Connect with us

Government

This startup is setting a DALL-E 2-like AI free, consequences be damned

DALL-E 2, OpenAI’s powerful text-to-image AI system, can create photos in the style of cartoonists, 19th century daguerreotypists, stop-motion animators…

Published

on

DALL-E 2, OpenAI’s powerful text-to-image AI system, can create photos in the style of cartoonists, 19th century daguerreotypists, stop-motion animators and more. But it has an important, artificial limitation: a filter that prevents it from creating images depicting public figures and content deemed too toxic.

Now an open source alternative to DALL-E 2 is on the cusp of being released, and it’ll have no such filter.

London- and Los Altos-based startup Stability AI this week announced the release of a DALL-E 2-like system, Stable Diffusion, to just over a thousand researchers ahead of a public launch in the coming weeks. A collaboration between Stability AI, media creation company RunwayML, Heidelberg University researchers, and the research groups EleutherAI and LAION, Stable Diffusion is designed to run on most high-end consumer hardware, generating 512×512-pixel images in just a few seconds given any text prompt.

Stable Diffusion sample outputs.

“Stable Diffusion will allow both researchers and soon the public to run this under a range of conditions, democratizing image generation,” Stability AI CEO and founder Emad Mostaque wrote in a blog post. “We look forward to the open ecosystem that will emerge around this and further models to truly explore the boundaries of latent space.”

But Stable Diffusion’s lack of safeguards compared to systems like DALL-E 2 poses tricky ethical questions for the AI community. Even if the results aren’t perfectly convincing yet, making fake images of public figures opens a large can of worms. And making the raw components of the system freely available leaves the door open to bad actors who could train them on subjectively inappropriate content, like pornography and graphic violence.

Creating Stable Diffusion

Stable Diffusion is the brainchild of Mostque. Having graduated from Oxford with a Masters in mathematics and computer science, Mostque served as an analyst at various hedge funds before shifting gears to more public-facing works. In 2019, he co-founded Symmitree, a project that aimed to reduce the cost of smartphones and internet access for people living in impoverished communities. And in 2020, Mostque was the chief architect of Collective & Augmented Intelligence Against COVID-19, an alliance to help policymakers make decisions in the face of the pandemic by leveraging software.

He co-founded Stability AI in 2020, motivated both by a personal fascination with AI and what he characterized as a lack of “organization” within the open source AI community.

Stable Diffusion Obama

An image of former president Barrack Obama created by Stable Diffusion.

“Nobody has any voting rights except our 75 employees — no billionaires, big funds, governments or anyone else with control of the company or the communities we support. We’re completely independent,” Mostaque told TechCrunch in an email. “We plan to use our compute to accelerate open source, foundational AI.”

Mostque says that Stability AI funded the creation of LAION 5B, an open source, 250-terabyte dataset containing 5.6 billion images scraped from the internet. (“LAION” stands for Large-scale Artificial Intelligence Open Network, a nonprofit organization with the goal of making AI, datasets and code available to the public.) The company also worked with the LAION group to create a subset of LAION 5B called LAION-Aesthetics, which contains AI-filtered images ranked as particularly “beautiful” by testers of Stable Diffusion.

The initial version of Stable Diffusion was based on LAION-400M, the predecessor to LAION 5B, which was known to contain depictions of sex, slurs and harmful stereotypes. LAION-Aesthetics attempts to correct for this, but it’s too early to tell to what extent it’s successful.

Stable Diffusion

A collage of images created by Stable Diffusion.

In any case, Stable Diffusion builds on research incubated at OpenAI as well as Runway and Google Brain, one of Google’s AI R&D divisions. The system was trained on text-image pairs from LAION-Aesthetics to learn the associations between written concepts and images, like how the word “bird” can refer not only to bluebirds but parakeets and bald eagles, as well as more abstract notions.

At runtime, Stable Diffusion — like DALL-E 2 — breaks the image generation process down into a process of “diffusion.” It starts with pure noise and refines an image over time, making it incrementally closer to a given text description until there’s no noise left at all.

Boris Johnson Stable Diffusion

Boris Johnson wielding various weapons, generated by Stable Diffusion.

Stability AI used a cluster of 4,000 Nvidia A1000 GPUs running in AWS to train Stable Diffusion over the course of a month. CompVis, the machine vision and learning research group at Ludwig Maximilian University of Munich, oversaw the training, while Stability AI donated the compute power.

