Digital Twin #3: Rounding Up 2023
A Newsletter on Engineering Simulation, Artificial Intelligence, and the Industrial Metaverse
Welcome to the Siml.ai Newsletter. It is a space of news, events, and curiosities from AI, scientific simulation, CFD, and the industrial metaverse.
Rounding Up 2023
We have prepared a surprise for all our subscribers! By helping us reach more readers, you get a chance to win 300 compute credits in Siml.ai 🎄🎁 Tag us in your post so we can make sure you enter the race!
Revealing Siml.ai v0.4
To finish the year strong, we present Siml.ai version 0.4. In this release, we have focused on introducing data-driven simulators, fixing bugs, and improving performance. Below are the key highlights of this update.
1. Data-driven Simulators
You can now leverage the power of data-driven models within Siml.ai. In addition to physics-based models, our platform is equipped to analyze and learn from user-provided data in CSV format. User has the choice to provide CSV datasets as a separate file or as part of an archive file, which includes zip and tar.gz formats.
2. Simulation Studio Improvements
We reworked the environment to a clean and minimalistic look for better visibility of the simulation data. Simulation Studio is now upgraded to Unreal Engine 5.3, allowing us, in the near future, to implement OpenVDB support to render hundreds of millions of points and also VR/AR capabilities on the web. Added ability to control the transparency of the geometry model, points and vector fields.
3. Enhancements
Hidden Unavailable Tiers:
Unavailable tiers within the environment are now appropriately hidden, streamlining the user experience.
Enhanced Running Cost Widget:
The running cost widget has been reshaped to provide more comprehensive and useful information, offering users greater insights into their simulation costs.
Improved Error Messaging for Training Failure:
Users will now receive informative error messages upon any start/stop training failures, addressing a previous bug where no error message was displayed.
Improved Responsiveness of Plan Limits:
Plan limits now display with improved responsiveness across different screens, providing a consistent user experience.
Clickable Environment and Simulator Card:
Users can now click on the entire environment and simulator card, not just the title, providing a more intuitive and user-friendly interface.
4. Bug Fixes
Corrected Estimated Cost Calculation:
The release addresses an issue where the estimated cost for the environment was being calculated incorrectly. Users can now rely on accurate cost estimations for their simulations.
Monthly Plan Display for Free Tier:
Users on the free tier will now see the appropriate monthly plan displayed, as a bug causing its absence has been fixed.
Prevented 'Train' Button Click During Loading:
A bug that allowed users to click on the 'train' button while a process was loading has been fixed, ensuring proper user interaction.
Resolved Color Contrast Issue on Environment Card:
The release fixes a bug related to the colour contrast of badges on environment cards, ensuring better visibility, especially when the environment is unavailable.
Mobile Compatibility Fixes:
The left sidebar can now be closed seamlessly on mobile devices, addressing an issue that hindered this functionality.
Expanded Simulator Loading:
Users will no longer encounter limitations on the number of loaded simulators, as the release resolves an issue that previously restricted the display.
Hidden Empty Simulators from Others:
Simulators from other users with no content are now appropriately hidden from view, ensuring a cleaner and more focused user interface.
Color Palette Consistency:
Resolved an issue in Simulation Studio where changes to the colour palette were inadvertently affecting all data points. Users can now expect a consistent colour representation of their data.
Optimized Gnomon Position:
Adjusted the gnomon position to the bottom middle to prevent it from being hidden in high DPI edge cases. This enhancement ensures a more reliable and visible orientation reference in the Simulation Studio.
Magnitude Source Selection Fix:
Fixed a bug that ignored the magnitude source selection on a new inference, leading to an incoherent state in the user interface. Users can now rely on accurate and consistent magnitude source selections.
Slicer Glitch Elimination:
Successfully eliminated glitches in the slicer functionality, ensuring a smooth and glitch-free experience during simulation analysis.
Introducing Compute Credits
We have introduced Siml.ai's new compute credits feature for seamless payments and hassle-free top-up. Users can now easily manage credits on the Billing page in real-time, with the flexibility to purchase credits using their preferred mode via Stripe Checkout.
Subscription plans remain, influencing functionality access on Siml.ai based on monthly or yearly plans. Enjoy a quick and secure payment experience, with the option to adjust credit amounts from $10 to $10,000 in a single transaction.
The new compute credits feature offers users a credit estimate based on current usage, providing insight into the credits needed for ongoing tasks. This ensures a more predictable and efficient experience with Siml.ai's training and inference infrastructure.
We made a few simple simulations to demonstrate the capabilities of Siml.ai 💥
Aerodynamics of a fusilli 🍝
The architecture of a fusilli is helical, and its aerodynamics are oddly satisfying. It's no wonder the #cfd community is already playing with the simulations around it. We couldn't resist either - here's a simulation of the aerodynamics of a fusilli made using Navier Stokes equations in Siml.ai. The model was trained on physics inputs into the neural network without any data involved.
Aerodynamics of a cow in a wind tunnel made using Navier Stokes equations 🐄
Have you ever wondered about the aerodynamics of a cow? We have! So, we made a digital twin of a cow in a wind tunnel in Siml.ai. And thanks to SciML and PINNs, once again, no data is required, only physics!
🚘 Aerodynamics of a moving car 🌪
Cars are constantly blasting through a variety of environments and wind tunnels when on the road. Through this type of simulation of the aerodynamics of a car tested with real-life wind speed values, engineers can fine-tune the aerodynamics of their designs and deliver the peak technology by learning from the performance of digital twins in Siml.ai.
We look forward to your feedback.
What we need at this stage of production is for you to report any bugs or issues you encountered while using Siml.ai. The fastest way to not only share your report but also get feedback and a solution is via our Discord community:
Alternatively, you can find us on ProductHunt, GitHub, and e-mail (hello@dimensionlab.org).
Want to try Siml.ai?
These are the plans we offer. They are modular, depending on the size of your organization and the challenges you face.
News from the field
2023 was a strong year for AI. At the end of 2022, we could see a glimpse of GPT. That started an avalanche, which isn't stopping soon. Let's recap what happened in AI in 2023:
👉 OpenAI announced the release of GPT-4 Turbo, a more capable model.
👉 Pika Labs released an AI-powered Text-to-Video platform that allows users to easily and quickly create high-quality videos.
👉 Other notable breakthroughs include DALL·E 3, a model that generates and edits images with natural language prompts. In AI, as in general in the digital world, open-source is essential.
👉 In AI, we could see LLaMA2, an open-source model that can generate long-form text with high coherence and accuracy, and Mistral AI, a Paris-based startup, released Mixtral 8x7B, a "mixture of experts" (MoE) model with open weights that reportedly truly matches OpenAI's GPT-3.5. The Mistral AI is also special because it's an open-source model.
👉 Google brought Bard, which is a chatbot developed by Google, introduced in 2023, and integrated into the Google search engine. It is based on a less powerful version of LaMDA, adapted not to overload the search engine. Bard simulates human conversations using natural language processing and machine learning.
👉 We got a pre-Christmas gift from Google introducing Gemini in December 2023, enabling more advanced reasoning, planning, and understanding. Gemini is their largest and most capable AI model, designed to be multimodal and optimized for three different sizes: Ultra, Pro, and Nano. It reasons seamlessly across text, images, video, audio, and code. Gemini was made available to developers and enterprise customers via the Gemini API in Google AI Studio or Google Cloud Vertex AI. It was integrated into Google products, such as the Bard chatbot, for more advanced reasoning, planning, and understanding.
Read out the annual Quanta Magazine 2023 recap in computer science:
In the landscape of physics, this year, there were also some exciting discoveries and breakthroughs: