AI Top-of-Mind for 5.6.24 — A Mysterious GPT
Top-of-mind for today is the ‘tease’ last week of a next OpenAI ChatGPT. Going by ‘gpt2-chatbot,’ it was briefly available via the LMSYS Chatbot Arena but has since been removed. Jim Clyde Monge posting in ‘Generative AI,’ and Gianluca Centualani put the new model through its paces and concluded it was beyond GPT-4 in its responses. We’ll need to wait and see. From the post:
At LMSYS:
Still keeping to models, more on the foundations of Retrieval Augmented Generation (RAG). Jules Damji covers the following in his tutorial:
Using published research papers, practical blogs, and code examples, this article surveys various aspects of RAG paradigms, including their methodologies, functionality, architecture, types, benefits, limitations, and future directions. We’ll also touch on RAG evaluations, a key aspect of research emerging in Generative AI applications.
Now turning to hardware, you are probably experiencing ‘AI PC’ hype, but what exactly is that? Alexandru Voica writing in ‘The Information’ explains capabilities to be delivered by NVIDIA, Qualcomm, and Apple, as well as describing three potential use cases — hyper-intelligent assistants, real-time multimedia intelligence, and photorealistic synthetic media.
Then, AMD has announced its second-generation Versal AI Edge Series. ‘EE Times’ reports on features, architecture, and availability. From the article:
AMD customer Subaru has already said it will use the second generation Versal AI Edge Gen 2 in its future EyeSight systems, which handle pre-collision braking, lane departure warnings, adaptive cruise control and lane keep assist. Subaru needs to run both application processors and real-time processors in lock-step for functional safety.
And another follow-up, this time to last week’s news on additional agreements between AI firms and media companies. ‘The Information’ looks into licensing challenges, with one datapoint:
One big concern is that tech firms’ offers on AI licensing put too little value on news content. This year, CNN executives have held discussions with OpenAI on potential deals to license articles from CNN’s website. The terms offered translated to paying “a fraction of a penny” per word, according to an executive involved in those discussions. When executives did the math on how many articles CNN would provide and how much it would get paid in total, it would have been a couple of million dollars annually. The CNN executives believed that was way below what their content was worth, said the executive.
If you are thinking of launching a corporate AI project, important reading. Elaine Lu publishing in ‘Towards Data Science’ describes why 85% of projects fail, detailing six reasons why:
· AI systems don’t solve the right problem
· AI innovation gap
· AI systems can’t achieve good enough performance and are not useful
· People miss low-hanging fruit
· AI systems don’t generate enough value
· Ethics, Bias, Societal Harm
I’ll leave you with a short video by Boston Dynamics highlighting the company’s choreography software.