Stable Diffusion can run on graphics cards with around 5GB of VRAM. That’s roughly the capacity of mid-range cards like Nvidia’s GTX 1660, priced around $230. Work is underway on bringing compatibility to AMD MI200’s data center cards and even MacBooks with Apple’s M1 chip (although in the case of the latter, without GPU acceleration, image generation will take as long as a few minutes).

“We have optimized the model, compressing the knowledge of over 100 terabytes of images,” Mosque said. “Variants of this model will be on smaller datasets, particularly as reinforcement learning with human feedback and other techniques are used to take these general digital brains and make then even smaller and focused.”

Stability AI Stable Diffusion

Samples from Stable Diffusion.

For the past few weeks, Stability AI has allowed a limited number of users to query the Stable Diffusion model through its Discord server, slowing increasing the number of maximum queries to stress-test the system. Stability AI says that over 15,000 testers have used Stable Diffusion to create 2 million images a day.

Far-reaching implications

Stability AI plans to take a dual approach in making Stable Diffusion more widely available. It’ll host the model in the cloud, allowing people to continue using it to generate images without having to run the system themselves. In addition, the startup will release what it calls “benchmark” models under a permissive license that can be used for any purpose — commercial or otherwise — as well as compute to train the models.

That will make Stability AI the first to release an image generation model nearly as high-fidelity as DALL-E 2. While other AI-powered image generators have been available for some time, including Midjourney, NightCafe and Pixelz.ai, none have open-sourced their frameworks. Others, like Google and Meta, have chosen to keep their technologies under tight wraps, allowing only select users to pilot them for narrow use cases.

Stability AI will make money by training “private” models for customers and acting as a general infrastructure layer, Mostque said — presumably with a sensitive treatment of intellectual property. The company claims to have other commercializable projects in the works, including AI models for generating audio, music and even video.

Stable Diffusion Harry Potter

Sand sculptures of Harry Potter and Hogwarts, generated by Stable Diffusion.

“We will provide more details of our sustainable business model soon with our official launch, but it is basically the commercial open source software playbook: services and scale infrastructure,” Mostque said. “We think AI will go the way of servers and databases, with open beating proprietary systems — particularly given the passion of our communities.”

With the hosted version of Stable Diffusion — the one available through Stability AI’s Discord server — Stability AI doesn’t permit every kind of image generation. The startup’s terms of service ban some lewd or sexual material (although not scantily-clad figures), hateful or violent imagery (such as antisemitic iconography, racist caricatures, misogynistic and misandrist propaganda), prompts containing copyrighted or trademarked material, and personal information like phone numbers and Social Security numbers. But Stability AI won’t implement keyword-level filters like OpenAI’s, which prevent DALL-E 2 from even attempting to generate an image that might violate its content policy.

Stable Diffusion women

A Stable Diffusion generation, given the prompt: “very sexy woman with black hair, pale skin, in bikini, wet hair, sitting on the beach.”

Stability AI also doesn’t have a policy against images with public figures. That presumably makes deepfakes fair game (and Renaissance-style paintings of famous rappers), though the model struggles with faces at times, introducing odd artifacts that a skilled Photoshop artist rarely would.

“Our benchmark models that we release are based on general web crawls and are designed to represent the collective imagery of humanity compressed into files a few gigabytes big,” Mostque said. “Aside from illegal content, there is minimal filtering, and it is on the user to use it as they will.”

Stable Diffusion Hitler

An image of Hitler generated by Stable Diffusion.

Potentially more problematic are the soon-to-be-released tools for creating custom and fine-tuned Stable Diffusion models. An “AI furry porn generator” profiled by Vice offers a preview of what might come; an art student going by the name of CuteBlack trained an image generator to churn out illustrations of anthropomorphic animal genitalia by scraping artwork from furry fandom sites. The possibilities don’t stop at pornography. In theory, a malicious actor could fine-tune Stable Diffusion on images of riots and gore, for instance, or propaganda.

Already, testers in Stability AI’s Discord server are using Stable Diffusion to generate a range of content disallowed by other image generation services, including images of the war in Ukraine, nude women, a Chinese invasion of Taiwan, and controversial depictions of religious figures like the Prophet Mohammed. Many of the results bear telltale signs of an algorithmic creation, like disproportionate limbs and an incongruous mix of art styles. But others are passable on first glance. And the tech, presumably, will continue to improve.

