• AI Disruptor
  • Posts
  • Stability AI undergoes restructuring, cuts 10% of workforce amid leadership changes

Stability AI undergoes restructuring, cuts 10% of workforce amid leadership changes

+ Llama 3: Meta's latest open AI models aim to outperform competitors

Welcome to AI Disruptor! If you want to join our growing community of readers, click the button below.


🚨 Stability AI undergoes restructuring, cuts 10% of workforce amid leadership changes

🦙 Llama 3: Meta's latest open AI models aim to outperform competitors

+ Other quick AI developments you should know…

Main Story: Stability AI, the startup behind the popular Stable Diffusion text-to-image model, has laid off more than 20 employees, amounting to about 10% of its global headcount, in an effort to "right-size" the business after a period of unsustainable growth.

Dive deeper:

  • The company's newly appointed co-CEOs Shan Shan Wong and Christian Laforte notified employees of the restructuring in an internal memo.

  • The layoffs follow the departure of controversial former CEO Emad Mostaque, who left the company last month amid reports questioning his credentials and the nature of a partnership with Amazon.

  • Affected employees are mostly on the operational side of the business and have been notified of their redundancies.

  • Despite the layoffs, Stability AI continues to operate as normal and has recently announced developer APIs for its Stable Diffusion 3 AI model.

Industry impact: The layoffs at Stability AI highlight the challenges faced by rapidly growing AI startups as they navigate the competitive landscape and aim to establish sustainable business models. The company's restructuring efforts and the appointment of new leadership suggest a focus on stabilizing operations and maintaining its position as a leading player in the text-to-image AI space.

Stability AI memo to team

Llama 3 benchmarks

Main Story: Meta has released the latest entry in its Llama series of open generative AI models, Llama 3, debuting two models in the new family: Llama 3 8B (8 billion parameters) and Llama 3 70B (70 billion parameters). Meta claims these models are among the best-performing generative AI models available today for their respective parameter counts.

Dive deeper:

  • Llama 3 models outperform other open models like Mistral's Mistral 7B and Google's Gemma 7B on various benchmarks, including MMLU, ARC, and DROP.

  • Llama 3 70B is competitive with flagship generative AI models, beating Google's Gemini 1.5 Pro on some benchmarks and outperforming Anthropic's Claude 3 Sonnet on others.

  • The Llama 3 models were trained on a collection of 15 trillion tokens, seven times the size of the Llama 2 training set, and include more code and non-English data.

  • Meta has developed new data-filtering pipelines and updated its generative AI safety suites to address toxicity and bias issues in the Llama 3 models.

Industry impact: The release of the Llama 3 models underscores Meta's commitment to advancing open generative AI and its ambition to compete with leading players in the field. By making the models widely available and optimizing them for various hardware platforms, Meta aims to drive innovation and adoption across the AI ecosystem. However, the company's restrictions on using Llama models for training other generative models and its discretionary licensing for large-scale app developers may limit the truly open nature of these releases.

Some other quick developments you should know…

A fake, partially AI-generated movie trailer featuring Henry Cavill as James Bond has gone viral on YouTube, racking up over 2.5 million views. Created by KH Studio, the misleading video uses a mix of AI imagery and real movie footage to fuel fan fantasies of Cavill as the next 007.

Meta is rolling out real-time AI image generation in beta for WhatsApp users in the US. As users type text-to-image prompts in chats with Meta AI, they'll see the image change in real-time as more details are added. Meta's Llama 3 model powers the feature, producing sharper and higher-quality images compared to previous iterations.

Meta's newly introduced AI agents have been spotted engaging in unusual interactions with Facebook users, causing confusion. In one instance, a chatbot with the official Meta AI label inserted itself into a private group conversation, claiming to have a child in a specific school district before apologizing and admitting it was just a language model. Meta acknowledges the technology is new and may not always provide the intended response.

What did you think of this edition of AI Disruptor?

Your feedback helps us create a better newsletter.

Login or Subscribe to participate in polls.

Join the conversation

or to participate.