s subject - Identify action items - Extract decisions or other specific information - Product teams need to define the desired outcome and detailed use case rather than relying on broad descriptions of AI capabilities. - Greater specificity helps designers, engineers, and stakeholders develop a shared understanding of what the feature should do. AI features are most effective when they address a concrete, measurable user problem and acknowledge the limits of current models. Teams should define the user outcome first, then determine whether AI is the appropriate and trustworthy way to achieve it.

Read original

Large Language Models AI User Experience Generative AI+2

figmaAug 22, 20232 min readCurated summary

Give ideas more space with Jambot | Figma Blog

Jambot is a FigJam widget that brings ChatGPT’s generative capabilities into a collaborative, visual canvas. Figma created it to move beyond the limitations of linear chat, allowing people to ideate, branch into related topics, summarize discussions, and explore ideas together. The project reflects Figma’s broader view that AI interfaces should become more spatial, tangible, and multiplayer. ## From Chatbots to Creative Collaboration - Large language models can simulate spontaneous brainstorming and provide a broad base of knowledge. - ChatGPT’s conversational format is useful but can feel one-sided and restrictive during creative work. - Jambot was designed to make AI interaction more collaborative and adaptable to group ideation. ## Limitations of Linear Chat - Chat conversations present ideas in a one-dimensional sequence. - When ChatGPT offers multiple possibilities, exploring one path makes it difficult to return to another without scrolling and repeating questions. - Linear chat makes it unnatural to branch into related topics, compare alternatives, or see how ideas connect. ## A Visual Alternative - Jambot began as an internal Figma AI hackathon project described as “a visual version of ChatGPT.” - Its concept draws on networked-thinking tools such as Roam Research and Logseq, which link and organize ideas across pages. - The team was also inspired by Albus, which gives AI interaction a more visual structure. - LangChain influenced the idea of making sophisticated AI workflows visually tangible rather than requiring users to write code. ## Rethinking AI Interfaces - The team argues that users are currently “stuck in chat boxes,” much as they became dependent on video-call interfaces like Zoom. - Existing AI interfaces can feel primitive and command-line-like, despite decades of progress in graphical user interfaces. - Designers have an opportunity to develop new interaction patterns that provide more context, identity, and flexibility than simple conversational prompts. - A visual canvas can make AI more approachable while supporting branching ideas and shared participation. ## What Jambot Enables - Ideation and brainstorming directly inside FigJam. - Summarizing conversations or collections of ideas. - Riffing on concepts and extending them in multiple directions. - Collaborative exploration of AI-generated output within a multiplayer workspace. Jambot’s central recommendation is to treat AI as something that can inhabit richer environments than a chat window. By placing generative AI on a shared visual canvas, Figma aims to give teams more space to explore, connect, and develop ideas together.

Read original

Large Language Models Generative AI Figjam Chatgpt+2