Nude women Stability AI

Nude women generated by Stable Diffusion.

Mostque acknowledged that the tools could be used by bad actors to create “really nasty stuff,” and CompVis says that the public release of the benchmark Stable Diffusion model will “incorporate ethical considerations.” But Mostque argues that — by making the tools freely available — it allows the community to develop countermeasures.

“We hope to be the catalyst to coordinate global open source AI, both independent and academic, to build vital infrastructure, models and tools to maximize our collective potential,” Mostque said. “This is amazing technology that can transform humanity for the better and should be open infrastructure for all.”

Stable Diffusion Zelensky

A generation from Stable Diffusion, with the prompt: “[Ukrainian president Volodymyr] Zelenskyy committed crimes in Bucha.”

Not everyone agrees, as evidenced by the controversy over “GPT-4chan,” an AI model trained on one of 4chan’s infamously toxic discussion boards. AI researcher Yannic Kilcher made GPT-4chan — which learned to output racist, antisemitic and misogynist hate speech — available earlier this year on Hugging Face, a hub for sharing trained AI models. Following discussions on social media and Hugging Face’s comment section, the Hugging Face team first “gated” access to the model before removing it altogether, but not before it was downloaded over a thousand times.

War in Ukraine Stability AI

“War in Ukraine” images generated by Stable Diffusion.

Meta’s recent chatbot fiasco illustrates the challenge of keeping even ostensibly safe models from going off the rails. Just days after making its most advanced AI chatbot to date, BlenderBot 3, available on the web, Meta was forced to confront media reports that the bot made frequent antisemitic comments and repeated false claims about former U.S. president Donald Trump winning reelection two years ago.

BlenderBot 3’s toxicity came from biases in the public websites that were used to train it. It’s a well-known problem in AI — even when fed filtered training data, models tend to amplify biases like photo sets that portray men as executives and women as assistants. With DALL-E 2, OpenAI has attempted to combat this by implementing techniques, including dataset filtering, that help the model generate more “diverse” images. But some users claim that they’ve made the model less accurate than before at creating images based on certain prompts.

Stable Diffusion contains little in the way of mitigations besides training dataset filtering. So what’s to prevent someone from generating, say, photorealistic images of protests, “evidence” of fake moon landings and general misinformation? Nothing really. But Mostque says that’s the point.

Stable Diffusion protest

Given the prompt “protests against the dilma government, brazil [sic],” Stable Diffusion created this image.

“A percentage of people are simply unpleasant and weird, but that’s humanity,” Mostque said. “Indeed, it is our belief this technology will be prevalent, and the paternalistic and somewhat condescending attitude of many AI aficionados is misguided in not trusting society … We are taking significant safety measures including formulating cutting-edge tools to help mitigate potential harms across release and our own services. With hundreds of thousands developing on this model, we are confident the net benefit will be immensely positive and as billions use this tech harms will be negated.”

Read More

Continue Reading

International

I’m headed to London soon for #EUBIO22. Care to join me?

Adrian Rawcliffe
It was great getting back to a live ESMO conference/webinar in Paris followed by a live pop-up event for the Endpoints 11 in Boston. We’re…

Published

on

Adrian Rawcliffe

It was great getting back to a live ESMO conference/webinar in Paris followed by a live pop-up event for the Endpoints 11 in Boston. We’re staying on the road in October with our return for a live/streaming EUBIO22 in London.

Kate Bingham

Silicon Valley Bank’s Nooman Haque and I are once again jumping back into the thick of it with a slate of virtual and live events on October 12. I’ll get the ball rolling with a virtual fireside chat with Novo Nordisk R&D chief Marcus Schindler, covering their pipeline plans and BD work.

After that I’ve teed up two webinars on mRNA research — with some of the top experts in Europe — and the oncology scene, building better CARs in Europe.

That afternoon, we’ll switch to a live/streaming hybrid event, with a chance to gather once again now that the pandemic has faded. I’ve recruited a panel of top biotech execs to look at surviving the crazy public market, with Adrian Rawcliffe, the CEO of Adaptimmune, SV’s Kate Bingham, Mereo CEO Denise Scots-Knight and Andrew Hopkins, chief of Exscientia.

