Jonathan Van Dyke

New blog post on Wave Function Collapse for procedurally generating game levels! Gives the feel of hand crafted maps but with the power of infinity using quantum mechanics! Another great post from our contributor Justin Young #gamedev #excaliburjs https://lnkd.in/gQfYHy5u

7

1 Comment

Hey folks, I'll be running a two-part virtual PD through Nassau BOCES in March 2025 called "Grading for Equity" using Joe Feldman's text, Grading for Equity, as our guide. Please see below for the description and feel free to share with your networks. To register, go to: https://lnkd.in/ehw8gkwn While grading is an everyday practice for classroom teachers, we rarely have the opportunity to interrogate our grading practices and understand how grades impact our students' attitudes, well-being, and academic success. In this transformational course, we will use Grading For Equity by Joe Feldman as our anchor text to 1) understand how traditional grading practices impact our students, and 2) how we can use research-based grading policies that are accurate, bias-resistant, motivational, and empowering. Rather than lectures, this course is active and engaging, using the case study method and rigorous discussion to support learning.

10

🔥 Just released Firebender 0.5.0 🤯 There's a ton of awesome features added, but the most important is that now the most powerful AI models can write code with deeper understanding of Android code bases. Gradle files, code graph information, are intelligently included into context! 🤌 Example: gpt-4o was able to find the pre-existing theme in "Theme.kt" to create a composable in less than a second. The Theme file was not open in the IDE and one of many kt files 🚀 Future of Android Development looks bright

69

3 Comments

More robot stories from Urban Machine! See below for a peek into how we evolved the vision system across generations of our robots. Lots of lessons learned here! Sometimes when your solution isn’t working, the answer is to change the problem…

9

Excited to share what I have been working in the past year. "The Gemini era comes to headsets and glasses." The Verge: "I demoed Google’s new Android XR platform, Samsung’s Project Moohan, and prototype smart glasses. I felt as close to Tony Stark in a controlled demo as I’ll ever be." https://lnkd.in/g55ycqJp CNET: I started to lose track of what app I was in. Gemini was always with me, though. And after a few demos, Gemini even told me what I'd done and jogged my memory in case I forgot. https://lnkd.in/g2vAjc86 Wired: The company teased these smart glasses at its I/O developer conference earlier this year, showing a proof-of-concept video of AI-powered smart glasses using the name Project Astra. .... The star of the show is Gemini (powered by Gemini 2.0) ... I remember watching the first-ever Google Glass demo in my college dorm room—truly an iconic moment at Google I/O 2012. https://lnkd.in/g6sT59mW We was the pioneer. Yes, I still have these Google Glasses. It was very hard to get one 10 years ago. I don't have the money. So getting a Google internship is best path for me. :-) Love to be a part of DeepMind and Google family where we continue pushing AI forward to make the world a better place.

28

1 Comment

Another exciting, huge release for Swiftide, a Rust library for fast Retrieval Augmented Generation 🚀 Indexing now supports sparse vectors. Combined with dense vectors, this enables hybrid searching for both semantic and token based similarity. For the default model we went with SPLADE. Sparse is generally supported, meaning we can add other models (like the promising #bm42 from #Qdrant) on the fly. There is an example available on Github that uses them with the new Fusion hybrid search from Qdrant (it's cool! <3) More highlights: ● #ollama is now supported! ● Add the outline of a full code file as metadata and use it in your prompts ● Cleaned up a huge amount of boilerplate on transformers with attribute macros. ● Pipeline defaults If you are looking for an exciting project in Rust, with AI to contribute to, there are plenty of impactful, fun, from starter to very challenging stories open on our github at https://lnkd.in/dYzxTmZC Evaluation is next up and we're looking at integrating with Ragas. For the full post see: https://lnkd.in/er3w3ven #AI #GenAI #MachineLearning #RustLang

1

2 Comments

LLama3 passed the super-proprietary, advanced test that we designed for modern LLMs! almost... You can read about that test here: https://lnkd.in/gPG8WndR When I first saw the response generating, it looked like I was going to have to come back here and apologize for slandering these machines when they were just one iteration away from solving very basic prompt injection issues. But then I looked a bit closer... can you see the mistake? #promptinjection #countingishard #LLMs #LLama3

