Collaborative Intelligence Week in Review - 04Nov2024
Best business article(s) I read this week…
This analysis from the Boston Consulting Group finds that “only 22% of companies have advanced beyond the [AI] proof-of-concept stage to generate some value, and only 4% are creating substantial value.” They categorize all companies studied as “not doing much”(25%), “still focusing on proof of concepts” (49%), “scaling value” (22%), and “operating value engines” (4%).
How AI agents are reshaping the future of work
This report from the Deloitte AI Institute describes how AI agents and multiagent systems are transforming work processes. Unlike generative AI, which primarily assists with single tasks, AI agents can autonomously manage entire workflows, use the experience to adapt over time, and connect with multiple digital tools to execute complex, multistep processes. Organizations using AI agents see benefits such as improved collaboration, faster decision-making, enhanced personalization, and significant scalability improvements. Key applications include automated customer support, personalized financial advising, dynamic pricing, and talent acquisition.
Best technical article(s) I read this week…
GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models
The paper, referenced in the next section, describes the limitations of large language models (including OpenAI’s new o1 model) in performing genuine mathematical reasoning because current LLMs rely on pattern-matching rather than formal logical processing. This lack of reasoning leads to inconsistent results, especially as complexity increases, and reveals that LLMs struggle with tasks that require logical depth beyond surface-level associations.
Other items I found valuable/interesting…
I Took a ‘Decision Holiday’ and Put AI in Charge of My Life
The author of this New York Times article “used two dozen generative AI tools for daily tasks and nearly 100 decisions” for a week. “It told me what to eat, what to wear, and what to do with my kids. It chose my haircut and what color to paint my office.” She includes text chains, photos, and other details. I won’t spoil the ending, where she assesses what worked and what didn’t.
These Apple researchers just showed that AI bots can’t think and possibly never will
This Los Angeles Times article discusses an Apple research paper (see previous section) highlighting significant limitations in AI models' reasoning abilities. Apple researchers demonstrated that large language models (including OpenAI’s new o1 model) often fail to solve math problems reliably, especially when problems contain irrelevant details. These AI systems frequently treat this irrelevant information as significant, leading to incorrect answers. The paper casts doubt on whether LLMs can ever handle abstract problem-solving reliably and raises concerns about their use in applications where “empirically perfect calculations are necessary.” (This quote is from an email to me from the article’s author.)
Walmart and Amazon are rolling out AI shopping assistants for Holiday 2024
Glimpse the future of online shopping experiences this holiday by using Walmart’s AI-driven shopping assistant (still in beta) or Amazon’s “Rufus,” which launched in Europe this week.
Coolest thing I saw…
Robots powered by insect brains could be used on Mars
Opteran, a company specializing in neuromorphic AI (inspired by insect brains), partnered with Airbus and space agencies to test its AI platform in Mars rover simulations. This technology aims to enable autonomous navigation in challenging Martian terrains, reducing reliance on human control.
A company that caught my eye…
It’s not quite flying cars, but they’re getting closer. This Vermont-based company, which makes small electric vertical take-off and landing (VTOL) planes, closed a $318M Series C round led by Qatar’s sovereign wealth fund. A spokesperson said the new money will move its aircraft closer to certification and commercialization.