The movies embedded on this put up are greatest considered on steveblank.com
Worldwide Coverage college students might be spending their careers in an AI-enabled world. We needed our college students to be ready for it. That is why we’ve adopted and built-in AI in our Stanford nationwide safety coverage class – Expertise, Innovation and Nice Energy Competitors.
Right here’s what we did, how the scholars used it, and what they (and we) discovered.
Expertise, Innovation and Nice Energy Competitors is a global coverage class at Stanford (taught by me, Eric Volmar and Joe Felter.) The course supplies future coverage and engineering leaders with an appreciation of the geopolitics of the U.S. strategic competitors with nice energy rivals and the function essential applied sciences are enjoying in figuring out the end result.
This course contains all that you’d count on from a Stanford graduate-level class within the Masters in Worldwide Coverage – complete readings, visitor lectures from present and former senior coverage officers/consultants, and deliverables within the type of written coverage papers. What makes the category distinctive is that that is an experiential coverage class. College students kind small groups and embark on a quarter-long challenge that bought them out of the classroom to:
- choose a precedence nationwide safety problem, after which …
- validate the issue and suggest an in depth answer examined in opposition to precise stakeholders within the know-how and nationwide safety ecosystem
The category combines a number of instructing instruments.
- Actual world – College students labored in groups on actual issues from authorities sponsors
- Experiential – They get out of the constructing to interview 50+ stakeholders
- Views – They get coverage context and insights from lectures by consultants
- And this 12 months… Utilizing AI to Speed up Studying
Rationale for AI
Utilizing this quarter to introduce AI we had three issues going for us: 1) By fall 2024 AI instruments had been good and getting exponentially higher, 2) Stanford had arrange an AI Playground enabling college students to make use of quite a lot of AI Instruments (ChatGPT, Claude, Perplexity, NotebookLM, Otter.ai, Mermaid, Lovely.ai, and so forth.) and three) many college students had been utilizing AI in courses but it surely was often ambiguous about what they had been allowed to do.
Coverage college students should learn reams of paperwork weekly. Our hypotheses was that our pupil groups may use AI to ingest and summarize content material, establish key themes and ideas throughout the content material, present an in-depth evaluation of essential content material sections, after which synthesize and construction their key insights and apply their key insights to resolve their particular coverage drawback. They did all that, and far, a lot, extra.
Whereas Joe Felter and I had arm-waved “we have to add AI to the category” Eric Volmar was the actual AI hero on the instructing crew. As an AI energy consumer Eric was most frequently forward of our college students on AI expertise. He threw down a problem to the scholars to repeatedly use AI creatively and informed them that they might be graded on it. He pushed them laborious on AI use in workplace hours all through the quarter. The outcomes beneath converse for themselves.
For those who’re not conversant in these AI instruments in observe it’s price watching these one minute movies.
Workforce OSC
Workforce OSC was attempting to know what’s the acceptable stage of economic threat for the U.S. Division of Protection to supply loans or mortgage ensures in know-how industries?
The crew began utilizing AI to do what we had anticipated, summarizing the stack of weekly coverage documentsusing Claude 3.5. And like all groups, the surprising use of AI was to create new leads for his or her stakeholder interviews. They discovered that they might ask AI to present them a listing of leaders that had been concerned in comparable applications, or that had been concerned of their program’s preliminary phases of improvement.
See how Workforce OSC summarized coverage papers right here:
For those who can’t see the video click on right here
Claude was additionally capable of create a listing of leaders with the Division of Power Title17 credit score applications, Exim DFC, and different federal credit score applications that the crew ought to interview. As well as, it created a listing of leaders inside Congressional Finances Workplace and the Workplace of Administration and Finances that will have the ability to present insights. See the demo right here:
For those who can’t see the video click on right here
The crew additionally used AI to transcribe podcasts. They seen that key leaders of the organizations their drawback got here from had produced podcasts and YouTube movies. They used Otter.ai to transcribe these. That offered extra context for after they did interview them and allowed the crew to ask insightful new questions.
For those who can’t see the video click on right here
Observe the ability of fusing AI with interviews. The interviews floor the information within the groups lived expertise.
The crew got here up with a use case the instructing crew hadn’t considered – utilizing AI to critique the crew’s personal hypotheses. The AI not solely gave them criticism however supported it with hyperlinks from printed students. See the demo right here:
For those who can’t see the video click on right here
One other use the instructing crew hadn’t thought was utilizing Mermaid AI to create graphics for his or her weekly displays. See the demo right here:
For those who can’t see the video click on right here
The surprises from this crew saved coming. Their final was that the crew used Lovely.ai with a view to generate PowerPoint displays. See the demo right here:
For those who can’t see the video click on right here
For all groups, utilizing AI instruments was a studying/discovery course of all its personal. By and enormous, college students had been largely unfamiliar with most instruments on day 1.
Workforce OSC urged that college students ought to begin utilizing AI instruments early within the quarter and experiment with instruments like ChatGPT, Otter.ai. Instruments that which have steep studying curves, like Mermaid needs to be began on the very begin of the challenge to coach their fashions.
Workforce OSC AI instruments abstract: AI instruments usually are not good, so ensure that to cross verify summaries, insights and transcriptions for accuracy and relevancy. Be actually essential of their outputs. The most important takeaway is that AI works greatest when ready with human efforts.
Workforce FAAST
The FAAST crew was attempting to know how can the U.S. enhance and scale the DoE FASST program within the pressing context of nice energy competitors?
Workforce FAAST began utilizing AI to do what we had anticipated, summarizing the stack of weekly coverage paperwork they had been assigned to learn and synthesizing interviews with stakeholders.
One of many options of ChatGPT this crew appreciated, and vital for a nationwide safety class, was the non permanent chat characteristic – knowledge they entered wouldn’t be used to coach the open AI fashions. See the demo beneath.
For those who can’t see the video click on right here
The crew used AI do a number of new issues we didn’t count on – to generate emails to stakeholders and to create interview questions. In the course of the quarter the crew used ChatGPT, Claude, Perplexity, and NotebookLM. By the tip of the 10-week class they had been utilizing AI to do a number of extra issues we hadn’t anticipated. Their use of AI expanded to incorporate simulating interviews. They gave ChatGPT particular directions on who they needed it to behave like, and it offered personalised and customized solutions. See the instance right here.
For those who can’t see the video click on right here
Studying-by-doing was a key a part of this experiential course. The massive thought is that college students be taught each the tactic and the subject material collectively. By studying it collectively, you be taught each higher.
Lastly, they used AI to map stakeholders, get recommendation on their subsequent coverage transfer, and requested ChatGPT to overview their weekly slides (by screenshotting the slides and placing them into ChatGPT and asking for suggestions and recommendation.)
The FAAST crew AI device abstract: ChatGPT was particularly good when utilizing photographs or screenshots, so in these multi-level duties, and while you needed to make use of type of extra customized directions, as we used for the stakeholder interviews. Claude was higher at extra conversational and human in writing, so used it when sending emails. Perplexity was higher for researchers as a result of it supplies citations, so that you’re capable of entry the online and truly get directed to the supply that it’s citing. NotebookLM was one thing we tried out, but it surely was not as profitable. It was a cool device that allowed us to summarize particular coverage paperwork right into a podcast, however the summaries had been usually fairly obscure.
Workforce NSC Power
Workforce NSC Power was engaged on a Nationwide Safety Council drawback, “How can the USA generate adequate power to assist compute/AI within the subsequent 5 years?”
At first of the category, the crew started by utilizing ChatGPT to summarize their coverage papers and generate tailor-made interview questions, whereas Claude was used to synthesize analysis for background understanding. As ChatGPT sometimes hallucinated data, by the tip of the category they had been cross validating the summaries through Perplexity professional.
The crew additionally used ChatGPT and Mermaid to arrange their ideas and decide who they needed to speak to. ChatGPT was used to generate code to place into the Mermaid flowchart organizer. Mermaid has its personal language, so ChatGPT was useful, so we didn’t should be taught all of the syntax for this language.
See the video of how Workforce NSC Power used ChaptGPT and Mermaid right here:
For those who can’t see the video click on right here
Workforce Alpha Technique
The Alpha Technique crew was attempting to find whether or not the U.S. may use AI to create a whole-of-government decision-making manufacturing unit.
At first of sophistication, Workforce Alpha Technique used ChatGPT.40 for coverage doc evaluation and abstract, as properly for stakeholder mapping. Nonetheless, they found going one after the other by way of the numerous numbers of articles was time consuming. So the crew pivoted to utilizing Pocket book LM, for doc search and cross evaluation. See the video of how Workforce Alpha Technique used Pocket book LM right here:
For those who can’t see the video click on right here
The opposite instruments the crew used had been customized Gpts to construct stakeholder maps and diagrams and manage interview notes. There’s going to be all kinds of specialised Gpts. One which was actually useful, they mentioned, was a scholar GPT.
See the video of how Workforce Alpha Technique used customized GPTs:
For those who can’t see the video click on right here
Like different groups, Alpha Technique used ChatGPT to summarize their interview notes and to create circulation charts to stick into their weekly displays.
Workforce Congress
The Congress crew was exploring the query, “if the Division of Protection got financial devices of energy, which instruments can be handiest within the present techno-economic competitors with the Folks’s Republic of China?”
As different groups discovered, Workforce Congress first used ChatGPT to extract key themes from tons of of pages of readings every week and from press releases, articles, and laws. In addition they used for mapping and diagramming to establish potential relationships between stakeholders, or to creatively recommend alternate visualizations.
When Workforce Congress weren’t capable of attain their sponsor within the preliminary two weeks of the category, very like Workforce OSC, they used AI instruments to faux to be their sponsor, a member of the protection modernization caucus. As soon as they realized its utility, they continued to do mock interviews utilizing AI function play.
The crew additionally used personalized fashions of ChatGPT however of their case discovered that this was restricted within the variety of paperwork they might add, as a result of that they had plenty of content material. So that they used retrieval augmented era, which takes in a consumer’s question, and matches it with related sources of their information base, and fed that again out because the output. See the video of how Workforce Congress used retrieval augmented era right here:
For those who can’t see the video click on right here
Workforce NavalX
The NavalX crew was studying how the U.S. Navy may develop its capabilities in Intelligence, Surveillance, and Reconnaissance (ISR) operations on common maritime site visitors.
Like all groups they used ChatGPT to summarize and extract from lengthy paperwork, organizing their interview notes, and defining technical phrases related to their challenge. On this video, observe their use of prompting to information ChatGPT to format their notes.
See the video of how Workforce NavalX used tailor-made prompts for formatting interview notes right here:
For those who can’t see the video click on right here
In addition they requested ChatGPT to function play a critic of our argument and answer in order that we may discover the weaknesses. In addition they started importing many interviews directly, and requested Claude to seek out themes or concepts in widespread that they may have missed on their very own.
Right here’s how the NavalX crew used Perplexity for analysis.
For those who can’t see the video click on right here
Like different groups, the NavalX crew found you possibly can customise ChatGPT by telling it the way you need it to behave.
For those who can’t see the video click on right here
One other shocking perception from the crew is that you need to use ChatGPT to let you know find out how to write higher prompts for itself.
For those who can’t see the video click on right here
In abstract, Workforce NavalX used Claude to translate texts from Mandarin, and located that ChatGPT was the very best for writing duties, Perplexity the very best for analysis duties, Claude the very best for studying duties, and pocket book LM was the very best for summarization.
Classes Discovered
- Integrating AI into this class took a devoted teacher with a mission to create a brand new solution to educate utilizing AI instruments
- The outcome was AI vastly enhanced and accelerated studying of all groups
- It acted as a useful collaborator
- Fusing AI with stakeholders interviews was particularly highly effective
- At first of the category college students had been conversant in a number of of those AI instruments
- By the tip of the category they had been fluent in lots of extra of them
- Most groups invented artistic use circumstances
- All Stanford courses we now educate – Hacking for Protection, Lean Launchpad, Entrepreneurship Inside Authorities – have AI built-in as a part of the course
- Subsequent 12 months’s AI instruments might be substantively higher
Filed below: Buyer Improvement, Gordian Knot Middle for Nationwide Safety Innovation, Nationwide Safety, Expertise Innovation and Nice Energy Competitors |