Dante T.

🚨 Excited to share our latest project: IndySafeBot! 🤖 At the Indy Civic Tech Hackathon 2025, our team built IndySafeBot, an AI-powered public safety assistant that answers emergency-related questions using a RAG (Retrieval-Augmented Generation) pipeline. 🧠 Ask it things like: “What should I do if a volcano erupts?” “How do I report vandalism or graffiti?” “What goes in an emergency kit?” 💬 Built with: Python + Streamlit for the UI FAISS + Sentence Transformers for retrieval FLAN-T5 for answer generation A user feedback loop to improve over time! 📚 The bot learns from corrections and updates its knowledge base—empowering communities to get accurate, real-time safety info. Check out the demo and code: 🔗 GitHub: https://lnkd.in/ddAwJuCX Huge thanks to the organizers and our amazing team! #IndySafeBot #CivicTech #AIForGood #Hackathon #Streamlit #RAG #NLP #PublicSafety #MachineLearning

13

1 Comment

We've already seen how AI can be weaponized against communities of color, just look at its use in criminal justice, where algorithms like COMPAS have falsely labeled Black defendants as high-risk at nearly twice the rate of white defendants. Are we ready for that same flawed technology to become the backbone of our education system? The Minnesota Spokesman-Recorder's powerful piece "AI in Schools: Revolution or Risk for Black Students" asks this exact question. At a glance, AI in classrooms sounds promising personalized learning, reduced administrative burdens, and faster feedback. However, for Black students, the reality is more complicated; Bias baked into the algorithm: From grading to discipline, AI tools are often trained on data that reflect society's worst prejudices. The digital divide is still very real: Nearly 1 in 4 Black households with school-age children have no access to high-speed internet at home. Whose perspective shaped the tech? A lack of Black developers and decision-makers means many AI systems fail to recognize or respond to our students' lived experiences. And yet, the rollout is happening—fast. One in four educators plans to expand their use of AI this year alone, often without meaningful policy guardrails. We must ask: Who is this tech designed to serve—and at whose expense? This article is a must-read for anyone in education, tech, or equity work. Let's make sure the "future of learning" doesn't repeat the mistakes of the past. #AI #GlobalEducation #publiceducation #CommunityEngagement #equity #Youthdevelopment #AIinEducation #DigitalJustice #EquityInTech #EdTechWithIntegrity Read the article here: https://lnkd.in/g9U7za_k

28

2 Comments

Illinois just banned AI therapists. Illinois has notoriously been on the forefront of AI regulation (see their prior facial recognition and AI hiring legislation). They usually get it right. But here they messed up. I get it, people are scared. There’s even real cases of people being told crazy and dangerous things by AI models. At the same time, ~50k people die by suicide annually in the US. AI isn’t a replacement for therapists, but it’s 24/7 and extremely affordable. Even vanilla ChatGPT, with its problems, is better than people having no accessible option to speak to. Layer on models and products (Sonia (YC W24) or Ash by Slingshot AI) specifically built for therapy with safeguards and best practices built in, and it’s a no brainer. Let’s not be idealists, let’s play the game in front of us. AI doesn’t have to be perfect. It just has to be better than our current options. This may be controversial, but I guarantee an AI therapist will cause harm to someone. An AI doctor will misdiagnose someone. A self-driving car will kill someone. But so do humans, and we need to look at the net impact on society, not the unfortunate anecdotes. Let’s have oversight, but make it reasonable. https://lnkd.in/gknXrGcu

9

1 Comment

The hard part about AI isn't actually using the tools, it's keeping up with what tools to use. It's a full time job just trying to absorb the information each day on what's happening in the space... How does everybody else keep up?

15

16 Comments

Here’s a list of papers I’d study before planning / developing a generative music model. They’ll give you a strong foundation. You’ll also get an idea of the state of the art, and what to expect. 1. Museformer: Transformer with Fine- and Coarse-Grained Attention for Music Generation  2. Mousai: Text-to-Music Generation with Long-Context Latent Diffusion 3. MusicLM: Generating Music From Text 4. Simple and Controllable Music Generation 5. Masked Audio Generation using a Single Non-Autoregressive Transformer 6. Long-form music generation with latent diffusion 7. Video2Music: Suitable Music Generation from Videos Using an Affective Multimodal Transformer Model 8. ACE-Step: A Step Towards Music Generation Foundation Model You have to be comfortable with the topics below to follow these papers: - Transformers  - LLMs - Diffusion / flow models - Encoder architectures - Spectrogram / Mel-spectrogram representations — 💼 Check my music tech advisor service: https://lnkd.in/divEFrTA 📩 Follow my content: https://lnkd.in/d7MPrPQT #GenAI #AIMusic #GenerativeMusic 

