AI investments are not finished and Anthropic is gaining a lot of attention, after the 4B$ investment by Amazon I reported in September post, in October Google added 500M$ with the promise to add 1.5B$ more. Since Anthropic tailors their product specifically for corporate and enterprise-level solutions this must come to no surprise. Currently Microsoft dominates this area and the last quarter report from the cloud business of Google testify that they’re facing difficulties, Amazon on the other hand, while strong on cloud still doesn’t have a tailored generative AI such as Azure OpenAI. Notably, we’re seeing a growing trend in institutional adoption of AI products/services; thus, Anthropic may be perfectly poised to try to fill the gap.
Another interesting news is related to the OpenAI multi modality stake, some subscribers (alas not yet me) have already received the integrated multi modal interface, where not only you can input text, voice or images, but you can also get back images automatically perform online searches or use the proper data analytics without switching interface.
AI regulation is an important and urgent topic, President Biden issued an Executive Order, while old EU is still trying to find a common position, This is sad and unfortunate, we’re definitely too slow and we’re under financing AI (5B€ compared to 50B$ in US). Anyway the Biden administration Executive Order on Safe, Secure, and Trustworthy AI tries to balance national security with consumer rights—creating some of the first international guardrails that could be fortified by global legislation. The Order mandates comprehensive actions across various sectors, from requiring stringent safety tests for powerful AI systems to establishing an advanced cybersecurity program (full list here). Notably, it requires AI developers to share safety results, and the National Institute of Standards and Technology to create standards to ensure AI tools are safe and secure before public release. Furthermore, It emphasizes the importance of protecting Americans’ privacy and supporting consumers by requiring The Commerce Department to label all AI-generated content to protect the mental health of US consumers. Lastly, the act encourages responsible innovation, competition, and entrepreneurship by building upon previous commitments from 15 big-tech companies (OpenAI, Microsoft, X & more).
Speaking of regulations, this month Azure AI Content Safety entered GA. This is a service that enables developers to build safer online environments by detecting and assigning severity scores to unsafe images and text across content categories and languages. These capabilities empower businesses to effectively prioritize and streamline the review of both human and AI-generated content, enabling the responsible development of next-generation AI applications.
Speaking of Azure, starting in October we can fine tune ChatGPT model (aka GPT-3.5-Turbo). Also You can now utilize two new base inference models (Babbage-002 and Davinci-002) and fine-tune Babbage-002 and Davinci-002, too. Label pixels in images through Semantic Segmentation: You can now add tags or labels to individual pixels within images, leverage the vendor workforce in labeling the images, and label categories through hierarchical labeling. Also OpenAI Whisper model has landed on Azure OpenAI in preview.
OpenAI’s Forthcoming Attractiveness: Lower Costs, Higher Capability
OpenAI is setting the stage for significant updates aimed at enticing developers by making it more cost-effective and quicker to create applications leveraging its AI models. This initiative includes adding memory storage to developer tools, potentially cutting costs by up to 20-fold. New tools like vision capabilities are on the horizon, expanding use cases across various fields. The anticipated stateful API and vision API are steps towards reducing costs and enhancing multimedia processing respectively. These improvements, set to unveil at OpenAI’s inaugural developer conference, are part of a broader strategy to bolster its appeal amidst a competitive landscape1.
NSA’s AI Security Center: A Shield Against Cyber Threats
The National Security Agency (NSA) is establishing an Artificial Intelligence Security Center to bolster the defense and intelligence sectors against evolving cyber threats. This initiative, announced by the outgoing director, aligns with the NSA’s proactive stance to secure AI technologies from potential misuse, while promoting its safe adoption within national security apparatus. The collaborative effort will engage private and international entities, focusing on deterring adversarial threats, chiefly from China and Russia, and ensuring that the U.S. maintains its AI advantage in the face of potential election interferences and cyber operations1.
Image generation competition is ramping up. Google and Adobe launch their solution following, among the others, DALL-E 3 and Bing Image Creator. Google’s AI search experiment, SGE, embarks on a multimodal venture, introducing an AI image generating tool. This innovation, fueled by Google’s Imagen text-to-image diffusion models, allows users to depict desired images, which SGE swiftly translates into four varied visual options1. Adobe introduced Firefly Image 2 Model, currently in beta, which is poised to redefine image generation through enhanced model capacity and speed, elevating both quality and efficiency compared to its predecessor1. This innovative tool, powered by AI, ushers in a plethora of advanced features, bridging the gap between creative ideas and photorealistic images2.
Lastly there are a couple of tools I’m particularly intrigued with I want to try out:
- Guidde provides AI-generated documentation that promises to help explain the most complex tasks and rubooks at AI speed.
- Grafts that promises to address many common use cases such as
- Use prompts to cleanup data
- Summarize reports in seconds with text summarization
- Deliver more accurate and relevant search results
- Enable visual search to shop by image
- Moderate content seamlessly with toxicity detection
- Organize data faster with clustering algorithms
- Automatically categorize customer support tickets
- Analyze customer sentiment to understand feedback
Quae Nocent Docent 