AI Top-of-Mind for Jan 16
Top-of-mind is a detailed analysis by the IMF on AI job disruption. The survey looks at different types of economies depicted in the graphic below, job distribution, and then the impact from AI. From the blog:
In advanced economies, about 60 percent of jobs may be impacted by AI. Roughly half the exposed jobs may benefit from AI integration, enhancing productivity. For the other half, AI applications may execute key tasks currently performed by humans, which could lower labor demand, leading to lower wages and reduced hiring. In the most extreme cases, some of these jobs may disappear.
Link to the report.
Turning to models, Ignacio de Gregorio in ‘Towards AI’ details ‘Mamba,’ said to be a massive improvement in speed and cost vs today’s transformer-based models.
From the article:
· Just like the attention module sits at the core of the Transformer, the Selective State Space Model (Selective SSM) sits at the core of Mamba.
· Yes, but the key thing here is that Selective SSMs keep only the context that matters in memory, which means they achieve linear and constant complexities for training and inference.
· In other words, if the sequence doubles the cost of training doubles (Transformers quadruple) while the inference cost remains constant no matter the length!
Pretty cool! He goes on to note that although the approach shows superior performance for smaller models, it is still untested with the largest of LLMs.
Also on the model front, enhancements by Microsoft to better assist with enterprise AI adoption. From the ‘CodeX’ article:
Microsoft Azure introduces GPT-RAG, an Enterprise RAG Solution Accelerator designed specifically for deploying LLMs using the Retrieval Augmentation Generation (RAG) pattern within enterprise settings. It emphasizes a robust security framework and zero-trust principles to handle sensitive data carefully. In addition, GPT-RAG employs a Zero Trust Architecture Overview and features like Azure Virtual Network, Azure Front Door with Web Application Firewall, etc., ensuring data security.
The code repository is here.
And in case you didn’t catch it, more evidence of the stealth release of GPT 4.5, but still no word from OpenAI. In fact, some claim it is a hallucination.
As a follow-up on the recently announced OpenAI GPT store, some good guidance by The PyCoach in ‘Artificial Corner’ on how to best monetize any custom GPTs.
- Use custom actions: This is a feature that allows your GPT to connect to an API. Connecting to APIs gives your GPT new functionalities that others won’t be able to replicate unless they have access to the API (here you can see my tutorial on how to add custom action to your GPT)
- Use knowledge: Knowledge is a feature that allows you to add files to your GPT. Adding exclusive information could enrich your GPT and help it stand out from the pack. Just remember that files can be downloaded when the code interpreter is enabled.
On the more practical front, a good article by ‘The New Stack’ on how ITOps are already integrating Gen AI and three relevant use cases:
· Status update automation
· Incident postmortems
· Process automation
We’ve all heard of Microsoft’s Copilot, and some good thoughts from Philip Grabenhorst in ‘UX Collective’ as to how it can and should evolve into something much more useful, a ‘coach’. As he writes:
Instead of just giving us fish, Microsoft should employ its (1) special relationship with OpenAI, (2) omnipresence in the desktop computing space, and (3) special relationship with our personal information, via Outlook, Onedrive, and Office, to use that shiny new Copilot button for a better purpose: teaching us to fish.
Finally, on the marketing front, more help with AI assist, this time by Microsoft for banner ads. Its Retail Media Creative Studio makes it simple to quickly generate ads, collect performance data, and then optimize, as ‘MediaPost’ describes. From the article:
The goal is to make creative building as easy as possible. In this platform, marketers can take a product photo and turn it into a lifestyle image for a campaign that consumers can relate with by simply prompting the technology to “put this sofa in a cozy living room.”