Andrew Hopkins
Denise Scots-Knight

That will be followed by my special, live fireside chat with Susan Galbraith, the oncology R&D chief at AstraZeneca. And then we’ll turn to Nooman’s panel, where he’ll be talking with Katya Smirnyagina with Oxford Science Enterprises, Maina Bhaman with Sofinnova Partners and Rosetta Capital’s Jonathan Hepple about navigating the severe capital headwinds.

You can review the full schedule and buy tickets here and review everything we have planned. It will be a packed day. I hope to see you there. It’s been several years now since I’ve had a chance to meet people in the Golden Triangle. I’m very much looking forward to it.

Read More

Continue Reading

Government

We can turn to popular culture for lessons about how to live with COVID-19 as endemic

As COVID-19 transitions from a pandemic to an endemic, apocalyptic science-fiction and zombie movies contain examples of how to adjust to the new norm…

Published

on

By

An endemic means that COVID-19 is still around, but it no longer disrupts everyday life. (Shutterstock)

In 2021, conversations began on whether the COVID-19 pandemic will, or even can, end. As a literary and cultural theorist, I started looking for shifts in stories about pandemics and contagion. It turns out that several stories also question how and when a pandemic becomes endemic.


Read more: COVID will likely shift from pandemic to endemic — but what does that mean?


The 2020 film Peninsula, a sequel to the Korean zombie film, Train to Busan, ends with a group of survivors rescued and transported to a zombie-free Hong Kong. In it, Jooni (played by Re Lee) spent her formative years living through the zombie epidemic. When she is rescued, she responds to being informed that she’s “going to a better place” by admitting that “this place wasn’t bad either.”

Jooni’s response points toward the shift in contagion narratives that has emerged since the spread of COVID-19. This shift marks a rejection of the push-for-survival narratives in favour of something more indicative of an endemic.

Found within

Contagion follows a general cycle: outbreak, epidemic, pandemic and endemic. The determinants of each stage rely upon the rate of spread within a specified geographic region.

Etymologically, the word “endemic” has its origins with the Greek words én and dēmos, meaning “in the people.” Thus, it refers to something that is regularly found within a population.

Infectious disease physician Stephen Parodi asserts that an endemic just means that a disease, while still prevalent within a population, no longer disrupts our daily lives.

Similarly, genomics and viral evolution researcher Aris Katzourakis argues that endemics occur when infection rates are static — neither rising nor falling. Because this stasis occurs differently with each situation, there is no set threshold at which a pandemic becomes endemic.

Not all diseases reach endemic status. And, if endemic status is reached, it does not mean the virus is gone, but rather that things have become “normal.”

Survival narratives

We’re most likely familiar with contagion narratives. After all, Steven Soderbergh’s 2011 film Contagion, was the most watched film on Canadian Netflix in March 2020. Conveniently, this was when most Canadian provinces went into lockdown during the early stages of the COVID-19 pandemic.

A clip from the film Contagion showing the disease spreading throughout the world.

In survival-based contagion narratives, characters often discuss methods for survival and generally refer to themselves as survivors. Contagion chronicles the transmission of a deadly virus that is brought from Hong Kong to the United States. In response, the U.S. Centers for Disease Control is tasked with tracing its origins and finding a cure. The film follows Mitch Emhoff (Matt Damon), who is immune, as he tries to keep his daughter safe in a crumbling Minneapolis.

Ultimately, a vaccine is successfully synthesized, but only after millions have succumbed to the virus.

Like many science fiction and horror films that envision some sort of apocalyptic end, Contagion focuses on the basic requirements for survival: shelter, food, water and medicine.

However, it also deals with the breakdown of government systems and the violence that accompanies it.

A “new” normal

In contrast, contagion narratives that have turned endemic take place many years after the initial outbreak. In these stories, the infected population is regularly present, but the remaining uninfected population isn’t regularly infected.

A spin-off to the zombie series The Walking Dead takes place a decade after the initial outbreak. In the two seasons of The Walking Dead: World Beyond (2020-2021) four young protagonists — Hope (Alexa Mansour), Iris (Aliyah Royale), Silas (Hal Cumpston) and Elton (Nicolas Cantu) — represent the first generation to come of age within the zombie-infested world.

