AI Top-of-Mind for 4.29.24 — Where is Plato?
Top-of-mind is one of the more novel uses of AI to decipher parts of the ‘Herculaneum Papyri.’ Back in February we looked at the ‘Vesuvius Challenge,’ but the latest revelations shed light on Plato’s Academy and life and where he may be buried.
And after a day of less substantial postings, we’ll move back to the business-at-hand. Some interesting news on the model and training front, first from ‘MIT IDE’ where a study on business performance gains shows promising results with a major caveat. If the entrepreneur is ‘high-performing,’ AI will create definite advantages, but the opposite holds true for ‘low-performing’ individuals. The key is the prompting, and from the article:
· Lower-performing managers tended to ask about business tasks that were more challenging and less specific.
· Perhaps the low-performing entrepreneurs received good advice from the AI mentor, but were unable to implement it. After all, there are costs to implementing advice, both in time and capital.
A link to the full paper by David Holtz et al.: The Uneven Impact of Generative AI on Entrepreneurial Performance.
And next from ‘Google DeepMind,’ where ‘VentureBeat’ reports on a new study focusing on ‘in-context-learning (ICL)’ and its ability to improve a model’s performance. The trick is the use of long context windows that support the entry of up to thousands of training examples in a prompt. Link to the original paper here.
Common knowledge by now, but both Microsoft and Alphabet (Google) have posted great quarterly results, each focusing on AI as a driver, and both plan to increase their investments in their cloud datacenters to support both LLM development and queries. From the Microsoft PR:
· Revenue was $61.9 billion and increased 17%
· Operating income was $27.6 billion and increased 23%
· Net income was $21.9 billion and increased 20%
· Diluted earnings per share was $2.94 and increased 20%
· Revenue in Intelligent Cloud was $26.7 billion and increased 21%
“Microsoft Copilot and Copilot stack are orchestrating a new era of AI transformation, driving better business outcomes across every role and industry,” said Satya Nadella, chairman and chief executive officer of Microsoft.
And from Alphabet (Google):
Sundar Pichai, CEO, said: “Our results in the first quarter reflect strong performance from Search, YouTube and Cloud. We are well under way with our Gemini era and there’s great momentum across the company. Our leadership in AI research and infrastructure, and our global product footprint, position us well for the next wave of AI innovation.”
But with investment comes addition energy consumption, and a few times I’ve posted about the coming crunch. To help address this, ‘Exowatt,’ one of several startups in this space, is looking to address the problem. But note that AI will also drive efficiency gains, something we also need to consider. ‘Data Centre Dynamics’ offers more details.
We’re all not the size of Microsoft, and Jim Clyde Monge writing in ‘Generative AI’ offers another tutorial on running Llama 3 locally with the help of Ollama and CodeGPT. In the same vein, Ric Raftis details the installation and use of GPT4ALL for secure personal document search. As Ric concludes:
The implications of GPT4ALL extend far beyond simple document search. As it does not require internet connectivity to function after installation and setup, it presents a robust tool for environments with strict privacy requirements or insufficient internet access. As mentioned, the ability to analyse local documents using AI opens new possibilities for academic research and personal information management, enhancing how individuals and organisations manage and interact with their own data.
Turning to security, a good article in ‘EE Times’ on how enterprises must prepare for ‘post-quantum cryptography’ and its inherent challenges. It looks at places where AI will help, and where it won’t. The posting also dives into IBM’s best practices:
And finally to medicine, where AI can help predict heart disease. Dr. Ashish Bamania in ‘Level Up Coding’ describes Imperial College London’s 3D model and what it implies for the future. Not entirely new, but good to file away to add to AI in healthcare. Link to the article in ‘Nature.’
