Elon Musk’s Grok launched a brand new AI image-generation characteristic on Tuesday evening that, identical to the AI chatbot, has only a few safeguards. That implies you’ll be able to generate faux photographs of Donald Trump smoking marijuana at the Joe Rogan display, for instance, and add it instantly to the X platform. Nevertheless it’s no longer in point of fact Elon Musk’s AI corporate powering the insanity; somewhat, a brand new startup — Black Woodland Labs — is the outfit in the back of the debatable characteristic.
The collaboration between the 2 was once printed Tuesday, when xAI introduced it’s running with Black Woodland to energy Grok’s picture generator the use of its FLUX.1 fashion. An AI picture and video startup that introduced on August 1, Black Woodland seems to sympathize with Musk’s imaginative and prescient for Grok as an “anti-woke chatbot,” with out the stern guardrails present in OpenAI’s Dall-E or Google’s Imagen. The social media website is already flooded with outrageous photographs from the brand new characteristic.
Black Woodland Labs is based totally in Germany and not too long ago got here out of stealth with $31 million in seed investment, led via Andreessen Horowitz, consistent with a press free up. Different notable traders come with Y Combinator CEO Garry Tan and previous Oculus CEO Brendan Iribe. The startup’s co-founders, Robin Rombach, Patrick Esser, and Andreas Blattmann, had been previously researchers who helped create Balance AI’s Strong Diffusion fashions.
In keeping with Synthetic Research, Black Woodland Lab’s FLUX.1 fashions surpass Midjourney’s and OpenAI’s AI picture turbines in the case of high quality, no less than as ranked via customers of their picture enviornment.
The startup says it’s “making our fashions to be had to a large target audience,” with open supply AI image-generation fashions on Hugging Face and GitHub. The corporate says it plans to create a text-to-video fashion quickly, as properly.
Black Woodland Labs didn’t in an instant reply to TechCrunch’s request for remark.
Oh my god. Grok has completely no filters for its picture era. This is likely one of the maximum reckless and irresponsible AI implementations I have ever noticed. percent.twitter.com/oiyRhW5jpF— Alejandra Caraballo (@Esqueer_) August 14, 2024
In its release free up, the corporate says it goals to “make stronger accept as true with within the protection of those fashions”; on the other hand, some would possibly say the flood of its AI generated photographs on X Wednesday did the other. Many photographs customers had been ready to create the use of Grok and Black Woodland Labs’ device, equivalent to Pikachu conserving an attack rifle, weren’t ready to be re-created with Google or OpenAI’s picture turbines. There’s indubitably certainly that copyrighted imagery was once used for the fashion’s coaching.
That’s more or less the purpose
This loss of safeguards is most likely a big reason why Musk selected this collaborator. Musk has made transparent that he believes safeguards in reality make AI fashions much less secure. “The chance of coaching AI to be woke — in different phrases, lie — is fatal,” stated Musk in a tweet from 2022.
Board director of Black Woodland Labs, Anjney Midha, posted on X a chain of comparisons between photographs generated on day one among release via Google Gemini and Grok’s Flux collaboration. The thread highlights Google Gemini’s well-documented problems with developing traditionally correct photographs of other folks, in particular via injecting racial range into photographs inappropriately.
“I’m satisfied @ibab and crew took this critically and made the appropriate selection,” stated Midha in a tweet, relating to FLUX.1’s seeming avoidance of this factor (and citing the account of xAI lead researcher Igor Babuschkin).
On account of this flub, Google apologized and became off Gemini’s skill to generate photographs of other folks in February. As of these days, the corporate nonetheless doesn’t let Gemini generate photographs of other folks.
A firehose of incorrect information
This normal loss of safeguards may just purpose issues for Musk. The X platform drew complaint when AI-generated deepfake particular photographs representing Taylor Swift went viral at the platform. But even so that incident, Grok generates hallucinated headlines that seem to customers on X nearly weekly.
Simply ultimate week, 5 secretaries of state recommended X to forestall spreading incorrect information about Kamala Harris on X. Previous this month, Musk reshared a video that used AI to clone Harris’ voice, making it seem as though the vp admitted to being a “range rent.”
Musk turns out intent on letting incorrect information like this pervade the platform. Through permitting customers to submit Grok’s AI photographs, which appear to lack any watermarks, immediately at the platform, he’s necessarily opened a firehose of incorrect information pointed at everybody’s X newsfeed.
