Google joins Meta in the text-to-video race

Google joins Meta in the text-to-video race

www.mgireservationsandbookings.co.uk

Google's Imagen takes on Meta as text-to-video AI models ramp up

Last week, Meta announced Make-A-Video, an AI system that allows users to turn text prompts into short, high-quality, one-of-a-kind video clips. Now, Google isn’t far behind. The text-to-video trend shows all the signs of getting ready to explode, much like text-to-image did over the past year with DALL-E, MidJourney and Stable Diffusion.

Announced just yesterday, Google’s Imagen Video is a text-to-video generative AI model capable of producing high-definition videos from a text prompt. It can generate videos up to a resolution of 1280×768 at 24 frames per second.

With the help of progressive distillation, Imagen Video can generate high-quality videos using just eight diffusion steps per sub-model. This speeds up video generation time by a factor of about 18 times.

‘Tis the season of generative artificial intelligence (AI). Last week, Meta announced Make-A-Video, an AI system that allows users to turn text prompts into short, high-quality, one-of-a-kind video clips. Now, Google isn’t far behind. The text-to-video trend shows all the signs of getting ready to explode, much like text-to-image did over the past year with DALL-E, MidJourney and Stable Diffusion.??

Announced just yesterday, Google’s Imagen Video is a text-to-video generative AI model capable of producing high-definition videos from a text prompt. The text-conditioned video diffusion model can generate videos up to a resolution of 1280×768 at 24 frames per second.?

Google’s Imagen Video offers high fidelity

In its recently published paper “Imagen Video: High definition video generation with diffusion models” Google claims Imagen Video is capable of generating videos with high fidelity and has a high-degree of controllability and world knowledge. The generative model’s capabilities include creating diverse videos and text animations in different artistic styles, 3D understanding, text rendering and animation. The model is currently in a research phase, but its arrival comes just five months after Imagen showed the rapid development of synthesis-based models.

A look at Imagen Video

Imagen Video consists of a text encoder (frozen T5-XXL), a base video diffusion model, and interleaved spatial and temporal super-resolution diffusion models. To create such an architecture, Google claims it transferred findings from the previous work on diffusion-based image generation to the video generation setting. The research team also inculcated progressive distillation into the video models with classifier-free guidance for fast, high-quality sampling.

Cascade of seven sub-video diffusion models

The video generation framework is a cascade of seven sub-video diffusion models that perform text-conditional video generation, spatial super-resolution, and temporal super-resolution. With the entire cascade, Imagen Video generates high-definition 1280×768 videos at 24 frames per second for 128 frames — approximately 126 million pixels. With the help of progressive distillation, Imagen Video can generate high-quality videos using just eight diffusion steps per sub-model. This speeds up video generation time by a factor of about 18 times.

The model’s several notable stylistic abilities also include generating videos based on the work of renowned painters like Vincent van Gogh, rendering rotating objects in 3D while preserving their structure and rendering text in various animation styles.

Google says that Imagen Video was trained on the publicly available LAION-400M image-text dataset, as well as 14 million video-text pairs and 60 million image-text pairs. The training datasets allowed it to generalize a variety of aesthetics. In addition, a benefit of cascading models discovered by Google’s development team was that each diffusion model could be trained independently — allowing one to train all seven models in parallel.

A Google data dilemma

As generative models may also be misused to generate fake, hateful, explicit or harmful content, Google claims that it has taken multiple steps to minimize such concerns. Through internal trials, the company affirmed that it was able to apply input text prompt filtering and output video content filtering, but warned that there are still several important safety and ethical challenges that must be worked through.?

Imagen Video and its frozen T5-XXL text encoder were trained on “problematic data.” While internal testing shows that much of the explicit and violent content can be filtered out, Google says that social biases and stereotypes still exist which can be challenging to detect and filter. This was a major reason that Google decided not to release the model or its source code publicly until the concerns are mitigated.

Generative AI at Google and beyond??

According to Dumitru Erhan, a staff research scientist for Google Brain, there are efforts to strengthen research behind Phenaki — another Google text-to-video system — that can turn detailed text prompts into two-minute-plus videos; the main drawback of which is lower video quality.?

The team working on Phenaki says that the model can take advantage of the vast text-image datasets to generate videos, where the user can also narrate and dynamically change scenes.

A generative AI trend that started with text-to-image and has begun to move to text-to-video, also seems to be slowly transforming towards text-to-3D — with models such as CLIP-Forge, a text-to-shape generation model that can generate 3D objects using zero-shot learning.

Google’s very own text-to-3D AI “DreamFusion”, released last week, is another prime example of generative AI moving towards a more aggressive 3D synthesis approach. DreamFusion utilizes Imagen to optimize a 3D scene.?

Insider risk: More prevalent than ever, according to Microsof

?

Insider risk: More prevalent than ever, according to Microsoft

When you think of insider risk, what comes to mind — fraud, IP theft, maybe even corporate espionage?

While those are all undoubtedly significant causes for concern, the reality is that the riskiest insiders in your organization don’t even know they’re doing anything wrong.?

The evolving impact of graph analytics_

This calls for a “holistic” approach to insider risk management that doesn’t put off employees — but, rather, educates and trains them, fosters their collaboration and gains their buy-in.

This, at least, is the key message of a new Microsoft Insider Risk Report.?

Risks inadvertent and malicious

Insider risk can be both inadvertent and malicious, as described in the report. It is defined as the potential for a person to use authorized access to an organization’s assets in a way that negatively affects the organization. This access can be physical or virtual, and assets can include information, processes, systems and facilities.?

