Cardboard is crucial for sheet mulching to kill a lawn, but critics say there's no science proving it works. All you really need, they say, is lots of wood chips.

What’s the fastest way to kill your lawn? Science says forget the cardboard

We asked Sundance filmmakers to keep a diary of one day in their working life. Raven Jackson ('All Dirt Roads Taste of Salt') describes screening her film at the New York Film Festival.

How filmmaker Raven Jackson performs the ‘ritual’ of showing ‘All Dirt Roads Taste of Salt’

Find the locations and how to find each Artifact in each cave below. Longitude – 45.4 Latitude – 40.4 Artifact of the Clever is found in a west-facing cave east of The Grand Hills. Continue straight through the cave. Unless there are bends on the main ...

How to find all 10 The Island Artifacts in Ark Survival Ascended

Learning SpaceactivitiesresourcesworkshopsprogramsTeachable MomentsBy Ota LutzExplore how and why the SWOT mission

NASA Mission Takes a Deep Dive Into Earth's Surface Water … – NASA Jet Propulsion Laboratory

Culture.pl presents here part of a collection belonging to Wiktoryn Grąbczewski, a Łęczyce-born and Warsaw-based collector of representations of and legends about the Polish devil. Can the devil be a patriot? Can he cry over a burnt church? A rich area ...

A Polish Halloween

Israeli troops push deeper into northern and central Gaza as the United Nations warns that airstrikes are hitting perilously close to hospitals.

Israel pushes deeper into Gaza and frees soldier held by Hamas; Netanyahu rejects calls for cease-fire

The federally funded Low Income Household Water Assistance Program was set to end this fall, but it will now remain available through March or until funding runs out.

Californians with past-due water bills can get help with payments. Here’s how

Located near the beaches where the majority of players will initially spawn in Ark: Survival Evolved, this clifftop base offers plenty of flat ground to build on and some great views.

Ark: Survival Ascended: Best base locations on The Island

Planned future missions to find water ice on Mars aim to send humans to the planet and find water ice. However, a question that remains to be answered is where exactly on the red planet should they land to do this?

Where exactly on Mars should Humans go? NASA scientists release a map

Have you ever asked yourself, what is a nautical mile? If so, you are not alone. This guide will explain everything cruise passengers should know.

What Is a Nautical Mile and Why Are They Used?

LAT Water

latwater

Authors:
(1) Hoon Kim, Beeble AI, and contributed equally to this work;
(2) Minje Jang, Beeble AI, and contributed equally to this work;
(3) Wonjun Yoon, Beeble AI, and contributed equally to this work;
(4) Jisoo Lee, Beeble AI, and contributed equally to this work;
(5) Donghyun Na, Beeble AI, and contributed equally to this work;
(6) Sanghyun Woo, New York University, and contributed equally to this work. Editor's Note: This is Part 11 of 14 of a study introducing a method for improving how light and shadows can be applied to human portraits in digital images. Read the rest below. Table of Links Abstract and 1. Introduction
2. Related Work
3. SwitchLight and 3.1. Preliminaries
3.2. Problem Formulation
3.3. Architecture
3.4. Objectives
4. Multi-Masked Autoencoder Pre-training
5. Data
6. Experiments
7. Conclusion Appendix A. Implementation Details
B. User Study Interface
C. Video Demonstration
D. Additional Qualitative Results & References A. Implementation Details We pre-train a single U-Net architecture during this process. In the subsequent fine-tuning stage, the weights from this pre-trained model are transferred to multiple U-Nets - NormalNet, DiffuseNet, SpecularNet, and RenderNet. In contrast, IllumNet, which does not follow the U-Net architecture, is initialized with random weights. To ensure compatibility with the varying input channels of each network, we modify the weights as necessary. For example, weights pre-trained for RGB channels are copied and adapted to fit networks with 6 or 9 channels. Data To generate the relighting training pairs, we randomly select each image from the OLAT dataset. Two randomly chosen HDRI lighting environment maps are then projected onto these images to form a training pair. The images undergo processing in linear space. For managing the dynamic range effectively, we apply logarithmic normalization using the log(1 + x) function. Architecture SwitchLight employs a UNet-based architecture, consistently applied across its Normal Net, Diffuse Net, Specular Net, and Render Net. This approach is inspired by recent advancements in diffusion-based models [12]. Unlike standard diffusion methods, we omit the temporal embedding layer. The architecture is characterized by several hyperparameters: the number of input channels, a base channel, and channel multipliers that determine the channel count at each stage. Each downsampling stage features two residual blocks, with attention mechanisms integrated at certain resolutions. The key hyperparameters and their corresponding values are summarized in Table. 4. IllumNet is composed of two projection layers, one for transforming the Phong lobe features and another for image features, with the latter using normal bottleneck features as a compact form of image representation. Following this, a cross-attention layer is employed, wherein the Phong lobe serves as the query and the image features function as both key and value. Finally, an output layer generates the final convolved source HDRI. The Discriminator network is utilized during both pretraining and fine-tuning stages, maintaining the same architectural design, although the weights are not shared between these stages. This network is composed of a series of residual blocks, each containing two 3×3 convolution layers, interspersed with Leaky ReLU activations. The number of filters progressively increases across these layers: 64, 128, 256, and 512. Correspondingly, as the channel filter count increases, the resolution of the features decreases, and finally, the network compresses its output with a 3x3 convolution into a single channel, yielding a probability value. Regarding the activation functions across different networks: NormalNet processes its outputs through ℓ2 normalization, ensuring they are unit normal vectors. IllumNet, DiffuseNet, and RenderNet utilize a softplus activation (with β = 20) to generate non-negative pixel values. SpecularNet employs a sigmoid activation fuction, ensuring that both the roughness parameter and Fresnel reflectance values fall within a range of 0 to 1. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Hoon Kim, Beeble AI, and contributed equally to this work; (2) Minje Jang, Beeble AI, and contributed equally to this work; (3) Wonjun Yoon, Beeble AI, and contributed equally to this work; (4) Jisoo Lee, Beeble AI, and contributed equally to this work; (5) Donghyun Na, Beeble AI, and contributed equally to this work; (6) Sanghyun Woo, New York University, and contributed equally to this work. Authors: Authors: (1) Hoon Kim, Beeble AI, and contributed equally to this work; (2) Minje Jang, Beeble AI, and contributed equally to this work; (3) Wonjun Yoon, Beeble AI, and contributed equally to this work; (4) Jisoo Lee, Beeble AI, and contributed equally to this work; (5) Donghyun Na, Beeble AI, and contributed equally to this work; (6) Sanghyun Woo, New York University, and contributed equally to this work. Editor's Note: This is Part 11 of 14 of a study introducing a method for improving how light and shadows can be applied to human portraits in digital images. Read the rest below. Editor's Note: This is Part 11 of 14 of a study introducing a method for improving how light and shadows can be applied to human portraits in digital images. Read the rest below. Editor's Note: This is Part 11 of 14 of a study introducing a method for improving how light and shadows can be applied to human portraits in digital images. Read the rest below. Editor's Note: This is Part 11 of 14 of a study introducing a method for improving how light and shadows can be applied to human portraits in digital images. Read the rest below. Table of Links Abstract and 1. Introduction 2. Related Work 3. SwitchLight and 3.1. Preliminaries 3.2. Problem Formulation 3.3. Architecture 3.4. Objectives 4. Multi-Masked Autoencoder Pre-training 5. Data 6. Experiments 7. Conclusion Abstract and 1. Introduction Abstract and 1. Introduction 2. Related Work 2. Related Work 3. SwitchLight and 3.1. Preliminaries 3. SwitchLight and 3.1. Preliminaries 3.2. Problem Formulation 3.2. Problem Formulation 3.3. Architecture 3.3. Architecture 3.4. Objectives 3.4. Objectives 4. Multi-Masked Autoencoder Pre-training 4. Multi-Masked Autoencoder Pre-training 5. Data 5. Data 6. Experiments 6. Experiments 7. Conclusion 7. Conclusion Appendix A. Implementation Details B. User Study Interface C. Video Demonstration D. Additional Qualitative Results & References A. Implementation Details A. Implementation Details B. User Study Interface B. User Study Interface C. Video Demonstration C. Video Demonstration D. Additional Qualitative Results & References D. Additional Qualitative Results & References A. Implementation Details We pre-train a single U-Net architecture during this process. In the subsequent fine-tuning stage, the weights from this pre-trained model are transferred to multiple U-Nets - NormalNet, DiffuseNet, SpecularNet, and RenderNet. In contrast, IllumNet, which does not follow the U-Net architecture, is initialized with random weights. To ensure compatibility with the varying input channels of each network, we modify the weights as necessary. For example, weights pre-trained for RGB channels are copied and adapted to fit networks with 6 or 9 channels. Data To generate the relighting training pairs, we randomly select each image from the OLAT dataset. Two randomly chosen HDRI lighting environment maps are then projected onto these images to form a training pair. The images undergo processing in linear space. For managing the dynamic range effectively, we apply logarithmic normalization using the log(1 + x) function. Data Architecture SwitchLight employs a UNet-based architecture, consistently applied across its Normal Net, Diffuse Net, Specular Net, and Render Net. This approach is inspired by recent advancements in diffusion-based models [12]. Unlike standard diffusion methods, we omit the temporal embedding layer. The architecture is characterized by several hyperparameters: the number of input channels, a base channel, and channel multipliers that determine the channel count at each stage. Each downsampling stage features two residual blocks, with attention mechanisms integrated at certain resolutions. The key hyperparameters and their corresponding values are summarized in Table. 4. Architecture IllumNet is composed of two projection layers, one for transforming the Phong lobe features and another for image features, with the latter using normal bottleneck features as a compact form of image representation. Following this, a cross-attention layer is employed, wherein the Phong lobe serves as the query and the image features function as both key and value. Finally, an output layer generates the final convolved source HDRI. IllumNet The Discriminator network is utilized during both pretraining and fine-tuning stages, maintaining the same architectural design, although the weights are not shared between these stages. This network is composed of a series of residual blocks, each containing two 3×3 convolution layers, interspersed with Leaky ReLU activations. The number of filters progressively increases across these layers: 64, 128, 256, and 512. Correspondingly, as the channel filter count increases, the resolution of the features decreases, and finally, the network compresses its output with a 3x3 convolution into a single channel, yielding a probability value. Discriminator Regarding the activation functions across different networks: NormalNet processes its outputs through ℓ2 normalization, ensuring they are unit normal vectors. IllumNet, DiffuseNet, and RenderNet utilize a softplus activation (with β = 20) to generate non-negative pixel values. SpecularNet employs a sigmoid activation fuction, ensuring that both the roughness parameter and Fresnel reflectance values fall within a range of 0 to 1. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. available on arxiv

