AI Top-of-Mind for 6.24.24 — ROI
Today: Use cases and ROI from CB Insights, 3D sculptures, Apple and the EU, Claude 3.5 Sonnet Artifacts, Alibaba’s Qwen2, Public Opinion from the Stanford AI Index Report and Lyft arrival times decoded.
Busy weekend so let’s dive in! Top-of-mind is new research by ‘CB Insights’ showing how enterprises are actually deploying AI and how they determine the ROI. The firm looked at multiple verticals and use cases, with the following bubbling up:
· Content creation
· Customer support assistance
· Natural language search
· Design and data creation
· Code generation
· Document automation
An ROI example:
This H2O.ai customer breaks down the ROI calculation further. To them, the $125K for the H2O AI Cloud (HAIC) product is justified by estimating 4 weeks of time saved annually per data scientist. With a team of 10 data scientists, this equates to around $500K in savings — a 4x return on the investment. The customer notes these savings max out their budget for such tools, showing they are spending at the limit of where they see a positive ROI.
Their conclusions:
But most interviews reveal a common theme: the rapidly evolving genAI market will significantly shape the future of AI development platforms. As businesses plan to develop their own AI applications, they are closely monitoring their vendors’ genAI roadmaps to inform their decision-making.
On creative, much of the coverage is on creating images or videos, 2D or 3D. But taking it a step further, what if AI could help with 3D sculptures. Robert Gonsalves writing for ‘UX Collective’ describes his journey in creating a modular piece with pluggable parts via DALL-E 3, 3dmaker.ai, and Blender.
Turning to corporate, an update by ‘The Information’ on Stability AI. The article introduces the company’s new CEO, Prem Akkaraju, as well as a ‘bailout’ by a group led by Sean Parker. Remember that the company is responsible for Stable Diffusion, an open-source AI image generator, but revenues were only $5M last quarter. Not to be left out, Mistral AI last week formally announced a $645 million round at a $6 billion valuation. Not too shabby for a company a year old.
Two weeks into Apple’s AI launch, and an update, this time involving the EU. The word is, reported by ‘The Information’ and others, that Apple will hold back many of its new AI features:
According to Bloomberg, Apple blamed its decision on rules imposed by the Digital Markets Act in Europe. Is that a valid excuse? Who knows? In a way, it doesn’t matter.
In this piece in The Wall Street Journal in January, columnist Greg Ip argued Europe was falling behind the U.S. and China in tech sectors such as AI and electric vehicles thanks to its rules. Apple’s decision suggests Europeans may soon have to be content with substandard consumer tech services as well.
We’ll see how this all plays out.
As an update to Anthropic’s Claude 3.5 Sonnet launch, Artifacts:
· Today, we’re also introducing Artifacts on Claude.ai, a new feature that expands how users can interact with Claude. When a user asks Claude to generate content like code snippets, text documents, or website designs, these Artifacts appear in a dedicated window alongside their conversation. This creates a dynamic workspace where they can see, edit, and build upon Claude’s creations in real-time, seamlessly integrating AI-generated content into their projects and workflows.
· This preview feature marks Claude’s evolution from a conversational AI to a collaborative work environment.
Keeping to Sonnet, Jim Clyde Monge writing in ‘Generative AI’ dives deeper into the new model and its capabilities. One area he focuses on is its vision capabilities and chart making. He also describes ‘Artifacts,’ introduced above.
Turning to Asia, Alibaba has been quiet lately, but has just released its new ‘Qwen2’ LLM. Thomas Ried in ‘AI Advances’ looks at its benchmarks in comparison to other leading models, including Llama 70B. His conclusion is that it still trails Llama 70B and GPT4-based models. From Alibaba:
Closing out the model front, a primer by Fabio Matricardi in ‘Generative AI’ on how to implement RAG on a PC using Llama.CPP. He also offers some good background on embeddings and context windows, and the use of RAG when working with models that support fewer tokens.
A last look at the ‘Stanford AI Index Report,’ this time Public Opinion. Some eye-opening stats on views as to the benefits of AI, its impact on jobs, and overall angst. The first chart is interesting in that less developed countries but with major internet presence as still probably in their ‘AI honeymoon’ phase vs more developed ones that are turning to issues such as privacy and job loss.
Lastly, if you use Lyft, and ever wondered as to how the algorithms calculate expected arrival times, ‘Lyft Engineering’ pulls back to curtain. The author introduces the concept of reliability, such as the likelihood that the driver will arrive when requested.