164

8 Comments

Vibe coding is fun. But real product teams don’t start from scratch. They iterate on what already exists. With Anima, you can clone your existing product into Playground and work directly on top of it. Need to add a new section? Select where it should go. Capture a section from another live site, or simply prompt Anima to generate one. It automatically adapts to your brand style, spacing, and visual language. That’s vibe coding with context. Not a blank canvas. Not generic AI layouts. An evolution of your real product with production-ready code. Extend. Refine. Ship. Chrome extension: https://lnkd.in/d_7hyJMi

14

One conclusion I keep coming back to with overfitted codecs is that they behave like living codecs. Unlike traditional codecs, an encode isn’t a one-shot process. If you keep the source file and the bitstream, you can resume encoding and push the optimization further without starting from scratch. That’s a valuable property. Imagine a video suddenly surging in popularity: instead of re-encoding from zero, you could refine it on demand, squeeze out more efficiency, and reduce distribution costs. With overfitted codecs, the bitstream is not the end of the story. It’s just a checkpoint. #VideoCompression #OverfittedCodecs #NeuralCompression

28

5 Comments

I'll be going to GDC next month! If you're working on GPU hardware or graphics tooling, I'd genuinely enjoy a conversation. A bit of context on what we're doing: 𝗧𝗿𝗮𝘃𝗲𝗿𝘀𝗲 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 focuses on advanced graphics programming research. We publish technical work on rendering, build open-source tools, and work directly with hardware vendors on performance analysis. 𝗘𝘃𝗼𝗹𝘃𝗲 is our GPU benchmarking suite. Instead of a single composite score, it measures performance across 7 categories. The idea is to give hardware teams and reviewers a clearer picture of where a GPU actually stands. We work with teams at Qualcomm, Intel, AMD, and ARM, among others. If any of that overlaps with what you're working on, or if you just want to grab a coffee, drop me a message and we'll find time.

38

3 Comments

Cursor's new Composer model is (not so quietly) Chinese under the hood. Cursor shipped the new version of Composer as its "first in-house frontier coding model," optimized through RL for real coding tasks. Within days, users noticed Composer's hidden thinking was full of Chinese characters. Industry press connected the dots: both Cursor's Composer and Windsurf's SWE-1.5 appear based on Chinese large models (likely Qwen bases), fine-tuned and wrapped in US products. Community reaction is split. Some call it a major downgrade: slower and struggling with complexity. Others report speed gains and slightly better performance. So why is the reception so mixed? Gavin Leech's AI Tigers post (link in comments) shows that on fresh benchmarks, Chinese models lose around 21% performance compared to roughly 10% for Western models. They look closer on leaderboards than they perform on novel problems. The cost story is similar: 3-6x cheaper per token, but often requiring 2-4x more tokens, which shrinks the cost advantage fast. So why would Cursor make this trade? Because they're not trying to replace GPT or Claude. They're owning a narrow, high-volume behavior and squeezing it with fine-tuning. Cursor has massive amounts of product-specific data (traces, diffs, accept/reject signals) and can invest in the RL loop to extract value from a cheaper base. This is where Chinese and open-source models actually fit in the stack. They're a strong fit for narrow problems with abundant training signal: tab complete, inline edits, code transforms. They're a weaker fit as primary reasoning for high-stakes, long-horizon workflows. Frontier US models still dominate the second category. Chinese open-source is taking over the first. Composer's rollout is a case study in how powerful Chinese models have become, and a live demo of where their limits still show.

39

12 Comments

Weird prediction: I think the role of “Context Engineer” will closely follow the arc of labubu dolls: For the next 2 years the odd creatures will be hard to find and deeply coveted, and in 5 years everyone will have forgotten about them. What is a Context Engineer? One of the hardest parts of adopting many agentic AI tools is that it is really hard to give the AI agents all the context they need to make smart decisions. At Luxury Presence we have Zach Wills and team doing some extremely innovative things to feed context to different AI agents that we use internally. It’s truly inventing a new type of engineering. Context Engineering. But in a few short years, hooking these systems together will just be a basic IT task and AI will build all the context it needs by itself. In the meantime, if you can find a good Context Engineer now, jump on the opportunity! If you want to be left behind, wait until they have all moved on and your labubu dolls are collecting dust.

16

1 Comment