☕️ Stability Makes a Stand

AI Tangle Newsletter

The latter part of the week is full of exciting news, from Groq's rapid processors that leave the competition in the dust to Intel-Microsoft's co-op custom chip venture. Meanwhile, Samsung remains tenacious in bringing Galaxy AI to millions of phones, and Google gives some insight into Gemini with the public release of Gemma. Join us at AI Tangle as we untangle this week's happenings.

Stability announces Stable Diffusion 3 as an answer to Sora and Gemini​

With the monstrous launch of both Sora and Gemini, Stability answered in kind with ​Stable Diffusion 3​ (SD3), its latest and most powerful version of the company's image generation AI model with updated tooling to make the model more accessible. Additionally, Stability Diffusion CEO and founder Emad Mostaque posted on X/Twitter that, theoretically, SD3 is multimodal capable, meaning that future releases could include audio and video generation.

​

How does it work?

SD3 leverages an updated "diffusion transformer," a technique similar to the principles OpenAI's Sora uses, and a "flow transformer" technique that similarly improves quality. SD3 comes in a variety of parameter ranges with the intent of being able to run on many different hardware setups, from the lightweight 800 million parameters to the more heavy-duty 8 billion. The company also emphasizes safety in its announcement, claiming that numerous safeguards have been implemented to prevent bad actors from abusing the model, though their exact nature is unknown.

Groq, the AI language model platform that's built different, built fast​

​Groq, not to be confused with Elon Musk's Grok AI, is a company that makes custom computer hardware designed for running AI - and they want to make it better and faster than anyone else, up to 75x faster than the average human can type. The way it achieves this is by developing and utilizing the company's specialized high-performance solutions and processors, which the company calls Large Processing Units, or LPUs for short. While Groq isn't developing any models on its own currently, the ones made by others available on its site respond like lightning.

​

​Intel and Microsoft shake hands to develop a new custom chip​

During its Intel Foundry event, the company announced that it would be partnering up with Microsoft to develop and produce custom chips in a deal that could be worth more than $15 billion. With Microsoft as the designer for the chip and Intel as the producer, the duo will leverage Intel's 18A architecture, something that has been on Intel's roadmap for a while now. Though neither company have specified what this chip would be used for, Bloomberg has noted that Microsoft has been planning in-house designs for its tech revolving around processors and AI.

​

​Google releases two versions of Gemma, the research predecessor of Gemini, to the public​

In a recent announcement on Wednesday, Google released Gemma 2B and Gemma 7B to the public, a pair of open-source language models that went into the research of Gemini. Though these models lack complexity, they more than make up for it with their lightweight and agile nature bundled with a "responsible AI toolkit," making them fit for smaller tasks. Both models will be available with a commercial license regardless of scale or what purpose Gemma would be used for, with obvious exceptions such as weapons development.

​

​Samsung's Galaxy AI will be rolling out to more devices than the S24​

​At this year's Galaxy Unpacked in January, Samsung's top-of-the-line S24 was required to access Galaxy AI's suite of features, being the only series with the option. However, starting in March, that will no longer be the case, as Samsung will introduce the full range of Galaxy AI features to the S23 series, recent foldable devices and tablets as part of the One UI 6.1 update in Samsung's to ship Galaxy AI to 100 million phones by the end of the year. Though Galaxy AI has been called a "mixed bag" by many, the option will be there - including the one to turn it off if it's not your cup of tea.

​

​Google disabled Gemini's image generation capabilities over racial inaccuracies​

After users on social media criticized Gemini for generating images of the U.S. Founding Fathers and others as a different ethnicity altogether, Google announced that it would be temporarily pausing Gemini's image generation while the company works on fixing the problem. In an X/Twitter post, a spokesperson for Google said that while Gemini can "generate a wide range of people," they couldn't deny that it is currently "missing the mark," stating that the company would be "re-releasing a new and improved version soon."

​

​Google DeepMind alumni startup Bioptimus raises $35 million in funding to develop biology-focused AI​

​Bioptimus, a biology-focused generative AI startup in Paris with many promising talents from the Google DeepMind Research team, recently emerged with $35 million in a round of seed funding, backed by venture firms like Sofinnova Partners, Frst, Bpifrance’s Large Venture fund, and more. The goal of the startup is to exclusively work on a universal biology-centered AI foundational model that would fit every scale of the field, from molecules to tissues, and partner with Amazon Web Services to potentially train Bioptimus' expansive model there.

Humanic - Humanic is an AI-enhanced platform that empowers revenue teams by providing valuable insights into potential buyers. It leverages Personality AI to identify early adopters and provide customer insights.

​Delibr - Delibr is a lightweight document-writing solution with generative AI baked into its core to help you and your team go from idea and feedback to a crisp document.

​Lensa - Branding itself as "an influencers' best kept secret," Lensa is an AI-powered photo enhancement tool and editor that offers a variety of features centered around boosting social media presence.

​Lindy - Put simple and repetitive processes on autopilot with the help of Lindy's simple and intuitive AI agent system that allows you to build your no-code agents in 5 minutes.

The Key to Limitless Clean Energy (2-min read)

Researchers harness AI to forecast plasma instabilities in nuclear fusion, a breakthrough that could pave the way for clean, limitless energy, which begs the question - how far is commercial viability away?

​

AGI and The Simulation (12-min watch)

Elon Musk and Neil deGrasse Tyson discuss the potential intersection of living in a simulation and the development of artificial general intelligence - Wes Roth, a news channel dedicated to emerging tech and artificial intelligence breakthroughs, discusses it.