Prompt to polished slides in minutes. No manual research. No formatting headaches. Manus AI turns a single prompt into a boardroom-ready deck, sentiment charts, comps to past releases, and PR-ready insights. Brad Pitt’s reputation, decoded in 60 seconds. Manus one prompt. It swept the major review sites, graphed sentiment, surfaced comps to past films, and packaged PR-ready talking point Main Street teams get decision-ready narratives without a research department. Point Manus at your brand, product line, or location and hit go. What would you analyze first? #ManusAI

Interesting read. Agree that everyone’s crowding upstream: NVIDIA, foundation models, the obvious infrastructure plays. Feels late, upside is capped, and will likely be a race to the bottom for many players who don’t get out early. The earlier opportunity? Fishing downstream (or at least anywhere other than upstream). Integrating AI into every workflow, business and personal. That’s going to take years of iteration and create wave after wave of new value. I'm personally betting on value creation for marketers, now that ideas are easier than ever to build into personalized customer experiences. One nit with the article: Most of the article focuses the value that investors can capture. That seems narrow. AI can (and should) create broad value for the world. The best companies will do both, but even if some just make work and life better without minting billionaires, that’s still a win.

Why do folks tend to put the 'TL;DR' at the end of what they've written? I can't remember where I saw this, a film, a show, but it seemed good advice at the time: ✅ Tell them what you're gonna tell them. ✅ Tell them. ✅ Tell them what you told them. I think it was a legal show, and it was advice from an experienced old hand to someone making their first case in court. (It's also been attributed to many great minds, including Aristotle.) Since then, I've encountered it a few times; as a presentation tool, use as a thesis preparation tool, countless others. Thing is, the idea that you can bury the summary of your ramblings in the middle, or at the end of them (as I've just done), is ridiculous, and so pre-AI. The internet, and its readers (user intent being all powerful with the advent of LLMs, and GenAI as 'search engines'), has moved on. Move with it.

The first "reasoning" model dropped just a year ago. While it was a game-changer, it suffered from the same problems many of its predecessors did. GPT-5 on high reasoning effort is a different animal, and I suspect models that will be released in the next few months will be too. What changed in 12 months? - Incredibly more depth and contextualisation - Much more tool fluency - Partial failures don't derail the model anymore - Much longer reasoning horizons It's incredible how quickly we normalised and became used to the rate of progress we are experiencing today. It's a ridiculous time to be building.

“𝗢𝗻𝗲 𝗵𝗲𝗮𝗱 𝗶𝘀 𝗻𝗲𝘃𝗲𝗿 𝗲𝗻𝗼𝘂𝗴𝗵. 𝗧𝗵𝗮𝘁’𝘀 𝘄𝗵𝘆 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗲𝗿𝘀 𝘂𝘀𝗲 𝗺𝗮𝗻𝘆." In my latest blog, I break down Multi-Head Attention how it works, why it matters, and how it powers modern GenAI models. Clear visuals, simple analogies, and a touch of fun so you’ll never look at attention the same way again. Read here 👉 https://lnkd.in/gyy7i_HD

I’ve been experimenting with GPT for rapid storyboarding, and here are some early results. The story: a macaw protecting the forest by transforming into a mystic, Godzilla-like creature. The frames turned out as expected for a first attempt—now the fun part begins: refining and enhancing with different composites to push the visual storytelling further. Exploring how AI can accelerate creative pre-visualization is exciting, even when the first drafts are a bit raw. Every frame is a step closer to sharper narrative design. #AIforCreatives #GenAI #AIStoryboard #InnovationInAction #FutureOfCreativity #DesignThinking #ImmersiveExperiences #CreativeExploration

Most AI tools forget the moment you close them. That’s fine for one-off chats but for developers and creators, it’s a dealbreaker. That’s why we brought Pieces Long-Term Memory directly into #Claude Desktop through #MCP: so your AI can finally remember the work you care about. - ask_pieces_ltm → Query your memories inside Claude 1. Ask Claude: “What video content have I been working on recently?” 2. Pieces fetches context from Final Cut Pro, QuickTime, and Screen Studio. 3. Claude responds with real project history. - create_pieces_memory → Add new memories from Claude Working on something new? Just ask Claude to “create a memory.” Pieces saves that knowledge privately to your long-term memory. Another Episode of Pieces Academy is live: How to use Pieces Long-Term Memory with Claude Desktop. 👉 Watch it here: https://lnkd.in/dgU7WmNt

🎯 As a self-proclaimed algorithm guy, I’m having a lot of fun right now optimizing short-form video. This edit of our episode with Aric Whiteley on his journey to founding his own agency pulled 11x more views than the unedited version. 📈 Short-form video is home to the shortest attention spans on the internet - every millisecond matters. Right now, I’m optimizing for engagement (view rate), which then fuels reach. But the real question is: what do you do with that attention once you have it 🤔 YouTube Shorts has become an unlimited playground for testing that. Side note: This video was largely human-edited, not AI. Tools like Opus Clip still can’t match the quality of a skilled human edit (at least from what I've seen so far). 💬 Have you found human edits or AI edits perform better for you?

✨ Can I say something? I don’t really like listening to audios generated by NotebookLLM 🎧🤖 — and that’s despite being an early adopter of the tool. I would listen if I knew exactly what the input was 📝➡️🔊. But I also worry about hallucinations 😬. For example, if I land on a blog post and see AI audio at the top, I usually ignore it 🚫👂 because: 👉 I don’t know the input 👉 Potential hallucinations So, if this is something you want to use, I’d recommend writing a quick summary 📌 of the input documents and clarifying whether the audio was human-reviewed 👩💻✅. Just a thought 💡.

I use a couple of models to create great work! GPT-4o for structured ideation, Claude 3 Opus for narrative polish, and a diffusion model fine-tuned on branded experiential references for visuals. But here’s the twist: models don’t win the game on their own. The real differentiator is the prompt architecture (how you layer context, constraints, and temperature) and the human curation on the back end. Think of it like a DJ set: the tracks (models) matter, but the magic comes from how you blend them, drop them, and know when to scratch.