AI Top-of-Mind for Dec 5
Interesting article in ‘The Information’ on the creation and use of custom GPTs, this time by Morgan Stanley. It speaks to the need for properly setting expectations, and net-net, adoption is an ongoing process vs just flicking the switch. What is clear is that there will be a learning curve on best practices as enterprises adopt the technology to replace older expert systems and bots. And yet another article on the past year in AI, this time from NY Times discussing the impact on Google, Meta, and others.
From marketing, as reported in ‘MediaPost,’ a disconnect in the percentage of marketeers who use AI vs those willing to admit it.
· But they don’t want to admit it. Of those polled, 52% express at least some concern over revealing their AI use to customers. Drilling down, 13% are greatly worried about possible negative perceptions, and 39% are somewhat concerned.
· What are they using it for? For 32%, personalized marketing is the biggest opportunity, while 25% cite automation of routine tasks. Content development is used by 22%, and data analysis by 19%.
Turning to hardware, Gavin Li in ‘Medium’ describes creative ways to optimize the GPU memory required to train LLMs. Pretty technical, but insightful.
Also on hardware, ‘EE Times’ on how AI is driving new memory architectures in terms of bandwidth and latency.
And, for those with a DIY mindset, ‘llamafile’ for creating LLMs on your laptop. From the description:
Our goal is to make open source large language models much more accessible to both developers and end users. We’re doing that by combining llama.cpp with Cosmopolitan Libc into one framework that collapses all the complexity of LLMs down to a single-file executable (called a “llamafile”) that runs locally on most computers, with no installation.