AI System Uses Teamwork to Create Picture-Perfect Lighting

Telehealth has reshaped the medical sector. Ever since the COVID-19 pandemic, platforms enabling remote access to healthcare services have skyrocketed, and they’re likely not going away any time soon. The telemedicine market was worth almost $72 billion in 2022, and experts predict it will surpass $244 billion by 2032. What does all this growth mean for patients and providers involved in the medical field? While some factors are uncertain, you can expect five telehealth trends in 2025 and beyond. 1. Increased AI Integration Like many sectors, telehealth will see several changes from artificial intelligence (AI) in the coming years. The medical industry in general will benefit from AI integration, but the efficiency-focused and data-driven field of telemedicine is particularly prone to disruption. Virtual nursing assistants are one of the biggest areas of growth, predicted to account for 26.4% of all revenue within the AI of the telehealth market. These advanced chatbots can talk to patients about their symptoms before connecting them to a human doctor, ensuring everyone gets immediate help around the clock. Some may also make notes for doctors to review for faster, more accurate diagnoses. Other companies may use AI to monitor patient data for insight into developing conditions. Such analysis would expand personalized care, which typically leads to better health outcomes. While medical AI still has a long way to go before it reaches its full potential, 2025 could see significant progress in this field. 2. Greater Support for Wearables Wearables and accompanying telemedicine apps are expected to take several steps forward in 2025. Telehealth and wearables pair together naturally, as the latter provides real-time data to improve the former’s offerings. This market may be on the verge of significant expansion thanks to recent breakthroughs. In May 2024, the Apple Watch became the first digital device to gain FDA clearance as a Medical Device Development Tool (MDDT). Organizations can use MDDTs in clinical trials. Now that a consumer-grade device meets this standard, telehealth and related wearables could play a key role in ongoing research. Apple receiving FDA approval further cements wearables’ usefulness as medical devices, making advanced telemedicine services more accessible. 3. Legal Challenges to Expanded Coverage While some areas of telehealth have a bright future in the next year, others face rising uncertainty. Several pre-pandemic Medicare limitations will take effect again in 2025, making telemedicine coverage less accessible for some patients. Most notably, Medicare will no longer cover telehealth services outside of rural areas or zones with a shortage of health professionals. Thankfully, these patients — who often need telemedicine the most — will retain coverage. For others, many of whom prefer virtual care even when in-person visits are possible, the regulatory change may limit their options. Mental health care has no geographic limits, and private insurers may still offer support, but these restrictions may hinder the explosive growth telehealth platforms have enjoyed over the past few years. 4. Continued Growth for Telepharmacy Applications As coverage for general telehealth applications wanes, telepharmacy support is expanding. Telepharmacy deals with remotely prescribing and distributing medication. This small but significant part of the industry will reach new heights in 2025. In November 2024, the Drug Enforcement Administration expanded current telepharmacy flexibilities, allowing doctors to digitally prescribe controlled substances through December 31, 2025. It’s the third consecutive year of such extensions, painting a promising future for remote pharmaceutical systems. Many people need prescription medications, but fewer can access or feel comfortable going to a pharmacy. Expanding the ability for providers to use telecommunications apps for this purpose means more patients could easily obtain their supplies. The shift particularly benefits fast-growing medication segments like GLP-1 weight loss treatments. 5. Growing Emphasis on Privacy and Security Finally, the telehealth sector will renew its emphasis on cybersecurity and data privacy. 2024 was a challenging year for healthcare security, peaking with the Change Healthcare ransomware attack in February, which directly impacted thousands of patients in the U.S. After such a massive incident, standards will likely rise for digital health platforms. Whether new legislation tightens cybersecurity controls is unclear, as legal changes take time. However, social and investor pressure to secure telehealth platforms will almost certainly rise. Telemedicine companies can differentiate themselves in 2025 and beyond by citing advanced cybersecurity controls. As more businesses pursue this, defense standards across the industry will heighten. Telehealth Faces Significant Changes in 2025 and Beyond The coming year has much in store for telehealth providers and patients. Many of these changes are beneficial. Others will prove challenging. Regardless of the outcome, one thing is certain — telemedicine tomorrow will look different than today. Both challenges and opportunities present chances for the telehealth sector to mature. As companies navigate changing shifts in 2025, they could take healthcare to new heights. Telehealth has reshaped the medical sector. Ever since the COVID-19 pandemic, platforms enabling remote access to healthcare services have skyrocketed, and they’re likely not going away any time soon. Telehealth has reshaped the medical sector. Ever since the COVID-19 pandemic, platforms enabling remote access to healthcare services have skyrocketed, and they’re likely not going away any time soon. The telemedicine market was worth almost $72 billion in 2022 , and experts predict it will surpass $244 billion by 2032. What does all this growth mean for patients and providers involved in the medical field? While some factors are uncertain, you can expect five telehealth trends in 2025 and beyond. $72 billion in 2022 $72 billion in 2022 1. Increased AI Integration Like many sectors, telehealth will see several changes from artificial intelligence (AI) in the coming years. The medical industry in general will benefit from AI integration, but the efficiency-focused and data-driven field of telemedicine is particularly prone to disruption. Virtual nursing assistants are one of the biggest areas of growth, predicted to account for 26.4% of all revenue within the AI of the telehealth market. These advanced chatbots can talk to patients about their symptoms before connecting them to a human doctor, ensuring everyone gets immediate help around the clock. Some may also make notes for doctors to review for faster, more accurate diagnoses. 26.4% of all revenue 26.4% of all revenue Other companies may use AI to monitor patient data for insight into developing conditions. Such analysis would expand personalized care, which typically leads to better health outcomes. While medical AI still has a long way to go before it reaches its full potential, 2025 could see significant progress in this field. 2. Greater Support for Wearables Greater Support for Wearables Wearables and accompanying telemedicine apps are expected to take several steps forward in 2025. Telehealth and wearables pair together naturally, as the latter provides real-time data to improve the former’s offerings. This market may be on the verge of significant expansion thanks to recent breakthroughs. In May 2024, the Apple Watch became the first digital device to gain FDA clearance as a Medical Device Development Tool (MDDT). the first digital device the first digital device Organizations can use MDDTs in clinical trials. Now that a consumer-grade device meets this standard, telehealth and related wearables could play a key role in ongoing research. Apple receiving FDA approval further cements wearables’ usefulness as medical devices, making advanced telemedicine services more accessible. 3. Legal Challenges to Expanded Coverage While some areas of telehealth have a bright future in the next year, others face rising uncertainty. Several pre-pandemic Medicare limitations will take effect again in 2025 , making telemedicine coverage less accessible for some patients. will take effect again in 2025 will take effect again in 2025 Most notably, Medicare will no longer cover telehealth services outside of rural areas or zones with a shortage of health professionals. Thankfully, these patients — who often need telemedicine the most — will retain coverage. For others, many of whom prefer virtual care even when in-person visits are possible, the regulatory change may limit their options. many of whom prefer virtual care many of whom prefer virtual care Mental health care has no geographic limits, and private insurers may still offer support, but these restrictions may hinder the explosive growth telehealth platforms have enjoyed over the past few years. 4. Continued Growth for Telepharmacy Applications Continued Growth for Telepharmacy Applications As coverage for general telehealth applications wanes, telepharmacy support is expanding. Telepharmacy deals with remotely prescribing and distributing medication. This small but significant part of the industry will reach new heights in 2025. In November 2024, the Drug Enforcement Administration expanded current telepharmacy flexibilities , allowing doctors to digitally prescribe controlled substances through December 31, 2025. It’s the third consecutive year of such extensions, painting a promising future for remote pharmaceutical systems. expanded current telepharmacy flexibilities expanded current telepharmacy flexibilities Many people need prescription medications, but fewer can access or feel comfortable going to a pharmacy. Expanding the ability for providers to use telecommunications apps for this purpose means more patients could easily obtain their supplies. The shift particularly benefits fast-growing medication segments like GLP-1 weight loss treatments. 5. Growing Emphasis on Privacy and Security Finally, the telehealth sector will renew its emphasis on cybersecurity and data privacy. 2024 was a challenging year for healthcare security, peaking with the Change Healthcare ransomware attack in February, which directly impacted thousands of patients in the U.S. After such a massive incident, standards will likely rise for digital health platforms. directly impacted thousands of patients directly impacted thousands of patients Whether new legislation tightens cybersecurity controls is unclear, as legal changes take time. However, social and investor pressure to secure telehealth platforms will almost certainly rise. Telemedicine companies can differentiate themselves in 2025 and beyond by citing advanced cybersecurity controls. As more businesses pursue this, defense standards across the industry will heighten. Telehealth Faces Significant Changes in 2025 and Beyond The coming year has much in store for telehealth providers and patients. Many of these changes are beneficial. Others will prove challenging. Regardless of the outcome, one thing is certain — telemedicine tomorrow will look different than today. Both challenges and opportunities present chances for the telehealth sector to mature. As companies navigate changing shifts in 2025, they could take healthcare to new heights.