The four youth spent their formative years in an infected world — similar to Jooni in Peninsula. For these characters, zombies are part of their daily lives, and their constant presence is normalized.

The trailer for the second season of AMC’s The Walking Dead: World Beyond.

The setting in World Beyond has electricity, helicopters and modern medicine. Characters in endemic narratives have regular access to shelter, food, water and medicine, so they don’t need to resort to violence over limited resources. And notably, they also don’t often refer to themselves as survivors.

Endemic narratives acknowledge that existing within an infected space alongside a virus is not necessarily a bad thing, and that not all inhabitants within infected spaces desire to leave. It is rare in endemic narratives for a character to become infected.

Instead of going out on zombie-killing expeditions in the manner that occurs frequently in the other Walking Dead stories, the characters in World Beyond generally leave the zombies alone. They mark the zombies with different colours of spray-paint to chronicle what they call “migration patterns.”

The zombies have therefore just become another species for the characters to live alongside — something more endemic.

The Walking Dead, Fear the Walking Dead (2015-), Z Nation (2014-18), and many other survival-based stories seem to return to the past. In contrast, endemic narratives maintain a present and sometimes even future-looking approach.

Learning from stories

According to film producer and media professor Mick Broderick, survival stories maintain a status quo. They seek a “nostalgically yearned-for less-complex existence.” It provides solace to imagine an earlier, simpler time when living through a pandemic.

However, the shift from survival to endemic in contagion narratives provides us with many important possibilities. The one I think is quite relevant right now is that it presents us with a way of living with contagion. After all, watching these characters survive a pandemic helps us imagine that we can too.

Krista Collier-Jarvis does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

Read More

Continue Reading

International

Xi Reemerges In 1st Public Appearance After ‘Coup’ Rumors

Xi Reemerges In 1st Public Appearance After ‘Coup’ Rumors

So much for the "coup in China" and "Xi is missing" rumor mill of the past week,…

Published

on

Xi Reemerges In 1st Public Appearance After 'Coup' Rumors

So much for the "coup in China" and "Xi is missing" rumor mill of the past week, which at one point saw Chinese President Xi Jinping's name trending high on Twitter...

"Chinese President Xi Jinping visited an exhibition in Beijing on Tuesday, according to state television, in his first public appearance since returning to China from an official trip to Central Asia in mid-September – dispelling unverified rumours that he was under house arrest."

He had arrived in Samarkand, Uzbekistan on September 15 - and attended the days-long Shanghai Cooperation Organization (SCO) summit - where he met with Russian President Vladimir Putin, among others.

Xi is "back"...image via state media screenshot

Importantly, it had been his first foreign trip in two years. Xi had not traveled outside of the country since before the Covid-19 pandemic began.

But upon returning the Beijing, he hadn't been seen in the public eye since that mid-September trip, fueling speculation and rumors in the West and on social media. Some pundits floated the idea that he had been under "house arrest" amid political instability and a possible coup attempt.

According to a Tuesday Bloomberg description of the Chinese leader's "re-emergence" in the public eye, which has effectively ended the bizarre rumors

Xi, wearing a mask, visited an exhibition in Beijing on Tuesday about China's achievements over the past decade, state-run news outlet Xinhua reported. The Chinese leader was accompanied by the other six members of the Politburo Standing Committee, a sign of unity after rumors circulated on Twitter about a challenge to his power.

He'll likely cinch his third five-year term as leader at the major Chinese Communist party’s (CCP) meeting on October 16. The CCP meeting comes only once every half-decade.

What had added to prior rumors was the fact that the 69-year old Xi recently undertook a purge of key senior security officials. This included arrests on corruption charges of the former police chiefs of Shanghai, Chongqing and Shanxi.

More importantly, former vice minister of public security Sun Lijun and former justice minister Fu Zhenghua were also sacked and faced severe charges.

Concerning Sun Lijun, state media made this shocking announcement a week ago: "Sun Lijun, former Chinese vice minister of public security, was sentenced to death with a two-year reprieve for taking more than 646 million yuan of bribes, manipulating the stock market, and illegally possessing firearms, according to the Intermediate People's Court of Changchun in Northeast China's Jilin Province on Friday." The suspended death sentence means he'll spend life in prison.

Tyler Durden Wed, 09/28/2022 - 14:05

Read More

Continue Reading

Trending