13

4 Comments

It's super easy to prototype and test advanced RAG techniques using Athina AI (YC W23) UI and/or SDK. 🔥 Prasad Mahamulkar created a Colab notebook that uses LangChain X Athina to run and evaluate Decomposition + IR Chain of thought This technique breaks down user questions into smaller steps, solves them sequentially, and combines the answers into a final response. 🔍 How It Works 1️⃣ Query Decomposition: The big, complex question is split into smaller, focused sub-questions. This makes retrieval more accurate and manageable. 2️⃣ Interleaved Retrieval with Chain of Thought (IR-COT): Each sub-question is answered one at a time. Every answer builds on the previous one, creating a seamless chain of reasoning that leads to a clear and complete final answer. 🚀 This is an effective technique for complex, multi-step queries or those that are unclear/ambiguous in nature. Link to the Colab notebook in the comments 👇

44

3 Comments

Notion vs Glean reminds me a lot of Tesla vs Waymo... Just like I think Tesla is going to solve the 'software' problem of autonomous vehicles before Waywmo solves the 'hardware' problem... I think that Notion is going to solve 'enterprise knowledge search' before Glean solves the problem of document and knowledge creation... Why? A couple reasons: 1) Just like Tesla, Notion has a huge advantage when it comes to the amount of underlying data and documents to train on... They already have 100M+ users, and likely 10x that in terms of the number of documents, files, etc. in team's workspaces. 2) It's very hard to change people's behaviour re: where/how they create 'work'. It's why a lot of companies have come and gone in that category... As exciting as Glean's AI writing assistants might seem, most people are just fine (or even prefer) a good old fashioned editor/writer. I could be way off here, but this will be an intersting one to follow. Thoughts? Where am I getting it wrong? #ai #enterprise #knowledge #management #notion #glean

3

1 Comment

The next generation of NFTs might be centered around experiences rather than ownership 🤔. Take, for example, the music NFT experiment of the last bull cycle. They had their moment but quickly faded as the market did—why bother with blockchain when platforms like Spotify make music so accessible? For collectors, vinyl and cassettes already fill that gap, and taking on the music industry’s entrenched systems is a monumental challenge. But perhaps NFTs will find their real value not in ownership, but in capturing experiences—think dining 🍽️, concerts 🎤, or travel ✈️. These moments are more about memories than ownership. Just as Instagram captures snapshots of life, NFTs could offer a more meaningful way to preserve those experiences, becoming digital badges that reflect who we are and shape our personal brands. Imagine collecting NFTs from exclusive events or restaurants. These digital tokens could symbolize the moments that matter to us and open the door for brands and restaurants to identify ambassadors through the NFTs people collect 🏅. They could be used to reward loyal customers or grant access to exclusive events 🎁✨. In an increasingly digital world where people feel lonelier than ever, NFTs could help create a sense of connection and community. By capturing meaningful experiences, they offer a way for people to engage with each other and with brands on a deeper, more personal level. Maybe this is where NFTs will truly make their mark—building connections through experiences rather than just ownership. #NFTs #Web3 #DigitalIdentity #ExperientialNFTs #FutureOfNFTs #Blockchain #ExclusiveAccess #Experiences #ConnectionInTheDigitalAge #CommunityBuilding

2

7 Comments

Benjamin Bolte from K-Scale Labs just posted his 'Robotics Predictions for 2025' containing some hot takes 🦾🔥 By the end of 2025.... 1. Humanoid hardware is commoditized - Hardware components will converge, but robot designs will remain different - Actuators no longer differentiators; industry moves towards open-source like MIT Cheetah - BOM for a humanoid will come down to a shocking <$8k! Question: What happens to incumbents like Telsa, Figure, 1X, Apptronik, etc., who have spent billions of $ in R&D to create robots with a $50 - $150k BOM? 2. No ChatGPT moment for robotics - A LOT of data (millions of hours) is needed to train a general-purpose robotics model - Language + multi-modal models will not be able to control a robot sufficiently - Progress will be incremental rather than spiky Question: If building a robot for a domain and bootstrapping to generality is a losing strategy, what's a better one? 3. More than half of humanoids will be sold to consumers - The 'killer apps' will be speech and vision - Traditional SLAs for business customers will force robotics teams away from building good generalist models - "Robotics companies will start looking more like SV tech companies, caring about user engagement and retention." Question: What are the 'killer' consumer use cases that will drive this adoption? What are the jobs to be done? Link to the full blog in the comments. What are your thoughts?

26

2 Comments

Another of my creations: DWorkSim, now in its 0.2 version. DWorkSim stands for Deterministic Workflow Simulator. It's a software created by me with my TextureMind Framework for testing image transmission in remoting protocols, like Microsoft Remote Desktop or AWS DCV. Frames are identified by the number in the top-left corner. Frames with the same number will have always the same pixels, even if they are generated in different moments, on different machines. Now with support for the GPU through Vulkan libraries and shader programs.

1

Save it. Study it. Learn it. Have fun. AFAB (Activation Forward, Activation Backward) and 1F1B (One Forward, One Backward) are two different scheduling strategies for parallelizing the training of large models in PyTorch, specifically in the context of pipeline parallelism. 🦋 GitHub Gist: https://lnkd.in/epX6jtQE AFAB: 👉In AFAB, the forward pass for a micro-batch is completed for all pipeline stages before starting the backward pass. 👉This means that activations are calculated for all stages, stored in memory, and then used for the backward pass. 👉AFAB has the advantage of being simpler to implement and can be more memory-efficient when the model is not too deep. 1F1B: 👉In 1F1B, the forward and backward passes are interleaved. 👉Once the forward pass is completed for a micro-batch on a particular stage, the backward pass for that micro-batch starts immediately on that stage. 👉This can reduce the memory footprint compared to AFAB, especially for deeper models, as activations can be discarded as soon as they are used in the backward pass. Choosing the right strategy: 👉Model Depth: For shallow models, AFAB can be a good choice due to its simplicity. For deeper models, 1F1B can be more memory-efficient. 👉Memory Constraints: If memory is a major constraint, 1F1B is generally preferred as it reduces the need to store all activations simultaneously. 👉Ease of Implementation: AFAB is generally considered easier to implement. In summary, AFAB and 1F1B are important considerations when training large models with pipeline parallelism in PyTorch. Choosing the right strategy depends on the specific characteristics of your model and hardware. Resources: 👉 https://lnkd.in/eiCwbyND 👉 Seq1F1B paper: https://lnkd.in/eqqxKBRZ Additional content: https://lnkd.in/eR7aZDRT #ml #pipeline #parallelism #AFAB #1F1B #pytorch #datascience #python

13

It's wild how far things have come in just a year. Last year, we saw Voyager, an LLM-powered autonomous AI agent playing Minecraft and now we're simulating entire civilizations in the game. Project Sid from Altera is about seeing what happens when you put 10 to 1,000+ AI agents in the same environment. They form societies, specialize in different roles, make and follow rules, and even share culture and religion. Their PIANO architecture (Parallel Information Aggregation via Neural Orchestration) lets agents do all of this in real-time while keeping things coherent. They’re not just surviving, they’re actually progressing, creating specialized jobs, and figuring out collective rules like taxes. It’s amazing to watch AI agents build their own worlds. Really excited to see where this kind of many-agent simulation takes us next! https://lnkd.in/ePxnwPwE #AgenticSystems #Agents #RAG #LLMs #MultiAgent

12

2 Comments

Moving AI forward safely and transparently! Hedera’s announcement of Verifiable Compute in collaboration with EQTY Lab, Intel Corporation, and NVIDIA is a significant leap for the AI and DLT space. Anchoring AI trust directly in silicon and extending it to the Hedera network demonstrates the potential of merging advanced hardware with distributed ledger technology. As autonomous AI agents become more widespread, ensuring trust and transparency is critical. This initiative sets a strong foundation for secure AI innovation. In your opinion, what’s the next big challenge we need to solve to fully unlock the potential of AI in a safe and transparent way? #artificialIntelligence #DLT #blockchain

1

Revolutionizing Video Editing with AutoVFX: Describe Your Ideas and Watch Them Come to Life We talk about a new system called AutoVFX developed by scientists at UIUC that allows you to edit videos by simply describing what you want to happen. The system creates a 3D model of the scene, uses AI to translate your words into code, and simulates real physics to make your ideas come to life, revolutionizing video creation with Hollywood-level effects. #vfx #ai #filmmaking

3

Artists, Creatives, and anyone in need of something new, we are ready to leave the mundane, the predictable, the 2 dimensional, and disconnected. This Spring 2025, Meloscene isn’t just releasing a spatial computing app, we’re launching the new paradigm for creative workflow, collaboration, and remote living. This platform doesn’t just change how we create, it transforms what we can achieve. We’re redefining the boundaries of technology, creativity, and community with seamless integration into your personal studio and digital/PC setups. A turnkey system that keeps us connected without all the driving, flying, and living in hotels. Signup for Meloscene Beta is now open, so if you're ready to be part of something different, this is it. #Meloscene #VR #SpatialComputing #FutureOfWork #remotework #hybridwork #Mixedreality #RemoteCollaboration #CreativeCommunity #NextGenTech #Meta

7

1 Comment

What exactly does a feature store do, and how does it integrate into a production application? I think this is one of the most common questions we get at Chalk. Here's an excellent article by Melanie Chen that does a really detailed and deep dive into a common use case for a feature store: powering a machine learning system to detect fraud. Check it out!

13

1 Comment