... Claude Code for most of what I build, and these go through it. The drafts are mine though: my thinking, my call, and nothing goes out that I haven't read line by line. Which is on topic, I suppose. That's the gate I was describing, and you just found the tell it doesn't catch. Taking the note on the dashes. Comment from Aadil Ghani(@aadilghani): @raffay_sajjad   Then the check I gave you has the same ... ... would sit with, and it is a limit on what I sell. A gate would not have helped. Nothing asked. The agent was doing things that were individually permitted, correctly permitted, and wrong in aggregate, and an approval layer tuned

... biggest update is in-scene entity editing you can swap or edit specific characters and objects within a single scene without breaking the overall visual composition. You can also trim, add, or delete individual clips post-generation, and fine-tune Foley, swap multi-language TTS voices, and edit subtitles right on the timeline. 5-Minute Video-to-Video (V2V) You can now transform real videos up to 5 minutes long. To prevent timeouts and memory crashes, our engine automatically slices ... ... generations fast and affordable, we pair silent models with a custom audio engine. It automatically detects visual actions to generate synced Foley and sound effects, or syncs visual cuts to custom music tracks up to 2 minutes long. Smart Topic

... experimenting with AI-assisted content creation for a highly technical B2B industry. One issue I keep running into is accuracy. AI can generate a polished article very quickly, but when the topic involves manufacturing processes, tooling, tolerances, materials, or failure analysis, the output often sounds convincing while missing important technical details. My current workflow is: Collect notes from engineers and production teams Use AI to organize the information Turn it into articles, short videos, FAQs, and social posts Have someone ... ... others are solving this. If you work in a technical or niche B2B industry, how do you keep AI-generated content accurate without slowing the whole workflow down? Are you using RAG, an internal knowledge base, custom prompts, fine-tuning

... notes, a recording, a screenshot, a question, or a topic, and turn it into the format they need. Underneath that, the system is meant to learn from behavior: which explanations work, where difficulty should move, when a visual or interactive question would help, and what should be revisited next. We are also training the Metis model family from scratch specifically for Lernex rather than fine-tuning

... positioning to make that the first thing people see. Comment from Sajid Javid(@sajid_javid): Love this format, Nolan, thanks for kicking it off. What I'm building: IntelliResearch, an all in one AI research assistant. You type a topic once and a dozen agents run in parallel: papers across six databases, then a literature review synthesized only from those papers (every claim cited, not hallucinated), plus a comparison table, research gaps, datasets, code, patents, conferences, and grants ... ... hardest part I'm solving right now: Low-latency native voice with proper echo cancellation, built in Rust on top of Tauri. Getting that audio engine feeling as snappy and reliable as Discord's took way more tuning

... direction. Comment from Ahmad(@blah_mad): My prior would be severity first, ratio second. One clearly bad high-risk outcome should tighten immediately. For normal misses, I d start around 2 bad outcomes in 20-30 governed actions, then tune per action class. A global threshold will feel clean in the model and wrong in production. Comment from Ahmad(@blah_mad): Yeah, I d keep that asymmetric. Auto-tighten when a bad outcome shows up. For loosening, I d make ... ... else. That's the seam Pushary's after, same gating idea but the policy is portable and cross-tool, so "auto-allow these, page me for those" is a saved thing you tune

... other services. Partnered with Bombora for high quality buying intent data and Prospeo / Hunters etc for B2B data. Agents generate the base line company profile and ICP along with Intent topics, users may edit / create new to fine tune

... trust it, and trust means it never sends or reschedules anything without you. So safe stuff runs on one click, anything touching other people or your calendar stops and asks first. And it is yours to tune, you can set each action to always ask or run on its own. Comment from Maxence Legu ry(@maxence_leguery): @fatima_shehzad  "A living space, not a column of discarded thoughts" is a lovely way to put it. That is basically the feeling ... ... hard part, finding the thing again six months later is. Buddy already pulls out what each note is actually about and links related ones together rather than leaving them as a flat pile, so you can follow a topic

... before the full reply is even done so audio starts ASAP instead of waiting on the whole response. Sentences are pipelined: while sentence 1 is playing, sentence 2's TTS is already being synthesized in parallel. TTS is tuned for first-audio (smaller server-side chunking), and frame prep (JPEG resize) runs off the event loop so it never stalls playback. Honest take: it's reaction-latency (~1-2s event voice), not twitch-reflex works great ... ... frame-perfect input. It's all open source if you want to dig into the pipeline: https://github.com/Alradyin/wall... Comment from Yelkhan (@alrady): @freya24  That's the dream There's already a cross-session memory layer (it remembers topics

... messiness, it s definitely synthetic! Do you find yourself checking for warped hands and hair first, or have the models finally mastered the "five-finger" challenge in your feed? Comment from Abdul-Hafiz Aderemi(@hafiz_aderemi): This is a topic I'm interested because almost everyone LinkedIn has turned into an expert at recognising AI content. I'll be reading the comments. Comment from Abdul-Hafiz Aderemi(@hafiz_aderemi): @lara_bishop  can you give examples? Comment from Maliik(@maliikb ... ... discussions. Comment from Stoyan Minchev(@stoyan_minchev): I don't. search for methods. As long as the content can genuinely make me better, smarter, does not really matter how it is created. Sometimes people use AI just to fine-tune

... where i also started working on "Global Memory Architecture" for this and this is going to be very essential. Now let's say if any micro lab or a team of small developers built any model or fine tuned any model that outperforms these big models at any specific tasks, and if in future we allow these people hosting their models on our platform and if chats with their model are ranking on the feed or the model is ranking ... ... provide real value to users? Comment from MAHENDRA BENIWAL(@mahendra_beniwal): @joey_tribbiani6  yes Comment from MAHENDRA BENIWAL(@mahendra_beniwal): @blakeskrable  Exactly... and one more thing i would say that sometimes you don't get enough topics

... heavy users. The live product is at lucidextractor.liceron.in if you want to see it. Happy to do a quick call to walk through the architecture for your research! Comment from Chris Conlee(@conleec): This is a great research topic, Ryan. As a solo founder launching PictaBase today, I can tell you that for a "pro-sumer" tool, the stack isn't just about what's easy it's about what's provable and sustainable . I built PictaBase ... ... have anyone they can really talk to without judgment, and Murror tries to be that space. Stack is React Native for mobile, Node/Express on the backend, PostgreSQL, and we're using a mix of Claude and some fine-tuned

... indirect relationships between ideas. Node coloring: Automatically color-code nodes by tags. Hover for tooltips showing topics, customize colors in settings. Multi-select & bulk actions: Hold Cmd/Ctrl and click to select multiple nodes, then perform bulk operations like deletion. Graph presets: Save filter settings, color schemes, and layouts as named presets. Switch instantly between different views of your knowledge. Graph settings: Fine-tune

... created a no-code AI Lab so that anyone can train their own small, specialized AI models using their proprietary data, without needing any technical skills. We abstract everything, from GPU configuration to data processing, pipelines, and hyperparameter tuning. Users simply upload their raw datasets, and we handle the entire process end-to-end. They go from raw data to their own small, specialized AI model, which they can run in our cloud, download and run locally, or deploy ... ... from Nika(@busmark_w_nika): @sup_nim  This is a problem of spanned attention. Peope should try a stronger will and reading books imo :D Comment from Nika(@busmark_w_nika): @nikhilshahane  Just out of curiosity, which topics

... started turning my personal habits into a checklist to build discipline and make these habits non-negotiable over time. Here s mine: 15 minutes of meditation Read at least 10 pages of a book Read one English article (any topic) Take notes on new vocabulary from that article Go for a run or work out Learn knitting Practice writing Python I can't do all of these every day although I really want. However my time budge has limits. Instead ... ... minutes to write a simple or complex problem I am trying to navigate, an affirmation to follow that problem, and a closing sentence to reflect on. After I finish, I let my body process the emotion so I can tune

... with AI, the greatest technology the world has ever seen, I predict the return of three pricing structures which are just so, inherently, human. And that, kinda, makes sense I've written a more in-depth Substack on the topic that I'll link to at the bottom, but here's the TL;DR to avoid you needing to switch platforms. A short summary of the thinking AI models are converging faster than we expected. For most users ... ... signals, and what it costs. In economic terms, some AI products may behave like Veblen goods where the price itself is part of the appeal. Second, fragmentation increases. Rather than one AI for everyone, we ll see many AIs tuned

