Amazon launches Q, Pika raises $55M, and more (11.28.23)

Author

Marshall Kirkpatrick
November 28, 2023

Friends, here are today's top stories in the AI community, plus an Early Stage highlight from a startup participating in AWS re:Invent this week.

As always, I’m selecting from stories rising to the top of our data tracking AI community engagement. I hope these summaries are useful and interesting. Please continue to share them with friends and co-workers! Thank you to all who have.

And now here’s today’s news

Marshall Kirkpatrick, Editor

First impacted: Enterprise knowledge workers
Time to impact: Medium

Amazon has introduced Amazon Q, an AI chatbot, developed by its cloud computing division, to aid employees in workplace tasks "such as summarizing strategy documents, filling out internal support tickets and answering questions about company policy." (NYT) Q is designed to provide enhanced security and privacy, and can interact with corporate data not stored on Amazon's servers, including platforms like Slack and Gmail. [Amazon Introduces Q, an A.I. Chatbot for Companies] Share by email

First impacted: New video content creators, Animators
Time to impact: Medium

Pika, a video-making platform, has launched Pika 1.0, a significant upgrade that includes a new AI model designed to generate and edit videos in various styles, including 3D animation, anime, cartoon, and cinematic. In addition to this product upgrade, Pika has also secured $55 million in funding from a pretty remarkable list of AI investors. Despite being only six months old, Pika says that it has already amassed a community of half a million users, producing millions of videos on a weekly basis. [ANNOUNCING PIKA] Share by email

First impacted: Application developers
Time to impact: Medium

Stability AI has introduced SDXL Turbo, a model that generates images from text in real time, as you type. It's impressively fast, though the quality doesn't appear to be what you get if you're willing to wait 30 seconds for DALL*E from OpenAI. The SDXL model utilizes a method known as Adversarial Diffusion Distillation (ADD). According to Stability AI, SDXL Turbo has shown superior performance and faster inference speed in blind tests compared to other models. The company has made this model available for testing on its image editing platform, Clipdrop. [Introducing SDXL Turbo: A Real-Time Text-to-Image Generation Model — Stability AI] Share by email

First impacted: Radiologists, Healthcare IT Professionals
Time to impact: Medium to long

At the 2023 Conference on Empirical Methods in Natural Language Processing, a paper titled "Exploring the Boundaries of GPT-4 in Radiology" was presented by researchers in partnership with Nuance, a Microsoft subsidiary. The paper emphasized the potential of the AI model in healthcare, especially radiology. The researchers claim that GPT-4 showed a 10-percent absolute improvement over existing models in certain tasks. [GPT-4's potential in shaping the future of radiology] Share by email

First impacted: Machine Learning Engineers, Data Scientists
Time to impact: Short

The Keras machine learning framework has launched its 3.0 version after five months of public beta testing. The new version allows users to run their workflows on JAX, TensorFlow, or PyTorch, and optimizes performance by dynamically choosing the best backend for a model. [Keras: Deep Learning for humans] Share by email

First impacted: Artificial Intelligence Researchers, Software Developers
Time to impact: Medium

Microsoft Research says it has successfully trained a small transformer model to accurately perform arithmetic tasks, even with large numbers. This is typically a difficult task for such models due to their dependence on positional information. The model reportedly excelled in 15-digit multiplication and addition tasks, a significant improvement over typical training results where models struggle with 4-digit multiplication and 3-digit addition. [Positional Description Matters for Transformers Arithmetic] Share by email

First impacted: Machine learning researchers, AI technology developers
Time to impact: Medium to long

The Forty-first International Conference on Machine Learning 2024 conference has announced a new track for position papers, designed to foster idea sharing in the machine learning community. Example topics include the strength and limitations of LLMs, open vs closed source, copyright, privacy, ethics, misinformation, and social impact. [ICML 2024 Call For Position Papers] Share by email

Early Stage: Big News From Smaller Companies

First impacted: Cloud Security Engineers, IT Administrators
Time to impact: Short

Security startup Tamnoon has integrated with the AWS Security Hub to improve cloud security remediation, focusing on prioritization, remediation, execution, and prevention. The company's AI platform identifies and ranks crucial security alerts, formulates custom execution playbooks, and automates encryption for new EC2 instances to decrease the chances of future security incidents. [From Prioritization to Prevention: Tamnoon’s Integration with AWS Security Hub] Share by email

That’s it! More AI news tomorrow.