What to Expect From Telehealth in 2025

AI That Remembers

I've been the happy owner of several Philips Hue-connected lights for a few years. Some of them are coloured, some of them regular. In addition, I bought a sensor to go along with the light I installed in my toilets: it turns on automatically when it detects a movement there. In this post, I want to document how I replaced the proprietary automation with Home Assistant's. The current Philips Hue automation I mentioned the light and the sensor in the introduction. To manage both directly or via Google Home, you need an additional component - the Hue Hub. The Hub needs an Ethernet connection, but you can't connect it to Wifi. The next step is to register the device. Installing the dedicated Philips Hue app on your phone or tablet is the first step. Once installed, you can proceed to register the Hub itself. You must then register the Hub itself. You must long push the button on top of the hub, probably for security reasons. If everything goes well, as it did for me, you should see your devices on the app. The process goes the same for the motion sensor. After registering it, you can configure it to turn on the light when it detects motion. Even better, you can set the light's intensity (and colour if it's a coloured bulb) to adapt to the time of the day. Indeed, it's better to have a low-intensity reddish light at night to keep your melatonin. Migrating to Home Assistant We will migrate step-by-step, as I did, to show you how straightforward and manageable it is. The first mandatory step is to add the Philips Hue integration. Integrating Philips Hue Go to the left Settings menu, then select the Devices & services item on the screen. The opening window displays four tabs: Integrations, Devices, Entities, and Helpers. The default tab is Integrations: click the Add Integration button at the bottom left corner. Select Philips > Hue, then configure the IP of the Hue Hub. After Home Assistant has detected the latter, it will display it and all associated devices. Note that you can add multiple hubs, even though one hub can manage (a bit more than) fifty connected devices. At this point, Home Assistant should display every device added to your Hub. At this point, the devices appear in your default dashboard overview. I'll probably write about it in a later post. Suffice it to say that the default dashboard overview automatically displays every device added. We can now turn on the lights via Home Assistant's UI. Binding the light to the sensor Before binding the light to the sensor in Home Assistant, we must unbind it in the proprietary Hub. Go to the Settings > Accessories menu in the Philips app. Select the motion sensor and click the Configure in another app button. You can ensure it's unbound if the sensor now appears in the previous screen with a Not configured in this app label. Additionally, the slider should be disabled. We are now ready to bind the light to the sensor in Home Assistant via an Automation. I mentioned Automation objects in the previous post on concepts. As a reminder, an Automation consists of three components: A when condition, the event that triggers the automation
A then clause, what to do when the event happens
An optional if condition. Each component can consist of one or multiple components assembled via boolean logic. It's time to dive a bit deeper. You can create an automation via the UI by going to the Settings > Automations & scenes menu. It brings you to a screen with three tabs; Automations is the default. Click on the bottom right Create Automation. The Automation is so standard that it has a dedicated menu item! The non-generic menu items use a Blueprint. I'll keep the explanation of Blueprint objects for a later post; suffice to say for now that it's akin to a template. It's pretty straightforward to configure it. The Blueprint is already selected. You must choose your motion sensor and the light(s) you want to turn on. The UI lists all available choices for the former. For the latter, you can choose a single light Device or a group in an Area or via a Label. The final parameter is self-describing: "Time to leave the light on after last motion is detected." Click on Save. Congratulations, you've successfully implemented your first automation! It's a significant step in your journey with Home Assistant, and it's just the beginning of what you can achieve. When the sensor detects a motion, it sends the event to Home Assistant. The latter sees there's a related automation and turns on the light. It also starts a timer and resets it every time it receives an event from the sensor. When it reaches zero, it turns the light off. You can achieve this by directly writing the YAML configuration. Locate the file automation.yaml file; by default, it should be under the /config folder. It displays the newly created automation: - id: '1719829445000'
  alias: Turn on the toilet light
  description: Turn on the toilet's light when motion is detected
  use_blueprint:
    path: homeassistant/motion_light.yaml
    input:
      motion_entity: binary_sensor.motion_sensor_toilettes_motion
      light_target:
        device_id: 0bd1f534865dae40ae66b805c9b2c499
      no_motion_wait: 240 You can update the YAML snippet, but creating a new automation entry in YAML doesn't reflect in the UI. We replaced the proprietary Hub automation with the Home Assistant one. Yet, the replacement was not on feature parity: we miss the light brightness! During the night, the brightness is the same as during the day, whereas we configured it differently in the previous setup. Conclusion In this, we moved from the proprietary Philips Hue automation between a sensor and a light to one in Home Assistant. It's a common scenario, so Home Assistant makes it straightforward via a Blueprint. Yet, in the end, we missed a crucial feature: brightness adjustment based on the time. In our next post, we are going to address this gap. To go further: Home Assistant
Philips Hue
Emulated Hue Originally published at A Java Geek on December 8th, 2024 I've been the happy owner of several Philips Hue-connected lights for a few years. Some of them are coloured, some of them regular. In addition, I bought a sensor to go along with the light I installed in my toilets: it turns on automatically when it detects a movement there. In this post, I want to document how I replaced the proprietary automation with Home Assistant's. The current Philips Hue automation I mentioned the light and the sensor in the introduction. To manage both directly or via Google Home, you need an additional component - the Hue Hub. The Hub needs an Ethernet connection, but you can't connect it to Wifi. need The next step is to register the device. Installing the dedicated Philips Hue app on your phone or tablet is the first step. Once installed, you can proceed to register the Hub itself. You must then register the Hub itself. You must long push the button on top of the hub, probably for security reasons. If everything goes well, as it did for me, you should see your devices on the app. The process goes the same for the motion sensor. After registering it, you can configure it to turn on the light when it detects motion. Even better, you can set the light's intensity (and colour if it's a coloured bulb) to adapt to the time of the day. Indeed, it's better to have a low-intensity reddish light at night to keep your melatonin . melatonin Migrating to Home Assistant We will migrate step-by-step, as I did, to show you how straightforward and manageable it is. The first mandatory step is to add the Philips Hue integration. Integrating Philips Hue Go to the left Settings menu, then select the Devices & services item on the screen. The opening window displays four tabs: Integrations , Devices , Entities , and Helpers . The default tab is Integrations : click the Add Integration button at the bottom left corner. Settings Devices & services Integrations Devices Entities Helpers Integrations Add Integration Select Philips > Hue , then configure the IP of the Hue Hub. After Home Assistant has detected the latter, it will display it and all associated devices. Philips > Hue Note that you can add multiple hubs, even though one hub can manage (a bit more than) fifty connected devices. At this point, Home Assistant should display every device added to your Hub. At this point, the devices appear in your default dashboard overview. I'll probably write about it in a later post. Suffice it to say that the default dashboard overview automatically displays every device added. We can now turn on the lights via Home Assistant's UI. Binding the light to the sensor Before binding the light to the sensor in Home Assistant, we must unbind it in the proprietary Hub. Go to the Settings > Accessories menu in the Philips app. Select the motion sensor and click the Configure in another app button. You can ensure it's unbound if the sensor now appears in the previous screen with a Not configured in this app label. Additionally, the slider should be disabled. Settings > Accessories Configure in another app Not configured in this app We are now ready to bind the light to the sensor in Home Assistant via an Automation . I mentioned Automation objects in the previous post on concepts . As a reminder, an Automation consists of three components: in Home Assistant Automation Automation previous post on concepts Automation A when condition, the event that triggers the automation A then clause, what to do when the event happens An optional if condition. A when condition, the event that triggers the automation when A then clause, what to do when the event happens then An optional if condition. if Each component can consist of one or multiple components assembled via boolean logic. It's time to dive a bit deeper. You can create an automation via the UI by going to the Settings > Automations & scenes menu. It brings you to a screen with three tabs; Automations is the default. Click on the bottom right Create Automation . The Automation is so standard that it has a dedicated menu item! Settings > Automations & scenes Automations Create Automation Automation The non-generic menu items use a Blueprint . I'll keep the explanation of Blueprint objects for a later post; suffice to say for now that it's akin to a template . Blueprint Blueprint template It's pretty straightforward to configure it. The Blueprint is already selected. You must choose your motion sensor and the light(s) you want to turn on. The UI lists all available choices for the former. For the latter, you can choose a single light Device or a group in an Area or via a Label . The final parameter is self-describing: "Time to leave the light on after last motion is detected." Click on Save . Blueprint Device Area Label Save Congratulations, you've successfully implemented your first automation! It's a significant step in your journey with Home Assistant, and it's just the beginning of what you can achieve. When the sensor detects a motion, it sends the event to Home Assistant. The latter sees there's a related automation and turns on the light. It also starts a timer and resets it every time it receives an event from the sensor. When it reaches zero, it turns the light off. You can achieve this by directly writing the YAML configuration. Locate the file automation.yaml file; by default, it should be under the /config folder. It displays the newly created automation: automation.yaml /config - id: '1719829445000'
  alias: Turn on the toilet light
  description: Turn on the toilet's light when motion is detected
  use_blueprint:
    path: homeassistant/motion_light.yaml
    input:
      motion_entity: binary_sensor.motion_sensor_toilettes_motion
      light_target:
        device_id: 0bd1f534865dae40ae66b805c9b2c499
      no_motion_wait: 240 - id: '1719829445000'
  alias: Turn on the toilet light
  description: Turn on the toilet's light when motion is detected
  use_blueprint:
    path: homeassistant/motion_light.yaml
    input:
      motion_entity: binary_sensor.motion_sensor_toilettes_motion
      light_target:
        device_id: 0bd1f534865dae40ae66b805c9b2c499
      no_motion_wait: 240 You can update the YAML snippet, but creating a new automation entry in YAML doesn't reflect in the UI. We replaced the proprietary Hub automation with the Home Assistant one. Yet, the replacement was not on feature parity: we miss the light brightness! During the night, the brightness is the same as during the day, whereas we configured it differently in the previous setup. Conclusion In this, we moved from the proprietary Philips Hue automation between a sensor and a light to one in Home Assistant. It's a common scenario, so Home Assistant makes it straightforward via a Blueprint . Yet, in the end, we missed a crucial feature: brightness adjustment based on the time. In our next post, we are going to address this gap. Blueprint To go further: To go further: Home Assistant Philips Hue Emulated Hue Home Assistant Home Assistant Philips Hue Philips Hue Emulated Hue Emulated Hue Originally published at A Java Geek on December 8th, 2024 Originally published at A Java Geek on December 8th, 2024 A Java Geek