... from Nika(@busmark_w_nika): @kavyashree_s1  Regarding public speaking, I have a few tips: practice in front of the camera practice in front of the mirror write down topics, prepare, and talk 15 20 minutes on some topic

... agents, fine-tuning, function calling, automation, and how teams can move faster with less manual work. I m here to answer any and all of your questions, so fire away! Comment from Henrik Thorenfeldt(@henrik_thorenfeldt): cool topic

... create interactive presentations in minutes I ve always believed that great ideas deserve more than static slides. Most of us spend hours designing decks, only to end up with something our audience quickly tunes out With Oina, we re changing that. You type a topic

... Lately, I have been experimenting with how to feed context into GPT models more effectively. For example, when fine-tuning or working with larger context windows, I have noticed that the dilemma is in organizing the surrounding information, rather than the prompt itself. Last week, I came to know that it's called Context Engineering. I am curious how others here approach it: How do you decide what information is essential vs noise when building AI workflows? Have you noticed ... ... model context. Comment from Ebony Belhumeur(@ebonybelle): @d_ferencha  We've built one actually, patent is pending, can't wait to show folks! Comment from Coire(@coire): Hi Ashok, I've been in the weeds on the same topic

... high touch engineering-in-loop . Check out his latest tweets on X or recent podcast with Dwarkesh Patel where he touches on the topic. At Navam we are vibe coding several non-trivial white label products across multiple AI stacks, models, and frameworks running up 177K lines of working production-ready code and months of tuning

... from Nika(@busmark_w_nika): @dzaitzow  Reddit is the place where people go for advice, so people trust people there. But dunno how often people click on your profile and links. I think it is better for discussing topics. As for me, I would prefer LinkedIn over Facebook (FB is too political and toxic for me). Comment from Nika(@busmark_w_nika): @charlene_zh1  But is it easy to build it? Do you have one? I would like ... ... Comment from Nika(@busmark_w_nika): @vladbalan  Congrats on a successful post, it seems that many of the comments are positive. That's rare for Reddit :D Comment from Nika(@busmark_w_nika): @rajan_batra3  Staying tuned

... question initiator" and think: "Why are middle-aged people more likely to be laid off? Is there data to support this? What angles can I use to explain this clearly?" You'll find that just around this topic, you can create tons of content ideas with different structures, facts, opinions, and styles. 2. From a "Single Role" to "Multi-Role Collaboration" Let's say you want to make a video about "overtime at work." Only showing the employee's perspective ... ... enjoyed this, @zhiqi_shi Zhiqi, the way you framed perspective-shifting as moving from your story to our story really stuck with me. Too often content gets stuck in one angle (rant, tutorial, or personal log), and the audience tunes

... your audience. Here s what it includes: Tailored Content Calendar: Organize your posts with a clear timeline and structured workflow. No more guessing what s next or scrambling for ideas! Engaging Themes & Content Ideas: Never run out of topics again! Get a list of themes that will keep your blog fresh, relevant, and engaging. SEO-Optimized Strategy: Craft each piece of content with search engines in mind. Learn how to seamlessly integrate keywords and boost your organic reach. Actionable ... ... rank higher and attract organic visitors. Engage Your Audience: Stay on top of your content goals and create posts your audience will love. Measure Success: Use KPIs to evaluate the performance of your content, so you can fine-tune

... mathematically mutually exclusive. That's why, the AI's mutability doesn't solve the fairness problem. It simply transforms it from an unconscious bias into a conscious, auditable, and strategic choice. Comment from Adi Singh(@adi_singh5): Fascinating topic! Do you think AI bias can ever be fully mitigated with better training data, or is human judgment still the safer bet despite its flaws? Comment from Dheeraj(@dheerajdotexe): This is really cool question, Dmytro. Personally, comparing two hypothetical human ... ... should apply their learnings in a specific way, there is certain to be bias for hiring, L&D, retentions, and so on. When we look at how UnitedHealthcare for example applies AI, we saw that it was basically tuned