Inadvertent cases can include employees taking unsafe actions, being untrained or distracted, misusing resources or causing other accidental data leakage.?

On the other hand, malicious insiders are intentionally seeking to cause harm in the way of fraud, IP theft, unauthorized disclosure, sabotage or corporate espionage.?

The survey’s most significant findings:?

? Data breaches arising from insider actions cost businesses an average of $7.5 million annually; that’s in addition to the reputational damage, IP loss, and legal expenses that 4 out of 5 security experts say insiders cost their organizations.

? Almost 40% of respondents said the average cost of a single data breach from an insider event was more than $500,000.?

? The highest-rated impacts of insider risk events on organizations included theft or loss of customer data (84%) and damage to brand or reputation (82%).?

? The average number of inadvertent events was roughly 12 per year.

? Malicious events totaled around eight a year.?

? One-third of respondents reported that insider risk event occurrence increased in the past year, with a majority (40%) expecting events to increase going forward.

? Two-thirds highly agreed that, “Data theft or data destruction from departing employees is a form of insider risk that is becoming more commonplace.”

? Based on the level of insider risk per department, IT (ironically, most often tasked with detecting and remediating insider risk), was most identified (60%), followed by finance/accounting (48%), operations (44%) and senior leadership (40%).?

Hybrid work a top culprit

Per the report, the number of businesses that are seeing increases in insider risk is far higher than those reporting declines.?

A few trends contribute to this, said Arsenault. First: The rise in hybrid work. Microsoft’s 2022 Work Trend Index found that hybrid work now accounts for 38% of the workforce.?

?“That shift has fundamentally changed how we connect with each other,” said Arsenault. “It’s also created massive data estates spread across functions and platforms.”?

All of which brings inherent risk, he said. “The same tools we use to communicate and collaborate can open doors to data theft, sensitive data leaks, harassment, and other forms of inadvertent and malicious insider risks.”

Companies across the country are at a crossroads as flexible work evolves into a standard practice for many employers, said Arsenault. “And with these digital transformations come new challenges for security and compliance teams as employees increasingly rely on collaboration tools and platforms from locations around the world,” he said.?

Fragmented programs weak against sophisticated attacks

A second contributor is the increase in the size and sophistication of cyberthreats. Microsoft’s recent Digital Defense Report showed that cybercriminals overwhelmingly rely on successfully manipulating insider behavior to steal data, said Arsenault.

Thirdly is the response many organizations have to this expanded threat landscape.?

?“A fragmented risk management program — one that over-indexes on negative deterrents, deprioritizes organizational buy-in, and treats the employee as a potential threat instead of a trusted partner — can drive the risks it’s supposed to mitigate,” said Arsenault.?

Microsoft undertook this report because it wanted to understand the costs of insider risk and how it can impact organizations, he said.

“But we also wanted to understand how to address it; what an effective response looks like,” said Arsenault. “And we found that the best risk management programs weren’t the most invasive, or focused on constraining employee behavior. They were focused on building trust, on balancing security and privacy, and on educating and empowering their workforce.”

Positive and negative deterrents

Still, many organizations cited challenges and negative consequences with insider risk programs.?

Many pointed to concerns over employee privacy rights (52%), loss of employee trust (51%), and general degradation of the working environment — investigations unfairly impacting employee careers and reputations, workplaces becoming more confrontational, negative impacts on employee retention and reduction in productivity.?

The report ultimately found that positive deterrents are proactive measures such as employee-morale events, more thorough onboarding, ongoing data security training and education, upward feedback and work-life balance programs.?

Negative deterrents check on and constrain employee behavior. This can include broad tools and solutions that block users from engaging with, accessing or sharing content — all of which can result in a more reactive environment.

Successful programs

The study developed the holistic insider risk management index (HIRMI), which identified three types of organizational risk management: “fragmented,” “evolving” and “holistic.”?

Fragmented organizations (or one-third self-identified in the survey) recognize the need for insider risk programs but are often misaligned on success measures. They see value in positive deterrents that reduce risk but have low current usage. They also think they understand what’s required to lower insider risk, but do not commit resources or gain company-wide buy-in, according to the survey.

By contrast, in holistic programs, privacy controls are used in the early stages of investigations. Holistic organizations get more buy-in from other departments such as legal, HR or compliance teams, per the survey. Leaders at holistic organizations also agreed that training and education are vital to proactively addressing and reducing insider risks.?

Other key characteristics of holistic insider risk management include more frequent use of positive deterrents and integrated tool usage.?

And, the tools deemed most useful in preventing insider risk:?

? Extended detection and response (XDR)

? Network detection and response (NDR)

? Privileged access management

? User activity monitoring

? Incident threat management

? Endpoint detection and response (EDR)

? Security and information event management

? User and entity behavioral analysis

Holistic versus fragmented

The study found that 29% of organizations treated insider risk in a “holistic” way. And, more than 90% of those categorized as holistic said a key element to success is striking a balance between employee privacy and company security.?

The ultimate key to establishing a holistic insider risk management program is building trust, said Arsenault. This means collaborating across functions, increasing employee training and awareness, and having strong privacy controls to ensure that employees feel respected and invested.?

“It’s critical for organizations to address insider risk. But it’s just as important that they do so in the right way,” said Arsenault.?

He added that, “the best risk management programs aren’t focused on constraining employee behavior. They’re focused on building trust, balancing security and privacy, and educating and empowering their workforce.”??