Building AI Agents’ Post

At Paper Crane, we understand that keeping up with AI can feel like running face first into a screen door. Last week, I published a list of cool things that happened in the AI space and what it means for regular humans - most importantly for your business. We’re back at it again for the week of April 28th: If you’d rather receive these updates in your inbox, send me a DM and I’ll add you to the distribution list. 1. Pushback on California's Proposed AI Bill SB - 1047 What it is: This week, California was challenged to improve their proposal for regulating AI systems. What it means: According to Dean Ball - an AI blogger - this piece of legislation caters to the crowd that believes AI is an existential risk to humanity. It does not, however, seem to have been looked over by those at the cutting edge, who are building real models, and have opinions on the trajectory of development. It relies heavily on hypothetical models that somehow escape from data centers and cause tens of millions of dollars in damage. Here is Dean's substack on how he thinks good AI policy should work instead. 2. Language Agent Tree Search (LATS) What it is: As companies continue to try to deploy LLMs augmented with their companies information like financial statements or internal knowledge bases, the race continues to build the scaffolding that helps find and make sense of the companies information in useful ways. LATS spins up an ‘agent’ - a gpt that is assigned a task - and uses a monte carlo tree search framework to allow for deliberate and adaptive problem-solving guided by external feedback and self-reflection. Prior implementations of agents are kind of like when a dog sees a squirrel, it chases the squirrel not caring about anything else. With this tree framework, the agent can consider multiple options before taking action. What it means: Each week we continue to see new innovations and approaches to the simple implementation of Retrieval Augmented Generation (RAG -  putting an LLM over large internal documentation). With a few developers and the development toolkits provided by Langchain and LlamaIndex, companies can improve the viability and performance of their RAG solutions and achieve advances in company productivity. 3. How Perplexity Builds Product What it is: Lenny Rachitsky is one of the most prominent voices in the product development world. This week he published a guest post alongside Perplexity. Perplexity is a fast growing start-up (valued over $1 billion) that is disrupting internet search. Many are turning to Perplexity over Google search because of its useful plain language interface. What it means: The newsletter talks about how the AI startup uses AI to make key decisions in their business. It’s a cool look at how to properly approach using large language models when starting and scaling a business. Read the post here. 

Is generative AI truly delivering the value it promises, or is it just impressive tricks? Some interesting reflections from Benedict Evans supported by data and analysis. With consumer AI experiences, you type something in and get magic back! But this "magic" might not be useful in its raw form and could even be wrong. This issue is amplified when we try to use these tools in an enterprise context. That's why it's crucial to focus on use cases and user experience, integrating AI so seamlessly into the workflow that users don't even realize some form of intelligence is being used. Trust and accuracy are foundational, so AI must securely leverage company data. Guess which company has been pursuing this strategy and allows you to plug in your preferred LLM? 😎 #salesforceistheway

Introducing "Five for Friday" – Your Weekly AI Digest! 🗞 Super excited to launch a brand-new feature in my newsletter, New World Navigator: Five for Friday! Every week, I’ll break down 5 of the most interesting and impactful AI news events, providing concise analysis so you can stay ahead of the curve. Here’s a sneak peek into this week’s lineup: 1. Rising from the Ashes 🔥 Following his OpenAI exit, Ilya Sutskever has raised $1B for his new venture, Safe Superintelligence (SSI), focused on creating "safe" AI that aims to surpass human intelligence. But are we repeating the same promises made by OpenAI? Only time will tell. 2. Latest Kid on the Block 🚸 Alibaba’s Qwen2-VL model takes on the GPT-4 class in visual problem-solving, language comprehension, and more. Despite US technology trade restrictions, Chinese tech giants continue to invest heavily in AI. 3. AI for Your Dishes and Laundry? 🤖 Forget AI art; how about a humanoid robot to do your dishes? 1X Technologies, backed by OpenAI, is pushing the boundaries with their NEO Beta household robot. 4. Claude Goes Corporate 👨💼 Anthropic is now supercharging businesses with enhanced security, a 500K context window, and integrations with GitHub, with the newly launched Claude Enterprise. Can they successfully take on Microsoft's Copilot and OpenAI's ChatGPT Enterprise? 5. If You Can’t Beat Them, Co-Opt Them 🔊 Amazon’s Alexa gets a makeover, powered by Anthropic’s Claude. With new features like AI-generated news summaries and conversational shopping, is this enough to put Amazon back in the AI race? Which of these new stories caught your attention the most? --- Want to dive deeper into the future of AI? Check out the full article (link in the comments). And if you’re eager for more insights, be sure to subscribe to New World Navigator! ⛵ Let’s connect and discuss how your team can harness AI to innovate and grow! 💡

Harnessing AI: Why Vigilance is Key In a recent post titled, "Forget fancy AI - Mobile apps still rule wealth management (for now)" https://lnkd.in/e3J6Usef I mentioned that AI systems are incredibly powerful but, in my opinion, they are currently best suited to act as assistants and not yet ready to take over from a professional person. It is a controversial viewpoint because there are live AI systems out there doing incredibly well. I got flamed from some friends for these views. But, today I felt vindicated. One of Googles latest AI services, called AI Overviews has had a few, lets say hiccups! Here are a few extracts from an article by MIT Technology Review: "Within days of AI Overviews’ release.... - It suggested that users add glue to pizza or, - eat at least one small rock a day, " WOW, that is bad! The full article is available here: https://lnkd.in/e9HHrwfa And this is Google. One of the leaders in world tech. If this could happen to them...? Imagine if your shiny new AI assistant, which was trusted to give clients advice and support, started making things like this up? Who would bear the ultimate responsibility? As with anything new in technology, start with simple, non critical tasks that can easily be measured for success. Build up in scope and responsibility over time. Make sure your team is aware of the strategy and on the look out for possible issues. Be wary of the fancy marketing pitch promising the dream! Please share your views in the comments below! #ai #wealthtech #mobileapps #wealthmanagement #fintech

Rags to riches LLMs thrive on data, and for businesses, no source is as valuable as your internal data. To securely leverage data for better, more grounded prompts, orgs are using an AI technique known as Retrieval Augmented Generation (RAG). What you need to know:

LLMs thrive on data, and for businesses, no source is as valuable as your internal data. To securely leverage data for better, more grounded prompts, orgs are using an AI technique known as Retrieval Augmented Generation (RAG). What you need to know:

LLMs thrive on data, and for businesses, no source is as valuable as your internal data. To securely leverage data for better, more grounded prompts, orgs are using an AI technique known as Retrieval Augmented Generation (RAG). What you need to know:

LLMs thrive on data, and for businesses, no source is as valuable as your internal data. To securely leverage data for better, more grounded prompts, orgs are using an AI technique known as Retrieval Augmented Generation (RAG). What you need to know:

LLMs thrive on data, and for businesses, no source is as valuable as your internal data. To securely leverage data for better, more grounded prompts, orgs are using an AI technique known as Retrieval Augmented Generation (RAG). What you need to know:

LLMs thrive on data, and for businesses, no source is as valuable as your internal data. To securely leverage data for better, more grounded prompts, orgs are using an AI technique known as Retrieval Augmented Generation (RAG). What you need to know: