Presentation 3: Experimenting with AI Deep Research
first given 27th June 2025 at the Pop Inn Community Cafe, Yate South Glos. UK
Link to first section The evolving AI environment.
A review of the evolving maturity of AI to give a context to the environment for Deep Research.
Link to second section AI tool summary.
A look at the main tools for Deep Research and what they do.
Analysis of some audience provided questions
These were tested with a selection of tools, as explained below. Please note that the original questions have all been expanded and otherwise altered to make them more AI tool friendly. Disclaimer! The responses are as written by the AI tool and have not been checked for accuracy and may contain hallucinations or other inaccuracies.
User question 1: Quantity Surveying (John S.)
'Taking Off' Question. In the UK one job of a Quantity Surveyor is "taking off" which is the process of looking at a planning document for a building including drawings and listing all of the materials and work needed to build it so that a cost estimate can be made. We need a report for our management board explaining to what extent AI tools should be expected to contribute to this within two years from now. This will inform how many trainees we need to take on, given a constant workflow. If we currently take on 5 per year, how many might we need by two years time.
[1.1] Using Grok in 'Think Mode'.
Below is an extract showing the very start of the report.
[1.2] Using Grok in 'DeepSearch'.
Extract: Research suggests AI will automate 30-40% of routine "taking off" tasks by 2027, but human oversight remains essential.
It seems likely that trainee numbers could decrease from 5 to 4 per year, given constant workflow, due to AI efficiency gains.
The evidence leans toward AI enhancing, not replacing, quantity surveyors, with controversy around job impacts still debated.
Here is Grok's full report in pdf form, badly formatted due to cut and paste. Grok's pdf report.
[1.3] Using Perplexity 'DeepSearch'.
Perplexity went through a long analysis process including writing some Python code that produced a chart projecting workforce needs forward in different scenarios. There is a pdf (link below).
Extract: Artificial Intelligence is poised to significantly transform quantity surveying, particularly the "taking off" process, within the next two years. Based on comprehensive analysis of current AI adoption rates, efficiency gains, and industry trends, we recommend reducing your annual trainee intake from 5 to 4 trainees by 2027 to maintain optimal workforce levels given projected productivity improvements.
Perplexity's pdf report
User question 2: Drones and the Last Mile (Kate K.)
'Last Mile' Question. Please provide a research report on how a large UK food delivery company can effectively integrate drones - particularly for deliveries outside urban centres - using ground-based interim hubs and aiming for long-term full automation. The report will cover AI driven coordination, drone fleet management, logistical infrastructure, vendor ecosystem, and best practices.
[2.1] Using ChatGPT o3 in 'Deep Research' mode.
Extract: Drone delivery offers a promising solution to optimize last-mile and middle-mile logistics in non-urban areas, providing faster service to customers and operational efficiencies for the company. By leveraging AI-driven fleet coordination, the company can achieve real-time route optimization, intelligent avoidance of obstacles, balanced utilization of drones, and proactive maintenance – all contributing to a safe and efficient network.
The necessary infrastructure – from ground hubs and charging stations to software integration – must be developed in parallel to support these operations, transforming existing delivery platforms into drone-compatible systems. A review of leading drone vendors (Wing, Zipline, Manna, Wingcopter, Matternet, among others) indicates there are mature solutions capable of handling UK’s weather and geography ...
[2.2] Using Perplexity in 'Deep Search' mode.
Extract of conclusion: The drone delivery market leaders demonstrate that successful AI systems for last-mile navigation and customer interaction require sophisticated integration of autonomous navigation, real-time communication, predictive analytics, and seamless customer experience design. Companies preparing for 2028 market entry should focus on developing scalable, safety-first systems that can operate with minimal human oversight while maintaining high reliability and customer satisfaction standards. The projected market growth to $12.3 billion by 2028 indicates substantial opportunities for well-designed AI systems that can effectively manage the complex requirements of autonomous last-mile delivery operations.
User question 3: 'Uncertain times' (Sandhya S.)
Using Gemini Deep Research
'Uncertain Times' Question. [original question was about predicting world events]. These are uncertain times for World political decision makers. With particular focus on the UK perspective and a five year horizon, please report on the five most important policy challenges and recommend the best responses.
Thinking process extract: Information Gathering Strategy. To gather the most relevant and reliable information, I'm prioritizing reputable sources. This includes government bodies, established think tanks, and academic institutions. My initial information gathering will cover global geopolitical shifts, economic volatility, and technological advancements, all viewed through the lens of their implications for the UK. I'll also be looking into key domestic policy challenges across social, economic, environmental, and governance sectors. I'm specifically looking for reports from prominent UK and international think tanks that focus on UK policy challenges for the upcoming five-year period.
Conclusion extract: The United Kingdom's policy landscape for 2025-2031 is shaped by a set of deeply intertwined challenges: persistent stagnant economic growth and productivity, the escalating demands on public services, the imperative of tackling climate change while ensuring energy security, the enduring housing crisis and its link to homelessness, and the fundamental need to enhance governance effectiveness and achieve regional rebalancing. These are not isolated issues; rather, they form a complex web where progress in one area often creates dependencies or opportunities in another.
Gemini also gives you the ability to share a link:
Shareable link: Link to Gemini's report
User question 4: 'Benefits and risks of ASI (Superintelligence)' (Joy C.)
Using ChatGPT o3 Deep Research
'Risks v Benefits of ASI' Question. What are the likely societal outcomes over the next 10 years if world leaders fail to establish coordinated AI policies — specifically weighing the risks of broad positive outcomes (AI Abundance) against significant negative consequences (AI Doom).
Conclusion extract: [5 mins, 24 sources, 77 searches]. On the whole, abundance appears more likely than doom, in the sense that most experts expect broadly positive outcomes to outweigh the negatives. AI capabilities are advancing rapidly, and even without unified governance, they are poised to deliver significant improvements in productivity, knowledge, and possibly human welfare. It is reasonable to anticipate that by 2035, AI will be deeply embedded in everyday life – largely for the better – from health and education to commerce and entertainment. Many millions of people could be lifted up by AI-enabled innovations (for example, better medical diagnostics in developing countries or personalized tutoring closing educational gaps). These optimistic outcomes have a strong probability given the sheer investment and momentum in AI development worldwide. However ...
The full ChatGPT response is not given here. Instead the content was copied and pasted as an input to a Google NotebookLM https://notebooklm.google.com Audio Overview (free). This creates two hosts discussing the topic.
Google NotebookLM Audio Overview
Google NotebookLM Audio Overview using analysis by OpenAI ChatGPT o3 of AI future risks v benefits. pic.twitter.com/YWDfgoSvaf
— The Silver AI Project - free AI training materials (@SilverAIProject) June 27, 2025