Gain More Control Over Your Smart Bulbs With This One Simple Handy Tool

Authors:
(1) Albert Gu, Machine Learning Department, Carnegie Mellon University and with equal contribution;
(2) Tri Dao, Department of Computer Science, Princeton University and with equal contribution. Table of Links Abstract and 1 Introduction 2 State Space Models 3 Selective State Space Models and 3.1 Motivation: Selection as a Means of Compression 3.2 Improving SSMs with Selection 3.3 Efficient Implementation of Selective SSMs 3.4 A Simplified SSM Architecture 3.5 Properties of Selection Mechanisms 3.6 Additional Model Details 4 Empirical Evaluation and 4.1 Synthetic Tasks 4.2 Language Modeling 4.3 DNA Modeling 4.4 Audio Modeling and Generation 4.5 Speed and Memory Benchmarks 4.6 Model Ablations 5 Discussion 6 Conclusion and References A Discussion: Selection Mechanism B Related Work C Mechanics of Selective SSMs D Hardware-aware Algorithm For Selective SSMs E Experimental Details and Additional Results 3.3 Efficient Implementation of Selective SSMs Hardware-friendly architectures such as convolutions (Krizhevsky, Sutskever, and Hinton 2012) and Transformers (Vaswani et al. 2017) enjoy widespread application. Here we aim to make selective SSMs efficient on modern hardware (GPU) as well. The selection mechanism is quite natural, and earlier works attempted to incorporate special cases of selection, such as letting ∆ vary over time in recurrent SSMs (Gu, Dao, et al. 2020). However, as previously mentioned a core limitation in the usage of SSMs is their computational efficiency, which was why S4 and all derivatives used LTI (non-selective) models, most commonly in the form of global convolutions. 3.3.1 Motivation of Prior Models We first revisit this motivation and overview our approach to overcome limitations of prior methods. • At a high level, recurrent models such as SSMs always balance a tradeoff between expressivity and speed: as discussed in Section 3.1, models with larger hidden state dimension should be more effective but slower. Thus we want to maximize hidden state dimension without paying speed and memory costs. • Note that the recurrent mode is more flexible than the convolution mode, since the latter (3) is derived from expanding the former (2) (Gu, Goel, and Ré 2022; Gu, Johnson, Goel, et al. 2021). However, this would require computing and materializing the latent state ℎ with shape (B, L, D, N), much larger (by a factor of N, the SSM state dimension) than the input x and output y of shape (B, L, D). Thus the more efficient convolution mode was introduced which could bypass the state computation and materializes a convolution kernel (3a) of only (B, L, D). • Prior LTI SSMs leverage the dual recurrent-convolutional forms to increase the effective state dimension by a factor of Nx (≈ 10 − 100), much larger than traditional RNNs, without efficiency penalties. 3.3.2 Overview of Selective Scan: Hardware-Aware State Expansion The selection mechanism is designed to overcome the limitations of LTI models; at the same time, we therefore need to revisit the computation problem of SSMs. We address this with three classical techniques: kernel fusion, parallel scan, and recomputation. We make two main observations: • The naive recurrent computation uses O(BLDN) FLOPs while the convolutional computation uses O(BLD log(L)) FLOPs, and the former has a lower constant factor. Thus for long sequences and not-too-large state dimension N, the recurrent mode can actually use fewer FLOPs. • The two challenges are the sequential nature of recurrence, and the large memory usage. To address the latter, just like the convolutional mode, we can attempt to not actually materialize the full state ℎ. The main idea is to leverage properties of modern accelerators (GPUs) to materialize the state ℎ only in more efficient levels of the memory hierarchy. In particular, most operations (except matrix multiplication) are bounded by memory bandwidth (Dao, Fu, Ermon, et al. 2022; Ivanov et al. 2021; Williams, Waterman, and Patterson 2009). This includes our scan operation, and we use kernel fusion to reduce the amount of memory IOs, leading to a significant speedup compared to a standard implementation. To avoid the sequential recurrence, we observe that despite not being linear it can still be parallelized with a work-efficient parallel scan algorithm (Blelloch 1990; Martin and Cundy 2018; Smith, Warrington, and Linderman 2023). Finally, we must also avoid saving the intermediate states, which are necessary for backpropagation. We carefully apply the classic technique of recomputation to reduce the memory requirements: the intermediate states are not stored but recomputed in the backward pass when the inputs are loaded from HBM to SRAM. As a result, the fused selective scan layer has the same memory requirements as an optimized transformer implementation with FlashAttention. Details of the fused kernel and recomputation are in Appendix D. The full Selective SSM layer and algorithm is illustrated in Figure 1. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Albert Gu, Machine Learning Department, Carnegie Mellon University and with equal contribution; (2) Tri Dao, Department of Computer Science, Princeton University and with equal contribution. Authors: Authors: (1) Albert Gu, Machine Learning Department, Carnegie Mellon University and with equal contribution; (2) Tri Dao, Department of Computer Science, Princeton University and with equal contribution. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 State Space Models 2 State Space Models 3 Selective State Space Models and 3.1 Motivation: Selection as a Means of Compression 3 Selective State Space Models and 3.1 Motivation: Selection as a Means of Compression 3.2 Improving SSMs with Selection 3.2 Improving SSMs with Selection 3.3 Efficient Implementation of Selective SSMs 3.3 Efficient Implementation of Selective SSMs 3.4 A Simplified SSM Architecture 3.4 A Simplified SSM Architecture 3.5 Properties of Selection Mechanisms 3.5 Properties of Selection Mechanisms 3.6 Additional Model Details 3.6 Additional Model Details 4 Empirical Evaluation and 4.1 Synthetic Tasks 4 Empirical Evaluation and 4.1 Synthetic Tasks 4.2 Language Modeling 4.2 Language Modeling 4.3 DNA Modeling 4.3 DNA Modeling 4.4 Audio Modeling and Generation 4.4 Audio Modeling and Generation 4.5 Speed and Memory Benchmarks 4.5 Speed and Memory Benchmarks 4.6 Model Ablations 4.6 Model Ablations 5 Discussion 5 Discussion 6 Conclusion and References 6 Conclusion and References A Discussion: Selection Mechanism A Discussion: Selection Mechanism B Related Work B Related Work C Mechanics of Selective SSMs C Mechanics of Selective SSMs D Hardware-aware Algorithm For Selective SSMs D Hardware-aware Algorithm For Selective SSMs E Experimental Details and Additional Results E Experimental Details and Additional Results 3.3 Efficient Implementation of Selective SSMs Hardware-friendly architectures such as convolutions (Krizhevsky, Sutskever, and Hinton 2012) and Transformers (Vaswani et al. 2017) enjoy widespread application. Here we aim to make selective SSMs efficient on modern hardware (GPU) as well. The selection mechanism is quite natural, and earlier works attempted to incorporate special cases of selection, such as letting ∆ vary over time in recurrent SSMs (Gu, Dao, et al. 2020). However, as previously mentioned a core limitation in the usage of SSMs is their computational efficiency, which was why S4 and all derivatives used LTI (non-selective) models, most commonly in the form of global convolutions. 3.3.1 Motivation of Prior Models 3.3.1 Motivation of Prior Models We first revisit this motivation and overview our approach to overcome limitations of prior methods. • At a high level, recurrent models such as SSMs always balance a tradeoff between expressivity and speed: as discussed in Section 3.1, models with larger hidden state dimension should be more effective but slower. Thus we want to maximize hidden state dimension without paying speed and memory costs. • Note that the recurrent mode is more flexible than the convolution mode, since the latter (3) is derived from expanding the former (2) (Gu, Goel, and Ré 2022; Gu, Johnson, Goel, et al. 2021). However, this would require computing and materializing the latent state ℎ with shape (B, L, D, N), much larger (by a factor of N, the SSM state dimension) than the input x and output y of shape (B, L, D). Thus the more efficient convolution mode was introduced which could bypass the state computation and materializes a convolution kernel (3a) of only (B, L, D). • Prior LTI SSMs leverage the dual recurrent-convolutional forms to increase the effective state dimension by a factor of Nx (≈ 10 − 100), much larger than traditional RNNs, without efficiency penalties. 3.3.2 Overview of Selective Scan: Hardware-Aware State Expansion 3.3.2 Overview of Selective Scan: Hardware-Aware State Expansion The selection mechanism is designed to overcome the limitations of LTI models; at the same time, we therefore need to revisit the computation problem of SSMs. We address this with three classical techniques: kernel fusion, parallel scan, and recomputation. We make two main observations: • The naive recurrent computation uses O(BLDN) FLOPs while the convolutional computation uses O(BLD log(L)) FLOPs, and the former has a lower constant factor. Thus for long sequences and not-too-large state dimension N, the recurrent mode can actually use fewer FLOPs. • The two challenges are the sequential nature of recurrence, and the large memory usage. To address the latter, just like the convolutional mode, we can attempt to not actually materialize the full state ℎ. The main idea is to leverage properties of modern accelerators (GPUs) to materialize the state ℎ only in more efficient levels of the memory hierarchy. In particular, most operations (except matrix multiplication) are bounded by memory bandwidth (Dao, Fu, Ermon, et al. 2022; Ivanov et al. 2021; Williams, Waterman, and Patterson 2009). This includes our scan operation, and we use kernel fusion to reduce the amount of memory IOs, leading to a significant speedup compared to a standard implementation. To avoid the sequential recurrence, we observe that despite not being linear it can still be parallelized with a work-efficient parallel scan algorithm (Blelloch 1990; Martin and Cundy 2018; Smith, Warrington, and Linderman 2023). Finally, we must also avoid saving the intermediate states, which are necessary for backpropagation. We carefully apply the classic technique of recomputation to reduce the memory requirements: the intermediate states are not stored but recomputed in the backward pass when the inputs are loaded from HBM to SRAM. As a result, the fused selective scan layer has the same memory requirements as an optimized transformer implementation with FlashAttention. Details of the fused kernel and recomputation are in Appendix D. The full Selective SSM layer and algorithm is illustrated in Figure 1. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Cutting-Edge Techniques That Speed Up AI Without Extra Costs

Əlavə Xərclər Olmadan Süni İntellektini Sürətləndirən Ən Müasir Texnikalar

Hochmoderne Techniken, die KI ohne Mehrkosten beschleunigen

Técnicas de vanguardia que aceleran la IA sin costes adicionales

تکنیک های پیشرفته ای که هوش مصنوعی را بدون هزینه اضافی سرعت می بخشد

Técnicas de vangarda que aceleran a IA sen custos adicionais

追加コストなしでAIを高速化する最先端技術

Teknika manafaingana ny AI tsy misy sarany fanampiny

Најсовремени техники кои ја забрзуваат вештачката интелигенција без дополнителни трошоци

අමතර වියදම් නොමැතිව AI වේගවත් කරන අති නවීන තාක්ෂණික ක්‍රම

Farsamooyinka Goynta-Edge ee Kor u qaadaya AI iyada oo aan kharash dheeraad ah lahayn

Teknikat më të fundit që përshpejtojnë AI pa kosto shtesë

Ekstra Maliyet Olmadan Yapay Zekayı Hızlandıran Son Teknoloji Teknikler

Many people in favor of self-driving cars believe they will dramatically improve safety, reduce traffic backups, and forever change transportation. However, some of those expected benefits are up for debate, especially since the vehicles are not yet widely used worldwide and could have unintended downsides. Ongoing research shows these high-tech cars could worsen traffic. Why is that, and how can developers proactively address the problem? They Stop When Encountering Problems One reason self-driving cars could exacerbate traffic issues concerns how they deal with unexpected things in their paths. If a driver sees a broken-down vehicle, accident, or something else blocking their way, they will usually notice it in time to steer around it. That may slow traffic but usually does not stop it. In contrast, some self-driving cars have stopped in the middle of the road when detecting something unusual. These instances even delay emergency services vehicles responding to accidents. Many self-driving cars have humans inside, supervising the journeys. However, when an autonomous vehicle operates independently and experiences a problem such as an overheated engine or a flat tire, would it know to pull over to the side of the road, or would it stop in place? The latter option would also cause traffic backups. Activists have even capitalized on how the cars stop when encountering obstacles. They’ve specifically placed traffic cones in the vehicles’ way, knowing that will create enough of an impediment to halt them. It’s easy to see how stopping in the presence of a barrier is a safety feature that could prevent collisions. However, it could also stop traffic because these vehicles lack the foresight, problem-solving capabilities, and experience humans naturally use when faced with them. Unfortunately, some people with advanced features on their cars have already begun depending on them too much. One study found that 58% of drivers with pedestrian-detecting cars have stopped looking for walkers themselves. They Do Not Understand Social Signs Humans in and around cars naturally give and receive signals to others around them. A pedestrian may make eye contact with a driver and watch for a nod from them before deciding to cross the street. Drivers regularly wave to each other, indicating they will let other cars out at busy intersections. Most people probably cannot pinpoint when or how they learned these things, but they understand how it makes driving and being near cars safer for everyone. Researchers analyzed YouTube videos of self-driving cars and found they have difficulty determining whether to yield or keep moving. Their programming also falls short of allowing them to detect body language. One example reviewed by the team involved a family that wanted to cross the street. Although they waved their hands to tell a self-driving car to pass them first, the vehicle did not recognize that cue. What happened next was particularly surprising because the vehicle stopped next to the family for 11 seconds. They took that as a sign to begin crossing. As soon as they did, the car started moving, making them run for the sidewalk. The nods, waves, and other brief acknowledgments exchanged between drivers, pedestrians, and others go a long way in keeping traffic flowing. People understand that although safe driving requires making personal decisions, it is also a collective effort. That partially explains why high-risk drivers are the most significant accident threats within fleets. They Need More Connectivity The programming and connectivity of self-driving cars affect how they respond to traffic signals. Researchers also determined that those able to receive signals from nearby infrastructure are more likely to move smoothly rather than get caught in backups. The team evaluated cars driven by humans, in addition to automated, connected, and connected automated vehicles. Those in the connected categories can receive information about the future state of nearby traffic lights and adjust their speeds. Therefore, they are more likely to encounter green lights rather than red or yellow ones. In contrast, automated vehicles without advanced connectivity are programmed to operate more conservatively and cautiously, which can reduce the number of cars that can flow through intersections. A computational model showed that automated cars without connectivity to traffic control systems slow intersection travel time, partly because their programming prioritizes collision reduction. Although accident prevention is an undeniably important aim, the researchers believe having more vehicles that connect and respond to traffic infrastructure would minimize the number sitting at red lights, allowing them to move with fewer interruptions. Exploring Potential Solutions These challenges highlight how self-driving car developers must consider multiple aspects and take thorough problem-solving approaches. A broad but practical step is to launch educational and awareness campaigns to help drivers learn more about the high-tech vehicles that increasingly appear on the roads. Some countries require people learning to drive to attach stickers to their cars, which helps others be more patient and understanding. Some self-driving cars have external branding that tells people what they are, but a better option may be to have a consistently applied designator. Signs could let people know that autonomous vehicles immediately stop in front of obstacles. That way, they’d be less likely to get frustrated and blow their horns when they don’t move. In one case, researchers determined that putting green lights on self-driving cars’ roofs effectively told pedestrians when it was safe to cross the road. However, the timing of the light’s activation was also important. People showed riskier behaviors if the automobiles slowed down or turned on the light too far away from intersections. It would be better if the light-activated as the vehicles arrived at the crossing and just before they stopped. Making self-driving cars reverse and go around obstacles directly in front of them or steer around them in time is also important from safety and traffic flow standpoints. Humans do this naturally, making those movements when it is safe. Similarly, autonomous vehicles should be programmed to steer to road shoulders when mechanical faults or other issues make it unsafe to continue moving. Addressing New Findings Those in the self-driving car industry should stay aware of new developments. Although these vehicles could ease traffic challenges, people should not assume that will happen. They must look deeper to find and address other possibilities to make the roads safer for all. Many people in favor of self-driving cars believe they will dramatically improve safety, reduce traffic backups, and forever change transportation. However, some of those expected benefits are up for debate, especially since the vehicles are not yet widely used worldwide and could have unintended downsides. Ongoing research shows these high-tech cars could worsen traffic. Why is that, and how can developers proactively address the problem? Many people in favor of self-driving cars believe they will dramatically improve safety, reduce traffic backups, and forever change transportation. However, some of those expected benefits are up for debate, especially since the vehicles are not yet widely used worldwide and could have unintended downsides. Ongoing research shows these high-tech cars could worsen traffic. Why is that, and how can developers proactively address the problem? They Stop When Encountering Problems They Stop When Encountering Problems One reason self-driving cars could exacerbate traffic issues concerns how they deal with unexpected things in their paths. If a driver sees a broken-down vehicle, accident, or something else blocking their way, they will usually notice it in time to steer around it. That may slow traffic but usually does not stop it. In contrast, some self-driving cars have stopped in the middle of the road when detecting something unusual. These instances even delay emergency services vehicles responding to accidents. stopped in the middle of the road stopped in the middle of the road Many self-driving cars have humans inside, supervising the journeys. However, when an autonomous vehicle operates independently and experiences a problem such as an overheated engine or a flat tire, would it know to pull over to the side of the road, or would it stop in place? The latter option would also cause traffic backups. Activists have even capitalized on how the cars stop when encountering obstacles. They’ve specifically placed traffic cones in the vehicles’ way, knowing that will create enough of an impediment to halt them. It’s easy to see how stopping in the presence of a barrier is a safety feature that could prevent collisions. However, it could also stop traffic because these vehicles lack the foresight, problem-solving capabilities, and experience humans naturally use when faced with them. Unfortunately, some people with advanced features on their cars have already begun depending on them too much. One study found that 58% of drivers with pedestrian-detecting cars have stopped looking for walkers themselves. 58% of drivers with pedestrian-detecting cars 58% of drivers with pedestrian-detecting cars They Do Not Understand Social Signs Humans in and around cars naturally give and receive signals to others around them. A pedestrian may make eye contact with a driver and watch for a nod from them before deciding to cross the street. Drivers regularly wave to each other, indicating they will let other cars out at busy intersections. Most people probably cannot pinpoint when or how they learned these things, but they understand how it makes driving and being near cars safer for everyone. Researchers analyzed YouTube videos of self-driving cars and found they have difficulty determining whether to yield or keep moving. Their programming also falls short of allowing them to detect body language. One example reviewed by the team involved a family that wanted to cross the street. Although they waved their hands to tell a self-driving car to pass them first, the vehicle did not recognize that cue. found they have difficulty determining found they have difficulty determining What happened next was particularly surprising because the vehicle stopped next to the family for 11 seconds. They took that as a sign to begin crossing. As soon as they did, the car started moving, making them run for the sidewalk. The nods, waves, and other brief acknowledgments exchanged between drivers, pedestrians, and others go a long way in keeping traffic flowing. People understand that although safe driving requires making personal decisions, it is also a collective effort. That partially explains why high-risk drivers are the most significant accident threats within fleets. the most significant accident threats the most significant accident threats They Need More Connectivity The programming and connectivity of self-driving cars affect how they respond to traffic signals. Researchers also determined that those able to receive signals from nearby infrastructure are more likely to move smoothly rather than get caught in backups. The team evaluated cars driven by humans, in addition to automated, connected, and connected automated vehicles. Those in the connected categories can receive information about the future state of nearby traffic lights and adjust their speeds. Therefore, they are more likely to encounter green lights rather than red or yellow ones. In contrast, automated vehicles without advanced connectivity are programmed to operate more conservatively and cautiously , which can reduce the number of cars that can flow through intersections. operate more conservatively and cautiously operate more conservatively and cautiously A computational model showed that automated cars without connectivity to traffic control systems slow intersection travel time, partly because their programming prioritizes collision reduction. Although accident prevention is an undeniably important aim, the researchers believe having more vehicles that connect and respond to traffic infrastructure would minimize the number sitting at red lights, allowing them to move with fewer interruptions. Exploring Potential Solutions These challenges highlight how self-driving car developers must consider multiple aspects and take thorough problem-solving approaches. A broad but practical step is to launch educational and awareness campaigns to help drivers learn more about the high-tech vehicles that increasingly appear on the roads. campaigns to help drivers learn more campaigns to help drivers learn more Some countries require people learning to drive to attach stickers to their cars, which helps others be more patient and understanding. Some self-driving cars have external branding that tells people what they are, but a better option may be to have a consistently applied designator. Signs could let people know that autonomous vehicles immediately stop in front of obstacles. That way, they’d be less likely to get frustrated and blow their horns when they don’t move. In one case, researchers determined that putting green lights on self-driving cars’ roofs effectively told pedestrians when it was safe to cross the road. However, the timing of the light’s activation was also important. People showed riskier behaviors if the automobiles slowed down or turned on the light too far away from intersections. It would be better if the light-activated as the vehicles arrived at the crossing and just before they stopped. told pedestrians when it was safe told pedestrians when it was safe Making self-driving cars reverse and go around obstacles directly in front of them or steer around them in time is also important from safety and traffic flow standpoints. Humans do this naturally, making those movements when it is safe. Similarly, autonomous vehicles should be programmed to steer to road shoulders when mechanical faults or other issues make it unsafe to continue moving. Addressing New Findings Those in the self-driving car industry should stay aware of new developments. Although these vehicles could ease traffic challenges, people should not assume that will happen. They must look deeper to find and address other possibilities to make the roads safer for all.

Could Self-Driving Cars Make Traffic Worse?

What Drives Inflation?

When we last asked, “What the heck is???”, we were looking at the __Apache Kafka__® protocol compatible data streaming platform WarpStream. With streaming data, you need to do something with it. Land it somewhere like Apache Iceberg, Snowflake, or Databricks, or query that in-flight data to enrich and/or filter it before landing it. There are some options out there for the latter, but the biggest one is undoubtedly the open-source project Apache Flink. Looking at Flink led me to DeltaStream, which is our subject today. What the heck is DeltaStream? How does it work with Flink, and how is it a companion to Kafka? What is Flink? Flink was initially accepted as an Apache project in December 2014, so it has been around for a while. The growth of stream processing has led to accelerated interest and use in the last few years. Flink is a somewhat challenging system to stand up and use internally, requiring dedicated engineering talent. Even the AWS MSF service, while easier, is still fairly complex. Utilizing Java, for example, requires you to write your Java query, generate a jar file, zip it, upload it to S3, set your permissions, and then execute it. I’m going to borrow from the Apache Flink web page here. Apache Flink is a framework and distributed processing engine for stateful computations over unbounded and bounded data streams. Flink has been designed to run in all common cluster environments and perform computations at in-memory speed and at any scale. I don’t want to replicate more of what is on their website, so if you aren’t sure about Flink, give it a look. Using DeltaStream At its most basic, DeltaStream gives you the power of Flink without having to deal with Flink. At all. For my test, I used their Quick Start Guide for the Web UI; note that they also have a CLI, which I did not test. When you sign up, you get a free 14-day trial. A sample Kafka cluster called “trial_store” is spun up that you can use instead of your own data to try out. The demo contains a number of topics to play with. The icons under ACTIONS allow you to delete a topic or view details and content about and in the topic. Here are the details of the pageviews topic for reference, which we’ll use later. Okay, we’ve got a Kafka cluster going and topics in it. Now, let’s do something interesting. As mentioned in the intro, the most interesting thing I can do is enrich and/or filter data while it is in flight before it lands at its ultimate destination, like a database/warehouse/lake. For those purposes, we go to the Workspace. This part took a little getting used to. A database and a schema in DeltaStream are just organizational folders. You can create ‘n’ number of databases, and within a database, you can have ‘n’ number of schemas. The schemas will hold the definitional layout of your DeltaStream Objects known as STREAM, CHANGELOG, MATERIALIZED VIEW, and TABLE. A Table relates to a database table in something like PostgreSQL, and the Materialized View is a way to persist the data in a result set without putting it somewhere specifically. I’m not going to do anything with either of those in this blog; I’m going to focus on Stream and Changelog. I’m creating a Stream of the pageviews topic in the Kafka cluster in the following screenshot. I think of it as making a table definition of the topic. We’ve assigned the names of the fields and told them what topic to use and what the data format is. We don’t have to fully justify the trial_store cluster as it is set as the default in the combo box at the top. Once that command is executed, it will show up under TestDB.public. I can then query it with something like SELECT * FROM PAGEVIEWS, and I’ll start seeing the data in the result pane at the bottom. Next, I declare a changelog backed by the user's topic and ordered by UserID. A changelog is similar to a stream but enables you to interpret events in a topic as UPSERT events. Events require a primary key; DeltaStream interprets each event as an insert or update for the given primary key. In this case, the changelog reflects specific details by user, such as gender and interests. Here is where we start to have some fun. I will create a new stream that enriches the pageviews stream with data from the users_log changelog using the userid as the key value to join on. This now gives me a unique topic in the cluster with data from two different topics. From here, I can filter it using something like regionid and write the results of that query to a final destination, such as a database, warehouse, or lake. This allowed me to enrich and filter data in-flight before landing it, thus improving latency and reducing compute and storage costs. Summary So, what the heck is DeltaStream? It’s a really simple way to use Apache Flink without knowing anything about it or directly doing anything with it. You saw from my example how straightforward it was to connect to Apache Kafka and then read, join, and filter the data. Other currently supported connections are Kinesis, PostgreSQL, Snowflake, and Databricks, and I’m told that ClickHouse and Iceberg will soon be available. Ultimately, DeltaStream gives you the power of Apache Flink without having to deal with Apache Flink, and you can do it using SQL instead of Java. If you are dealing with streaming data or looking to implement it, this is definitely a very clever and convenient solution. Check out my other What the Heck is… articles at the links below: What The Heck Is DuckDB?
What the Heck Is Malloy?
What the Heck is PRQL?
What the Heck is GlareDB?
What the Heck is SeaTunnel?
What the Heck is LanceDB?
What the heck is SDF?
What the Heck is Paimon?
What the Heck is Proton?
What the Heck is PuppyGraph?
What the Heck is GPTScript?
What the Heck is WarpStream? ** When we last asked, “What the heck is???”, we were looking at the __ Apache Kafka __® protocol compatible data streaming platform WarpStream . With streaming data, you need to do something with it. Land it somewhere like Apache Iceberg , Snowflake , or Databricks , or query that in-flight data to enrich and/or filter it before landing it. There are some options out there for the latter, but the biggest one is undoubtedly the open-source project Apache Flink . Looking at Flink led me to DeltaStream , which is our subject today. What the heck is DeltaStream? How does it work with Flink, and how is it a companion to Kafka? Apache Kafka WarpStream WarpStream Apache Iceberg Apache Iceberg Snowflake Snowflake Databricks Databricks Apache Flink Apache Flink DeltaStream DeltaStream What is Flink? Flink was initially accepted as an Apache project in December 2014, so it has been around for a while. The growth of stream processing has led to accelerated interest and use in the last few years. Flink is a somewhat challenging system to stand up and use internally, requiring dedicated engineering talent. Even the AWS MSF service, while easier, is still fairly complex. Utilizing Java, for example, requires you to write your Java query, generate a jar file, zip it, upload it to S3, set your permissions, and then execute it. AWS MSF AWS MSF I’m going to borrow from the Apache Flink web page here. Apache Flink is a framework and distributed processing engine for stateful computations over unbounded and bounded data streams. Flink has been designed to run in all common cluster environments and perform computations at in-memory speed and at any scale. I don’t want to replicate more of what is on their website, so if you aren’t sure about Flink, give it a look. Using DeltaStream At its most basic, DeltaStream gives you the power of Flink without having to deal with Flink. At all. For my test, I used their Quick Start Guide for the Web UI; note that they also have a CLI, which I did not test. Quick Start Guide Quick Start Guide When you sign up, you get a free 14-day trial. A sample Kafka cluster called “trial_store” is spun up that you can use instead of your own data to try out. The demo contains a number of topics to play with. The icons under ACTIONS allow you to delete a topic or view details and content about and in the topic. Here are the details of the pageviews topic for reference, which we’ll use later. pageviews Okay, we’ve got a Kafka cluster going and topics in it. Now, let’s do something interesting. As mentioned in the intro, the most interesting thing I can do is enrich and/or filter data while it is in flight before it lands at its ultimate destination, like a database/warehouse/lake. For those purposes, we go to the Workspace. This part took a little getting used to. A database and a schema in DeltaStream are just organizational folders. You can create ‘n’ number of databases, and within a database, you can have ‘n’ number of schemas. The schemas will hold the definitional layout of your DeltaStream Objects known as STREAM, CHANGELOG, MATERIALIZED VIEW, and TABLE. A Table relates to a database table in something like PostgreSQL, and the Materialized View is a way to persist the data in a result set without putting it somewhere specifically. I’m not going to do anything with either of those in this blog; I’m going to focus on Stream and Changelog. I’m creating a Stream of the pageviews topic in the Kafka cluster in the following screenshot. I think of it as making a table definition of the topic. We’ve assigned the names of the fields and told them what topic to use and what the data format is. We don’t have to fully justify the trial_store cluster as it is set as the default in the combo box at the top. Once that command is executed, it will show up under TestDB.public. I can then query it with something like SELECT * FROM PAGEVIEWS, and I’ll start seeing the data in the result pane at the bottom. pageviews Next, I declare a changelog backed by the user's topic and ordered by UserID. A changelog is similar to a stream but enables you to interpret events in a topic as UPSERT events. Events require a primary key; DeltaStream interprets each event as an insert or update for the given primary key. In this case, the changelog reflects specific details by user, such as gender and interests. user's Here is where we start to have some fun. I will create a new stream that enriches the pageviews stream with data from the users_log changelog using the userid as the key value to join on. This now gives me a unique topic in the cluster with data from two different topics. From here, I can filter it using something like regionid and write the results of that query to a final destination, such as a database, warehouse, or lake. This allowed me to enrich and filter data in-flight before landing it, thus improving latency and reducing compute and storage costs. pageviews users_log userid Summary So, what the heck is DeltaStream? It’s a really simple way to use Apache Flink without knowing anything about it or directly doing anything with it. You saw from my example how straightforward it was to connect to Apache Kafka and then read, join, and filter the data. Other currently supported connections are Kinesis, PostgreSQL, Snowflake, and Databricks, and I’m told that ClickHouse and Iceberg will soon be available. Ultimately, DeltaStream gives you the power of Apache Flink without having to deal with Apache Flink, and you can do it using SQL instead of Java. If you are dealing with streaming data or looking to implement it, this is definitely a very clever and convenient solution. Check out my other What the Heck is… articles at the links below: What The Heck Is DuckDB? What the Heck Is Malloy? What the Heck is PRQL? What the Heck is GlareDB? What the Heck is SeaTunnel? What the Heck is LanceDB? What the heck is SDF? What the Heck is Paimon? What the Heck is Proton? What the Heck is PuppyGraph? What the Heck is GPTScript? What the Heck is WarpStream? What The Heck Is DuckDB? What The Heck Is DuckDB? What The Heck Is DuckDB? What the Heck Is Malloy? What the Heck Is Malloy? What the Heck Is Malloy? What the Heck is PRQL? What the Heck is PRQL? What the Heck is PRQL? What the Heck is GlareDB? What the Heck is GlareDB? What the Heck is GlareDB? What the Heck is SeaTunnel? What the Heck is SeaTunnel? What the Heck is SeaTunnel? What the Heck is LanceDB? What the Heck is LanceDB? What the Heck is LanceDB? What the heck is SDF? What the heck is SDF? What the heck is SDF? What the Heck is Paimon? What the Heck is Paimon? What the Heck is Paimon? What the Heck is Proton? What the Heck is Proton? What the Heck is Proton? What the Heck is PuppyGraph? What the Heck is PuppyGraph? What the Heck is PuppyGraph? What the Heck is GPTScript? What the Heck is GPTScript? What the Heck is GPTScript? What the Heck is WarpStream? What the Heck is WarpStream? What the Heck is WarpStream? **

What The Heck is DeltaStream?

Co je sakra DeltaStream?

¿Qué diablos es DeltaStream?

دلتا استریم چیست؟

מה זה לעזאזל DeltaStream?

Mi a fene az a DeltaStream?

DeltaStream って一体何ですか?

អ្វី​ដែល​ The Heck ​គឺ​ជា​ DeltaStream?

O que diabos é DeltaStream?

DeltaStream යනු කුමක්ද?

Waa maxay Heck waa DeltaStream?

DeltaStream คืออะไรกันแน่?

Що таке DeltaStream?

