Edmar Mendizabal

Very interesting article. When we say "A.I." it often means "statistical computing feature that outputs a prediction". A.I. represents the future and always will be, as it is a very anthropomorphic notion that evolves in time as we do. Spellcheck is for us a useful feature, while for Gutemberg it would have been called witchcraft. Somewhere in the middle it could have been called an AI proofreader. LLMs predict the next most probable word ("token") based on the distribution of the data it has used to "learn" (actually "set statistical weights"). (1) If we talk about "computing statistical feature" we'll say something like "the result is wrong". But as we talk about A.I. we say things such as "it hallucinates". As such we project beliefs that can feed themselves back. Now the fun part. It is exactly the kind of technique used by psychics. 👇 https://lnkd.in/eMACW4r6 (1) check this very cool demo 👇 https://moebio.com/mind/

24

3 Comments

Navigating the complexities of porting to Apple Vision Pro? Unity's guide offers 6 essential tips to streamline the process, from leveraging Unity's AR Foundation to optimizing your app's performance on Vision Pro. Check out the full guide for expert insights and practical advice to ensure your app shines on Apple's latest platform - https://lnkd.in/dCcxYKem. #AR #VR #AppleVisionPro #Unity #realtime3D

6

Come learn about Xsolla's strategies to improve your mobile games revenue.

89

Will say it again - using game engines as the core logic and adding an AI based "post process" will be huge. We will likely stylize content using data sets provided within proper parameters. Likely be the next phase and completely change how we think of graphics technology. https://lnkd.in/gSFJZNk7

2

Here are three recordings done with the Saramonic Audio SR-VRMIC Ambisonic mic currently retailing for between $90 and $120. All recordings were made using the Zoom F6 and converted to stereo using the free SoundField by RØDE plug-in with some light shelving EQ to tame the plug-ins brightness. They were boosted by between 30 to 50 dB from the original raw 32-bit files to reach suitable loudness without using dynamic range compression. https://lnkd.in/gxTJ4n7c

7

The open source AI community is full of people building all sorts of things, and sharing them with the world. I'm one of those people, though my personal contributions are miniscule in the big picture. Most recently though, I had the idea to merge a version of two of my all-time favorite fine tuned versions of the Mistral AI 7B language model: the Dolphin version and the OpenHermes 2.5 version. Without getting too technical, you can basically create "new" AI models by blending the weights of other, previously fine tuned models. There are a few other details involved, but basically that is how merged models are made. By blending them "just right", you can get a bit of the best of both versions. You could also get the worst, so this can involve some trial and error, also known as time and patience. That's what I did, and when I ran it through a benchmark tool, it spanked any previous models I've built based on that same Mistral model. So, of course I have to share it here as well. I call it "Rudolph Finn Hermes", or RuDolph-Hermes-7B, and it's free to download and mess around with via Hugging Face. Get it here: https://lnkd.in/g_ptG4zg

1

It had been a while I had not been impressed by an AI breakthrough (you know, the "my LLM is bigger than your" race in the tech bro space before they realize size is not everything), but this one got my attention : 👇 https://fugatto.github.io/ As there are few examples linking Audio and instructions to create it in order to train a model, they focus on a dataset generation method. Excerpt from the paper (link in the comment) : "(...) audio data does not inherently contain the instructions that were used to generate it. To overcome this challenge, we introduce a specialized dataset generation approach optimized for producing a wide range of audio generation and transformation tasks, ensuring the data reveals meaningful relationships between audio and language" "We focus on building as large and diverse a dataset in order to collect demonstrations of as many audio tasks and contexts as possible. We emphasize that our ultimate goal is not to just excel on such tasks, but to drive, as a community, towards a future where unsupervised multitask learning emerges from data and model scale." The resulting demo suggests a very versatile audio synthesis and transformation engine. Still, let's be sure it's not "one more tech demo". Thanks Olivier for the link !

20

3 Comments

Microsoft is making waves with Copilot, its AI assistant program targeted for Xbox and PC games. Unveiled at a recent Surface and AI event, Copilot promises to revolutionize in-game assistance. Imagine having a built-in guide in Minecraft! That's what Copilot offers. Players can ask questions like "How do I craft a sword?" and receive real-time guidance on finding the necessary materials. This functionality positions Copilot as an always-on game guide, similar to the PlayStation 5's non-AI powered help system. Minecraft is the first game to integrate Copilot, but Microsoft hints at broader implementation across titles. Whether this initially targets first-party Xbox games remains unclear. 🤔 AI for the Future of Gaming? Copilot's potential impact extends beyond basic assistance. Microsoft envisions AI influencing narrative design, writing, and even customer service through AI chatbots. This integration of AI could fundamentally change how games are created and experienced. However, some gamers question the necessity of Copilot. For obscure game mechanics, an AI assistant might be helpful. However, readily available online guides already fulfill this role for many players. The true value proposition of Copilot, beyond a predetermined solution, remains to be seen. 🌴 Microsoft's commitment to carbon neutrality by 2030 clashes with the energy-intensive nature of AI models. The company's own greenhouse gas emissions rose 30% last year, raising concerns about the environmental cost of this technology. 🤔 Copilot's future remains uncertain. Will it become an indispensable gaming companion, or an eco-conscious nightmare? Only time will tell. (source: https://lnkd.in/etGNMjGu)

13

6 Comments

Learning and practicing new music 🎵with your voice 🎤and instrument 🎷has become easier with Metronaut’s Sight-Read, AI Feedback and AI Tempo modes, available on our entire catalog of 12k+ titles (and counting). 😇🎹🎻

12

Microsoft's Research team just introduced the VASA-1 model for Generating Lifelike Talking Faces from a single image (and audio clip!) 🗣️ The model is able to produce lip movements synchronized with the audio as well as realistic facial expressions and head movements. Key features of VASA-1 include: -Ability to generate 512x512 resolution videos at up to 40 FPS ⚡ -Works with just a single input image and audio clip -Captures a wide range of facial nuances and natural head motions The potential applications for virtual characters and avatars using this technology are going to be pretty extensive! However, they note the technology is still in the research stage with no immediate plans for a public release or API. 🚫 Which if I'm being honest, is probably best at the current time. 😜 Still, quite an advancement! More info can be found on their official research page, here: https://lnkd.in/gYYKzYzn #vfx #visualeffects #technology #ai #research

32

8 Comments

#TBT ... It Runs on the Computer in my Basement ... Exactly 21 years ago, as a final project for a mathematical methods for computation course at Stanford, I figured out how to trick the newly-released ATI 9700 graphics card to simulate cloth… doing the math for a few million springs per second on hardware designed to do no such thing. Yesterday I had a similar feeling - I ran some 80-billion parameter LLMs on my desktop GPU - a ChatBot that passes the Turing Test with flying colors. Moore’s Law is a powerful force. More details (and the paper I wrote in 2003) on the blog: https://lnkd.in/g_NrtedU Re-acknowledging Ian Buck, Josh Bao, Michael Mandel, and Geoffrey Irving who all helped me on that paper in 2003. 

31

4 Comments

Webinar Announcement 👇   🎨 Are your artists and modelers storing and managing assets in generic cloud storage solutions not designed for modern game development? ⏳They likely spend countless hours searching for necessary assets—or worse, they might not even know that the assets they need exist. 🟢🟢🟢If this sounds familiar, we have great news for you! 🟢🟢🟢 💡Asset Manager is an engine-agnostic, API-first solution tailored for the demands of 3D asset creation. For Unity developers, it's already included with your license. 🎯If you're using a different engine, explore Unity Cloud plans to find the one that fits your needs. 🎫Join us for the Asset Manager webinar, featuring a live demo and Q&A session, on August 14th at 3:30 PM UTC. Share this link with your artists and modelers—they'll thank you for it!

7

GPT4o + OUR MARKET EXPERTS’ INPUTS= The future of game ideation is now available on the Supersonic platform Supersonic from Unity Unity Grow

14

SDL2 viewport in a Win32 app. I plan on using this as a base later on for a level editor! #cpp

3

Optimizing XR Performance for Flawless Interactions This blog by QualityReality explores how to optimize XR performance for smooth user experiences. Discover techniques to reduce latency, optimize assets, and leverage 5G for the future of XR! Read here: https://lnkd.in/dUyStzfd #XR #PerformanceOptimization #QualityReality #GameCloud

2

Sometimes it's the most subtle AI announcements that are the biggest deal... Being able to access GitHub repositories directly from Claude now makes Claude an even bigger winner for coding. So much of my coding time is spent copying and pasting examples from github to fix Claude's stale/incorrect recollection of APIs. ...Not to mention the ability to hook this up directly to a database.

8

1 Comment

🔎 Testing Arium on Apple's Safari delivered a high-quality stream in HD with 60fps. 👀 Our focus is on two primary streaming protocols: SRT and WebRTC. Haivision' SRT and W3C's WebRTC are both used for streaming media, but they are designed for different use cases and have distinct features. SRT is designed for low-latency (starting from 20 ms), high-quality video streaming over unreliable networks. It primarily uses UDP for transport, with mechanisms like ARQ (Automatic Repeat reQuest) and FEC (Forward Error Correction) to handle packet loss and ensure reliable delivery, which is a great advantage. WebRTC offers ultra-low latency suitable for real-time interactions. It uses SRTP (Secure Real-time Transport Protocol) for media encryption and DTLS (Datagram Transport Layer Security) for data security. 🎮 Keen to learn more about Arium? Find out at https://arium.gg/about #Testing #Streaming #SRT #WebRTC #CloudGaming #Apple

33

2 Comments

I did another one of the deeper articles for PocketGamer.biz about all of the Supercell Action RPGs, (3 of those 😅) for those that like a little bit more detailed read. Its really fascinating seeing all of the iteration, especially with R.I.S.E. which is a rogue-lite with extraction elements. On the other hand, notice how Supercell is not really going for the "easy" path of making Clash of Clans 2 or Hay Day 2, which it could already, but rather trying to expand the Clash IP over different genres and sub-genres. I think we have seen a lot of diminishing returns of that "more of the same" strategy, especially with Zenless Zone Zero these days, having a rough time not only competing with its portfolio bigger brothers, but also many big Anime RPGs released this summer 😅. #gamedev #gamedevelopment #gamedesign #gameindustry #mobilegames #mobilemarketing #mobileadvertising #userexperiencedesign #userexperience #gamedeveloper #games #gameart #midcore #RPG

19