May Habib on LinkedIn: Conversation with May Habib, CEO of Writer on Successfully Transforming… (2024)

Stanford University's latest Foundation Model Transparency Index (FMTI) includes Writer LLMs, and the results are impressive! This was the first time Stanford Institute for Human-Centered Artificial Intelligence (HAI) evaluated Writer’s Palmyra-X model in FMTI, reflecting our growing impact to building customer trust through transparency, especially in enterprise applications where security, accuracy, and data protection are paramount. Palmyra-X achieved some of the highest marks for enterprise-grade models based on a set of 100 indicators. But in so many ways, our commitment goes far beyond FTMI expectations. At Writer, we understand that enterprises need top-notch performance paired with solid assurances about data management and protection. Our strategy isn't just to meet industry standards — it’s to surpass them. 🔍 Here are just some of the additional ways we're setting higher standards with our customers:Zero data retention: We never store data longer than is needed to provide our service, and give customers the right to delete data at any time.No training on customer data:
We don’t use the data you share with us to create, modify, or train our models.Zero model distilling & quantization: We do not distill or quantize our models, so customers can achieve peak reliability of accuracy, quality, and performance at all times.Toxic filter transparency: We share the mechanisms of our toxic content filters, ensuring our customers understand the safeguards we have in place to prevent harmful content generation.Bias mitigation: We’re transparent with our customers about our ongoing efforts and methodologies to identify and mitigate bias in our models.We’re providing our customers with technology that's not only powerful but also secure, and empowering them with the knowledge to make informed decisions. Our participation in this research is just the first step. We’re on a journey to raise the bar on transparency and redefine industry standards. Stay tuned for more to come!Link: https://lnkd.in/g8sMU2Qx

205

7 Comments

Is there any better feeling for a startup team than an operator they respect and admire SCRUBBING INTO THEIR BUSINESS?When Alex Wettreich and I met Andy Shorkey IRL for the first time, it was clear that the hyperscalingjourney we're on is one that Andy knows and loves.And in a nascent but exploding industry that's competitive AND dynamic — we needed a revenue leader who can share our obsessive focus on making customers successful. That's how we rise above the noise: by being the only enterprise generative AI platform that's delivering real ROI.Two of our leadership principles are "lead from the front" and "know our customers intimately + care for them passionately" — and it's incredibly rare to find experienced executives who can be as connected to customers and the people on our front line as Andy is.Can't wait for all of our customers and partners to meet you, Andy! Welcome to the team!💥💥💥And — he's hiring, folks!

264

14 Comments

🇬🇧🇬🇧🇬🇧It’s a beautiful day in London to announce…Writer's new GM of International, the incredible Brian O'Reilly, and our EU headquarters on Wimpole Street!💥💥💥 We're getting the international AI party started with a BANG! 1️⃣ I'm giving a keynote at the Generative AI Summit at Tottenham Stadium TOMORROW — and we'll share our learnings from helping enterprises unleash AI into production.2️⃣ We're hosting an invite-only AI Leaders Forum for AI execs in London at the end of May. Featuring an incredible lineup of speakers (all AI practitioners in the enterprise) and hands-on AI workshops with leaders from Salesforce, HSBC, Accenture, Adore Me, and Mars. They’ll be sharing an insiders view on how they’re scaling generative AI in their companies.DM / comment if you'll be at the Generative AI Summit this week and / or can join us at AI Leaders Forum on May 30.3️⃣ And of course — WE ARE HIRING IN LONDON:•AI research (much more to come on this)•Engineering (applied, front-end, back-end, product design)•GTM teams (sales, customer success, solution architects, field marketing, BDRs)It's a killer EU founding team on all fronts — JOIN THE FUN!

452

24 Comments

Google has burned the boats.The years of handwringing around generative AI and the core search business have resulted in a conclusive decision in favor of pushing innovation.It's great for consumers but scary for businesses. AI search summaries, which Google announced will officially *replace* the just-announced SGE (which people were just wrapping their heads around), will be the main search experience by the end of the year.Marketers, your job is going to get *even tougher*.Some thoughts on this week's product announcements:(1) Augmented reality has gotten a generative AI rebrand, and it might actually stick this time!Will Google introduce ways for brands to advertise on this new Astra platform? *Probably* — but it's going to be hella discrete and not bubbles popping up on your phone as you look at something.They will have to build in a way to connect what a user's phone is looking at with who is advertising in that place or against that object. Goodbye, keywords, hello, keyplaces or keythings.What it means for brands (and I think this is exciting): branding of interactivity / real life experiences take on new importance.(2) *all thing equal* AI search summaries most certainly mean fewer clicks to your website — which is actually going to most impact B2B websites.Does that mean you stop investing in content? No. But understanding that the content is for training LLM data scrapers (for LLMs that are updated max 3-4 times a year, for now) and agents that are browsing the web on BEHALF of your users (not just Google's) is important — imo it means that you'll need to close the loop to connect your brand to your users in the channels you KNOW are more likely to be their human eyeballs and not their AI ones:the almighty social.(3) We're going to see AI-agent web lanes emerge. What I mean — websites (or parts of your website or experience) that are *optimized for AI agents doing stuff on behalf of users.*Think of this as kinda like prompt engineering — figuring out how to talk and reason with the agents that come to your website.Search engine optimization (SEO) becomes AI agent optimization (AIAO).We're going to use Writer's own beta agents to experiment on our own website and will report back.Marketers, you gotta fight AI with AI.

821

62 Comments

Great to catch up with Caroline Hyde and Edward Ludlow at Bloomberg Tech in San Francisco yesterday! https://lnkd.in/gbGX24-6

269

13 Comments

There's a whole swath of the enterprise that does not know where to start when it comes to working with generative AI. It feels daunting, expensive, and time consuming — so they are letting the biggest tech shift in history just steamroll past them.We want to encourage people to dive in. That's why I'm hosting a workshop tomorrow at Maven on how to leverage AI across your enterprie. This will be PRACTICAL. You'll learn:• How to embrace the the full-stack approach to generative AI• How to make your data work for you• How to move out of POC purgatoryI'll cover some of the best use cases for generative AI today and the future use cases of tomorrow with real examples.RSVP for tomorrow: https://bit.ly/44qwxVZ

91

9 Comments

The U-curve of AI rollouts by Accenture's Marc Appel.Great read on helping users get through their own personal trough of disillusionment with AI.Going from "using" a tool to "interacting" with one is a new experience for most people, and Marc has some great learnings on how to do this at scale👇🏽

53

6 Comments

Generative AI transformation starts at the top. Full stop. If you don't have hands-on leaders who talk about it, build with it, showcase it, promote it, LOVE it, it simply won't take root in your company. Too many of the processes generative AI touches are cross-functional // cross-application // cross TIME for bottoms-up adoption to be the way work gets re-invented with AI. But it's worth the leadership investment. Our customers who have CxOs who have - played with the technology themselves- understood the use cases deeply- allocated headcount towards generative AI program mgmt- put their own chiefs of staff on itare the ones who are succeeding.

223

16 Comments

If powerful LLMs were all that was needed to get to great enterprise generative AI programs, then hundreds of thousands of open-source and closed-source LLMs would mean we'd all have the AI programs of our dreams.But most enterprise use cases require retrieval-augmented generation (RAG), which basically is a way to give an LLM-based AI app a lot more supporting information about a question being asked / workflow being run.Despite a lot of time and money invested, many RAG-based apps are still failing at being intelligent, and at Writer, we've found graph-based RAG to very often fix an implementation.Visualizations are helpful to explain WHY you can get better, deeper, more accurate answers when you integrate knowledge graphs into RAG.Column 1 is vector-based RAG. Here's what happens:1️⃣ Data processing: Text data is split into smaller chunks, converted into numbers (vector embeddings), and stored in a vector database. Numerical distance between embeddings is how we know how things are related.2️⃣ Query and retrieval: In vector-based retrieval, the user's question is converted to an embedding too, and we use those query data points to retrieve the closest other data points to the query, using "nearest neighbour" type of algos — it’s a similarity search.3️⃣ Answer generation: The retrieved vector embeddings (the neighbours!) are remapped to the original text data and sent to the LLM for answer generation.Column 2 is Writer's graph-based RAG. And here's what happens there:1️⃣ Data processing: Writer's uses a built-for-purpose LLM to *first* split the data you want for RAG into nodes and builds relationships as edges, and stores the node/edge data in a graph structure. The edges are SEMANTIC relationships that would otherwise be lost if you turned text into numbers/vectors.2️⃣ Query and retrieval: Graph-based retrieval breaks down the query into entities and semantic relationships, and then uses NLP algorithms, heuristic algorithms, and machine learning techniques to map those relationships to semantic connections in the graph of your data, and retrieves both relevant nodes/entities and relationships as potentially relevant. Much more similar to how a human thinks. We think in relationships, not text similarity. 3️⃣ Answer generation: The retrieved entities and relationships are extracted from the graph and sent to the LLM for answer generation.So which is better? It depends on the use case. 🔸 Vector-based RAG can be used for simpler tasks where contextual understanding is not as important — use cases with vast amounts of data but straightforward relationships.🔸 Graph-based RAG is better for more complex tasks where deeper contextual understanding is critical, aspects of the answer will sit across multiple sets of data/docs, and more "reasoning" is required. Use cases with important relationships—medical diagnosis, legal research, etc.

557

52 Comments