Artificial intelligence is rampant in schools, but education professionals aren’t teaching students the importance of ethical utilization. The problem is that it’s a complex topic, and educators simply don’t have the time or resources to tackle it. Where do they get the relevant materials? How do they fit it into their existing lesson plans? Here’s how questions like these are holding AI ethics classes back — and what decision-makers are doing to fix the problem. The Importance of Teaching AI Ethics in School AI ethics is a set of principles that govern the development, deployment and utilization of algorithm-based tools. They are meant to guide stakeholders, minimizing the damage such technologies can do to the environment, society or the economy. On a smaller scale, they focus on preventing bias, discrimination, misinformation and abuse. Teaching AI ethics in school may seem unnecessary, but its importance cannot be overstated. AI has already become a staple in many industries. According to a McKinsey & Company report, 72% of organizations use it for at least one business function. If its upward trajectory continues, its impact could rival that of computers, smartphones or the internet. Avoiding talking about AI in the classroom because of its newness needlessly repeats history. It is the same as years ago when students were told they wouldn’t always have a calculator in their back pocket right after the smartphone was invented. This technology’s popularity is growing in both personal and professional domains. Students must be prepared to use it responsibly when entering the workforce to ensure better outcomes for their industry and society. AI is powerful. Advanced algorithms can analyze massive datasets exponentially faster than humans. They excel at pattern recognition, trend prediction and content creation. Whether educators like it or not, AI will likely change the world. If they prepare their students accordingly, they can ensure that change is positive. Ethical Dilemmas Students Face When Using AI Typical ethical risks students face when using AI include technology abuse, harassment and plagiarism. The latter is one of the most common. Thanks to generative tools, academic dishonesty is easy to get away with. Google Trends reveals that searches for “AI essay writing” increased by a staggering 2,041% after ChatGPT’s release. Unfortunately, plagiarism isn’t the most severe offense. Lately, university, high school and even middle school students have been caught using generative tools to take part in deepfake-related sex crimes. In South Korea, one journalist uncovered one such group that had over 2,000 members. Many victims were believed to be under 16 — under Korea’s age of consent. Even if kids aren’t using regular photos of their classmates to create explicit material, AI use still revolves around consent. This concept revisits data privacy and security principles — all data end users enter is no longer private. The platform or service provider may use it to train their model or enhance their marketing strategies. Many of today’s students view generative and machine learning models as entertaining, harmless tools — even when they are actively harming others. Teachers have a unique opportunity to educate them early on about the consequences of such actions, stopping unacceptable behavior like this at the source. Is Ethical AI Use Being Taught in Schools Today? School board standards, classroom policies and state laws vary dramatically, so AI adoption has been erratic at best. This split between those who embrace it and those who don’t has stalled standardization. However, some are still taking a chance on ethics courses. Elementary School While various academics and industry experts argue ethics discussions should start in elementary school, relatively few institutions even use algorithm-based tools that early. There are virtually no public resources available for grade school administrators. Middle and High School Many middle and high schools aren’t discussing the intricacies of ethical AI. Instead, they are cracking down on AI use. Studies show that the accuracy of AI detection tools is typically about 50%, so most instances slip by unnoticed. As a result, kids often learn bad habits. Promisingly, some decision-makers have seen the need for AI ethics. For instance, California Governor Gavin Newsom signed Bill No. 2876 into law in September 2024. When curriculum frameworks are revised in January 2025, schools must incorporate AI literacy content — which guides critical and ethical use — into the classroom. Not all educators receive playbooks from experts. Some, like Jeff Simon, a mathematics teacher at Sage Creek High School, are navigating this new field on their own. Simon said he is trying to “teach the ethical use” of AI since it is  “just going to keep on growing.” College and University Most AI ethics classes are offered at higher education institutions because working those discussions into data science or computer engineering courses is easier. At the very least, many have developed policies for generative AI use. Reasons Why AI Ethics Isn’t Being Taught in School AI is not just present in education — it is thriving. For reference, in 2024, 86% of students admitted to using AI for schoolwork. Despite its popularity, many schools have no guiding principles on how to use these models ethically. Time is one of the main constraints — educators face the impossible task of doing more with less. They already spend 54 hours at work per week. How will they learn about AI and develop classroom policies that align with ethics in addition to teaching curriculum, grading papers and creating lesson plans? The field of AI is complex and rapidly evolving. Even with district-wide support, educational materials may quickly become outdated. Many schools still use decades-old textbooks to teach, so any solution would require extensive retrofitting. Moreover, how do they send home ethics homework? What would tests look like? How would they assign fair grades? Educators may be willing to embrace the concept of ethical AI classes if given the resources, but there’s no clear framework on how that could be done. Should they begin teaching it in elementary school? Will it replace a class, be offered as an elective or be integrated into the existing curriculum? There’s no one-size-fits-all solution. Many teachers have adopted “not in my classroom” policies in response to these concerns. As a result, students likely won’t learn how to use these tools the right way. Many will take whatever inaccurate, unregulated knowledge they cobble together from peers, parents and the internet as the truth — which may have far-reaching impacts. Addressing Concerns to Make AI Ethics Classes a Reality There are several ways education professionals can address why AI ethics isn’t being taught in schools. Place AI Ethics Wherever It Fits Integrating AI ethics into the classroom can mean creating a new elective or squeezing it in at the end of the period. Placing it where it naturally appears is best. For example, when students are assigned an eight-page essay, they can hear about responsible generative AI use. Teachers and administrators should adapt these new courses to students’ grade, technical expertise and knowledge levels to make the subject matter age-appropriate and engaging. Assignments should be relatively open-ended but follow a clear ethical framework. Tailor Subject Matter for End Users Typically, conversations about ethical AI focus on data privacy and output bias, centering around people who develop, deploy or manage these tools. End users interact with the user interface — not the backend — so their approach must differ slightly. Core principles should be accountability, value alignment, explainability, fairness and integrity. Pupils should learn critical thinking to combat bias, reasoning to prevent overreliance and objective analysis to mitigate the spread of misinformation. Model Responsible AI Use If state law or school board policies make teaching the ethics of AI in the classroom difficult, educators should take advantage of workarounds. Modeling responsible, honorable behavior goes a long way. Alternatively, they can invite a guest speaker. Facilitate In-Class Ethics Discussions Facilitating discussions, debates and group work in class lets high school teachers and college professors act as moderators. Posing questions about data de-identification, copyright infringement and summarization as they relate to AI can initiate productive conversations. Modern Lessons for a Modern Age Digitalization’s popularity is exponentially increasing. Soon, AI may be a part of every industry in the world. If educators modernize their lessons for this modern age, they can ensure today’s youth have a headstart once they reach the workforce. Artificial intelligence is rampant in schools, but education professionals aren’t teaching students the importance of ethical utilization. The problem is that it’s a complex topic, and educators simply don’t have the time or resources to tackle it. Artificial intelligence is rampant in schools, but education professionals aren’t teaching students the importance of ethical utilization. The problem is that it’s a complex topic, and educators simply don’t have the time or resources to tackle it. Where do they get the relevant materials? How do they fit it into their existing lesson plans? Here’s how questions like these are holding AI ethics classes back — and what decision-makers are doing to fix the problem. The Importance of Teaching AI Ethics in School AI ethics is a set of principles that govern the development, deployment and utilization of algorithm-based tools. They are meant to guide stakeholders, minimizing the damage such technologies can do to the environment, society or the economy. On a smaller scale, they focus on preventing bias, discrimination, misinformation and abuse. Teaching AI ethics in school may seem unnecessary, but its importance cannot be overstated. AI has already become a staple in many industries. According to a McKinsey & Company report, 72% of organizations use it for at least one business function. If its upward trajectory continues, its impact could rival that of computers, smartphones or the internet. 72% of organizations use it 72% of organizations use it Avoiding talking about AI in the classroom because of its newness needlessly repeats history. It is the same as years ago when students were told they wouldn’t always have a calculator in their back pocket right after the smartphone was invented. This technology’s popularity is growing in both personal and professional domains. Students must be prepared to use it responsibly when entering the workforce to ensure better outcomes for their industry and society. AI is powerful. Advanced algorithms can analyze massive datasets exponentially faster than humans. They excel at pattern recognition, trend prediction and content creation. Whether educators like it or not, AI will likely change the world. If they prepare their students accordingly, they can ensure that change is positive. Ethical Dilemmas Students Face When Using AI Typical ethical risks students face when using AI include technology abuse, harassment and plagiarism. The latter is one of the most common. Thanks to generative tools, academic dishonesty is easy to get away with. Google Trends reveals that searches for “AI essay writing” increased by a staggering 2,041% after ChatGPT’s release. increased by a staggering 2,041% increased by a staggering 2,041% Unfortunately, plagiarism isn’t the most severe offense. Lately, university, high school and even middle school students have been caught using generative tools to take part in deepfake-related sex crimes. In South Korea, one journalist uncovered one such group that had over 2,000 members. Many victims were believed to be under 16 — under Korea’s age of consent. uncovered one such group uncovered one such group Even if kids aren’t using regular photos of their classmates to create explicit material, AI use still revolves around consent. This concept revisits data privacy and security principles — all data end users enter is no longer private. The platform or service provider may use it to train their model or enhance their marketing strategies. Many of today’s students view generative and machine learning models as entertaining, harmless tools — even when they are actively harming others. Teachers have a unique opportunity to educate them early on about the consequences of such actions, stopping unacceptable behavior like this at the source. Is Ethical AI Use Being Taught in Schools Today? School board standards, classroom policies and state laws vary dramatically, so AI adoption has been erratic at best. This split between those who embrace it and those who don’t has stalled standardization. However, some are still taking a chance on ethics courses. Elementary School While various academics and industry experts argue ethics discussions should start in elementary school, relatively few institutions even use algorithm-based tools that early. There are virtually no public resources available for grade school administrators. Middle and High School Many middle and high schools aren’t discussing the intricacies of ethical AI. Instead, they are cracking down on AI use. Studies show that the accuracy of AI detection tools is typically about 50% , so most instances slip by unnoticed. As a result, kids often learn bad habits. is typically about 50% is typically about 50% Promisingly, some decision-makers have seen the need for AI ethics. For instance, California Governor Gavin Newsom signed Bill No. 2876 into law in September 2024. When curriculum frameworks are revised in January 2025, schools must incorporate AI literacy content — which guides critical and ethical use — into the classroom. signed Bill No. 2876 into law signed Bill No. 2876 into law Not all educators receive playbooks from experts. Some, like Jeff Simon, a mathematics teacher at Sage Creek High School, are navigating this new field on their own. Simon said he is trying to “teach the ethical use” of AI since it is  “just going to keep on growing.” trying to “teach the ethical use” trying to “teach the ethical use” College and University Most AI ethics classes are offered at higher education institutions because working those discussions into data science or computer engineering courses is easier. At the very least, many have developed policies for generative AI use. Reasons Why AI Ethics Isn’t Being Taught in School AI is not just present in education — it is thriving. For reference, in 2024, 86% of students admitted to using AI for schoolwork. Despite its popularity, many schools have no guiding principles on how to use these models ethically. 86% of students admitted 86% of students admitted Time is one of the main constraints — educators face the impossible task of doing more with less. They already spend 54 hours at work per week. How will they learn about AI and develop classroom policies that align with ethics in addition to teaching curriculum, grading papers and creating lesson plans? spend 54 hours at work spend 54 hours at work The field of AI is complex and rapidly evolving. Even with district-wide support, educational materials may quickly become outdated. Many schools still use decades-old textbooks to teach, so any solution would require extensive retrofitting. Moreover, how do they send home ethics homework? What would tests look like? How would they assign fair grades? Educators may be willing to embrace the concept of ethical AI classes if given the resources, but there’s no clear framework on how that could be done. Should they begin teaching it in elementary school? Will it replace a class, be offered as an elective or be integrated into the existing curriculum? There’s no one-size-fits-all solution. Many teachers have adopted “not in my classroom” policies in response to these concerns. As a result, students likely won’t learn how to use these tools the right way. Many will take whatever inaccurate, unregulated knowledge they cobble together from peers, parents and the internet as the truth — which may have far-reaching impacts. Addressing Concerns to Make AI Ethics Classes a Reality There are several ways education professionals can address why AI ethics isn’t being taught in schools. Place AI Ethics Wherever It Fits Integrating AI ethics into the classroom can mean creating a new elective or squeezing it in at the end of the period. Placing it where it naturally appears is best. For example, when students are assigned an eight-page essay, they can hear about responsible generative AI use. Teachers and administrators should adapt these new courses to students’ grade, technical expertise and knowledge levels to make the subject matter age-appropriate and engaging. Assignments should be relatively open-ended but follow a clear ethical framework. Tailor Subject Matter for End Users Typically, conversations about ethical AI focus on data privacy and output bias, centering around people who develop, deploy or manage these tools. End users interact with the user interface — not the backend — so their approach must differ slightly. Core principles should be accountability, value alignment, explainability, fairness and integrity. Pupils should learn critical thinking to combat bias, reasoning to prevent overreliance and objective analysis to mitigate the spread of misinformation. Model Responsible AI Use If state law or school board policies make teaching the ethics of AI in the classroom difficult, educators should take advantage of workarounds. Modeling responsible, honorable behavior goes a long way. Alternatively, they can invite a guest speaker. Facilitate In-Class Ethics Discussions Facilitating discussions, debates and group work in class lets high school teachers and college professors act as moderators. Posing questions about data de-identification, copyright infringement and summarization as they relate to AI can initiate productive conversations. Modern Lessons for a Modern Age Digitalization’s popularity is exponentially increasing. Soon, AI may be a part of every industry in the world. If educators modernize their lessons for this modern age, they can ensure today’s youth have a headstart once they reach the workforce.

Why Aren't We Teaching AI Ethics in Schools?

Waarom onderrig ons nie KI-etiek in skole nie?

لماذا لا نقوم بتدريس أخلاقيات الذكاء الاصطناعي في المدارس؟

¿Kunatsa Yatiqañ Utanakan jan AI Ética Yatichktanti?

Per què no estem ensenyant l'ètica de la IA a les escoles?

Hvorfor underviser vi ikke i AI-etik i skolerne?

¿Por qué no enseñamos la ética de la IA en las escuelas?

Pourquoi n’enseignons-nous pas l’éthique de l’IA dans les écoles ?

なぜ学校でAI倫理を教えないのか?

Почему мы не преподаем этику ИИ в школах?

Neden Okullarda Yapay Zeka Etiğini Öğretmiyoruz?

Ha Yini Hi Nga Dyondzisi Mahanyelo Ya AI eSwikolweni?

Tại sao chúng ta không dạy đạo đức AI ở trường học?

Home Assistant is a massive beast. It can be overwhelming for a newcomer; it was for me. In this post, I want to describe the underlying model of Home Assistant, which is a good entry point for your home automation journey. The biggest issue in describing the Home Assistant is the number of conflicting sources for this model: The helpers package of the GitHub repository. The database; disclaimer: I didn't find the schema generation in the code, and I wasn't bold enough to check the deployed SQLite database. The documentation. The web interface. As far as I understand it, here's the conceptual model. Integrations Integrations are pieces of software that allow Home Assistant to connect to other software and platforms. For example, a product by Philips called Hue would use the Philips Hue integration and allow Home Assistant to talk to the hardware controller Hue Bridge. Any Home Assistant compatible devices connected to the Hue Bridge would appear in Home Assistant as devices.
-- Home Assistant documentation When you switch on Home Assistant for the first time, it's a blank slate. You must make it communicate with your different devices. That's the role of Integration; thus, I think it's a critical entity. You can add an Integration on the integration dashboard: there's an ADD INTEGRATION in the lower left corner. Alternatively, you can add it from its documentation page, via a ADD INTEGRATION TO MY HOME ASSISTANT button. In this case, clicking on the latter opens a new page. You can change it; the URL is stored on the browser. Some integrations require additional configuration, e.g., authentication. Devices Devices are a logical grouping for one or more entities. A device may represent a physical device, which can have one or more sensors. The sensors appear as entities associated with the device. For example, a motion sensor is represented as a device. It may provide motion detection, temperature, and light levels as entities. Entities have states such as detected when motion is detected and clear when there is no motion.
-- Home Assistant documentation Once you've added the integration, you can add Device: Some categorize as Service when they need a Cloud connection, e.g., Google Cloud SDK or GitHub.



Some as Device when they need one, e.g., Philips Hue. Entities Entities are the basic building blocks to hold data in Home Assistant. An entity represents a sensor, actor, or function in Home Assistant. Entities are used to monitor physical properties or to control other entities. An entity is usually part of a device or a service. Entities have states.
-- Home Assistant documentation For example, the Philips Hue motion sensor device has three different available sensors: Illuminance in Lux
Motion as a boolean
Temperature in Celsius Unexpectedly, a motion sensor provides a temperature sensor. Check your devices for such sensors; you might be in for a good surprise. Other Objects Areas:
An Area represents a location. You can put a physical  Device in an Area.

Automations:
An Automation is pretty self-explanatory. Each consists of two main components: a trigger and an action. Note that you can add an optional condition.For example, let's imagine a Welcome Home Automation:
The trigger is when the user enters a geofenced perimeter around one's home.
The action is to switch on the lights
Yet, switching on the lights in daylight doesn't make sense.

The condition would be "only during the night."

Scripts:
As per the documentation, a Script is akin to an Automation but with no trigger; it's a named action. It lets you define the action once and reuse it across several Automation objects. Think DRY.

Add-ons:
While an Integration allows interfacing with an external device or service, an Add-On provides a service. In other words, an Add-On extends the platform itself. For example, I installed the Let's Encrypt add-on to my Home Assistant installation to automatically manage a TLS certificate.

Scenes:
A Scene for a Device is akin to a profile for your phone. It's a named set of configurations that you save and reuse. I must admit that I don't use profiles on my phone; I don't have any Scene objects at the moment, and I have to understand their usage beyond lighting. Conclusion In this post, I described the model objects available in Home Assistant. In the next post, I'll migrate from the proprietary Philips Hue automation to Home Assistant. To go further: Home Assistant
Concepts and terminology Originally published at A Java Geek on December 2nd, 2024 Home Assistant is a massive beast. It can be overwhelming for a newcomer; it was for me. In this post, I want to describe the underlying model of Home Assistant, which is a good entry point for your home automation journey. The biggest issue in describing the Home Assistant is the number of conflicting sources for this model: The helpers package of the GitHub repository. The helpers package of the GitHub repository . helpers GitHub repository The database; disclaimer: I didn't find the schema generation in the code, and I wasn't bold enough to check the deployed SQLite database. The database; disclaimer: I didn't find the schema generation in the code, and I wasn't bold enough to check the deployed SQLite database. The documentation. The documentation . documentation The web interface. The web interface. As far as I understand it, here's the conceptual model. Integrations Integrations are pieces of software that allow Home Assistant to connect to other software and platforms. For example, a product by Philips called Hue would use the Philips Hue integration and allow Home Assistant to talk to the hardware controller Hue Bridge. Any Home Assistant compatible devices connected to the Hue Bridge would appear in Home Assistant as devices. -- Home Assistant documentation Integrations are pieces of software that allow Home Assistant to connect to other software and platforms. For example, a product by Philips called Hue would use the Philips Hue integration and allow Home Assistant to talk to the hardware controller Hue Bridge. Any Home Assistant compatible devices connected to the Hue Bridge would appear in Home Assistant as devices. -- Home Assistant documentation Home Assistant documentation When you switch on Home Assistant for the first time, it's a blank slate. You must make it communicate with your different devices. That's the role of Integration ; thus, I think it's a critical entity. Integration You can add an Integration on the integration dashboard : there's an ADD INTEGRATION in the lower left corner. Alternatively, you can add it from its documentation page, via a ADD INTEGRATION TO MY HOME ASSISTANT button. In this case, clicking on the latter opens a new page. Integration integration dashboard ADD INTEGRATION ADD INTEGRATION TO MY HOME ASSISTANT You can change it; the URL is stored on the browser. Some integrations require additional configuration, e.g. , authentication. e.g. Devices Devices are a logical grouping for one or more entities. A device may represent a physical device, which can have one or more sensors. The sensors appear as entities associated with the device. For example, a motion sensor is represented as a device. It may provide motion detection, temperature, and light levels as entities. Entities have states such as detected when motion is detected and clear when there is no motion. -- Home Assistant documentation Devices are a logical grouping for one or more entities. A device may represent a physical device, which can have one or more sensors. The sensors appear as entities associated with the device. For example, a motion sensor is represented as a device. It may provide motion detection, temperature, and light levels as entities. Entities have states such as detected when motion is detected and clear when there is no motion. -- Home Assistant documentation Home Assistant documentation Once you've added the integration, you can add Device : Device Some categorize as Service when they need a Cloud connection, e.g., Google Cloud SDK or GitHub. Some as Device when they need one, e.g., Philips Hue. Some categorize as Service when they need a Cloud connection, e.g., Google Cloud SDK or GitHub. Some categorize as Service when they need a Cloud connection, e.g. , Google Cloud SDK or GitHub . Service e.g. Google Cloud SDK GitHub Some as Device when they need one, e.g., Philips Hue. Some as Device when they need one, e.g. , Philips Hue . Device e.g. Philips Hue Entities Entities are the basic building blocks to hold data in Home Assistant. An entity represents a sensor, actor, or function in Home Assistant. Entities are used to monitor physical properties or to control other entities. An entity is usually part of a device or a service. Entities have states. -- Home Assistant documentation Entities are the basic building blocks to hold data in Home Assistant. An entity represents a sensor, actor, or function in Home Assistant. Entities are used to monitor physical properties or to control other entities. An entity is usually part of a device or a service. Entities have states. -- Home Assistant documentation Home Assistant documentation For example, the Philips Hue motion sensor device has three different available sensors: Illuminance in Lux Motion as a boolean Temperature in Celsius Illuminance in Lux Motion as a boolean Temperature in Celsius Unexpectedly, a motion sensor provides a temperature sensor. Check your devices for such sensors; you might be in for a good surprise. Other Objects Areas:
An Area represents a location. You can put a physical  Device in an Area. Automations:
An Automation is pretty self-explanatory. Each consists of two main components: a trigger and an action. Note that you can add an optional condition.For example, let's imagine a Welcome Home Automation:
The trigger is when the user enters a geofenced perimeter around one's home.
The action is to switch on the lights
Yet, switching on the lights in daylight doesn't make sense.

The condition would be "only during the night." Scripts:
As per the documentation, a Script is akin to an Automation but with no trigger; it's a named action. It lets you define the action once and reuse it across several Automation objects. Think DRY. Add-ons:
While an Integration allows interfacing with an external device or service, an Add-On provides a service. In other words, an Add-On extends the platform itself. For example, I installed the Let's Encrypt add-on to my Home Assistant installation to automatically manage a TLS certificate. Scenes:
A Scene for a Device is akin to a profile for your phone. It's a named set of configurations that you save and reuse. I must admit that I don't use profiles on my phone; I don't have any Scene objects at the moment, and I have to understand their usage beyond lighting. Areas: An Area represents a location. You can put a physical  Device in an Area. Areas: An Area represents a location. You can put a physical Device in an Area . Area physical Device Area Automations: An Automation is pretty self-explanatory. Each consists of two main components: a trigger and an action. Note that you can add an optional condition. For example, let's imagine a Welcome Home Automation: The trigger is when the user enters a geofenced perimeter around one's home.
The action is to switch on the lights
Yet, switching on the lights in daylight doesn't make sense. The condition would be "only during the night." Automations: An Automation is pretty self-explanatory. Each consists of two main components: a trigger and an action. Note that you can add an optional condition. Automation For example, let's imagine a Welcome Home Automation : Automation The trigger is when the user enters a geofenced perimeter around one's home. The action is to switch on the lights Yet, switching on the lights in daylight doesn't make sense. The trigger is when the user enters a geofenced perimeter around one's home. The action is to switch on the lights Yet, switching on the lights in daylight doesn't make sense. The condition would be "only during the night." Scripts: As per the documentation, a Script is akin to an Automation but with no trigger; it's a named action. It lets you define the action once and reuse it across several Automation objects. Think DRY. Scripts: As per the documentation, a Script is akin to an Automation but with no trigger; it's a named action. It lets you define the action once and reuse it across several Automation objects. Think DRY. Script Automation Automation Add-ons: While an Integration allows interfacing with an external device or service, an Add-On provides a service. In other words, an Add-On extends the platform itself. For example, I installed the Let's Encrypt add-on to my Home Assistant installation to automatically manage a TLS certificate. Add-ons: While an Integration allows interfacing with an external device or service, an Add-On provides a service. In other words, an Add-On extends the platform itself. For example, I installed the Let's Encrypt add-on to my Home Assistant installation to automatically manage a TLS certificate. Integration Add-On Add-On Let's Encrypt Scenes: A Scene for a Device is akin to a profile for your phone. It's a named set of configurations that you save and reuse. I must admit that I don't use profiles on my phone; I don't have any Scene objects at the moment, and I have to understand their usage beyond lighting. Scenes: A Scene for a Device is akin to a profile for your phone. It's a named set of configurations that you save and reuse. I must admit that I don't use profiles on my phone; I don't have any Scene objects at the moment, and I have to understand their usage beyond lighting. Scene Device profile Scene Conclusion In this post, I described the model objects available in Home Assistant. In the next post, I'll migrate from the proprietary Philips Hue automation to Home Assistant. To go further: To go further: Home Assistant Concepts and terminology Home Assistant Home Assistant Concepts and terminology Concepts and terminology Originally published at A Java Geek on December 2nd, 2024 Originally published at A Java Geek on December 2nd, 2024 A Java Geek

Stories, News & Mentions about LATWATER

Latwater’s Stories on HackerNoon

Latwater's News & Mentions