What Level of AI Model Are You Qualified to Work With?

How should we train students to work with PhD level AI they can access for $20/month? Do you need a PhD to work with a PhD level researcher? How can schools manage students learning from PhD-level AI?

Early last week, The Information circulated a story about what OpenAI (ChatGPT) is planning to charge for access to various AI abilities.

I saw one person react to the announcement by saying that a person really needs a PhD to take advantage of Level 5/PhD Researcher.

It got me thinking about how we might soon judge a job applicant’s abilities by how well they can use/take advantage of a certain level or type of AI bot, combination of bots, or combination of AI bots and humans at similar levels.

In education and in professional development for educators, most of the focus has been on helping people work with a basic chatbot (level 1), something that can already do most school work, especially K-12 and some early undergraduate level, pretty well. More and more people are figuring out how to use it for this level of work through basic training, online courses, and simple word of mouth.

These bots have also become easier to use, as simple natural language conversation and follow-up (continued conversation) with the bot can get you pretty far. When I advise people about how to prompt, I tell them to simply be clear about what they are asking for, but to imagine they are asking a person who likely knows the answer to something, but making sure they provide enough context to get the answer they need. As “memory grows,” the need to supply all of the context will decline. I encourage them to ask follow-up questions to clarify and go deeper, as if they were interacting with someone.

The PhD-level bots are at the other end of the spectrum, with some claiming that one needs a PhD to interact with this level of bot. At first, I agreed with this; I found that sometimes this level produced reports on debate topics I was working on where I didn’t fully understand some of the economics and/or science the reports discussed.

But I don’t know if that’s the case for several reasons.

First, I often read studies conducted by people with expertise that I lack. While I often don’t understand all the details of how the study was constructed or some of the concepts, I can certainly understand the basic ideas and conclusions.

Second, this isn’t true in most contexts. Take, for example, a nurse. A nurse receives a lot of medical training. The level of training doesn’t reach the level of training that a doctor gets, but it is high enough that the nurse can work with a doctor to treat a patient. To extend this analogy, imagine a world of AI/robot doctors that possess an enormous amount of medical knowledge (this is sort of here already) and many human nurses who are great communicators and have a substantial “bedside manner.” These nurses can work with the robot doctors and translate a treatment plan and sometimes difficult news to a human patient.

So, it’s not hard to imagine a world where significantly more medical care could be delivered in rural areas with a shortage of doctors by nurses who receive enough medical training to interact well with robot doctors.

Some doctors, of course, will still be needed. They’ll be needed (at least for now) to help advance the state of medical knowledge, working with AI “co scientists.” They’ll be needed (at least for now) to help solve complex medical cases, just as two or more doctors would work together to do that now.

Obviously, people with PhDs will still be best able to interact with PhD-level AI models, but it’s not required.  People just need enough knowledge to interact productively with the PhD-level bot.

Third, of course, some people run companies and divisions of companies that oversee PhDs doing work there; overseers do not understand at all.

One does not need a PhD (or the equivalent), something only 2% of Americans have, to interact with another PhD. Nurses interact with doctors. Local business owners know enough to take advice from economists. Teachers can implement advice from education leaders.

We don’t need PhD level knowledge to interact with a PhD level bot for most things. What we do need is some basic foundational knowledge, and we need to understand how to communicate with (teams of) bots and humans.

In some ways, this is similar to the way the world already works, but the difference is that the costs of interacting with someone with PhD knowledge are collapsing. Even at the highest subscription cost ($20,000/month) for enterprise, you get to work with an infinite number of PhDs in all subject areas. So, you are not hiring one subject area PhD for $20,000 per month, but perhaps 10 (if that’s what you need) for $2,000/month each. That’s the equivalent of paying 10 PhDs the equivalent of $24,000 a year each….

Anyhow, to get back to the original point, access to knowledge, including PhD-level knowledge, is becoming incredibly cheap. As the quality (accuracy/thoroughness) of PhD AIs improves and even starts to exceed that of all or nearly all people with PhDs (some, including leading PhDs in their field such as Tyler Cowan) believe it already has, the cost of accessing such knowledge will decline due to ever-growing exponential gains on computational efficiency. It’s not absurd to think that everyone will eventually be able to work with a team of (super) intelligent PhDs for $100/month rather than $20,000 per month.

I need to think this through more, but I think this has a number of implications for education.

First, I think education needs to focus on preparing students to interact well with bots at the top of the intelligence spectrum. Whatever remains of traditional “knowledge work” will almost certainly require this. Students who start their own businesses will benefit from being able to interact with a team of people who have PhD-level knowledge in all or nearly all subject areas for a very low price.

Second, we need to fully understand what level of intelligence students already have access to. The $20/month ChatGPT subscription, which many K-16+ students already have, gives access to 10 PhD interactions per month with this level of knowledge. Perplexity offers a similar level of AI, with 500 interactions for $20/month. This is also now available in Grok-3 and Manus out of China (see below).

Schools need to understand that this level of AI can do nearly any school assignment. As Richard Socher, founder of You.com, recently pointed out, students can simply drag and drop assignments into You.com, and it will read the instructions and engage agents to complete the assignment. No prompting is even needed.

Third, schools need to consider what level of AI they are giving students access to and whether or not they will use it. Systems such as Magic School and School AI provide access to Level I with restrictions common to school systems.

But many students use Level 5 systems with no restrictions and pay only $20/month for individual use (limited Level 5 use is available for $20/month). Why will they use such “dumb” AIs, and how do these Level I systems help train students to work with more advanced AIs?

Third, what do you think about how to prepare students for this world?

PS. A Manus Report.

New AI model out of China —

For those of you haven't heard of Manus from China yet, this is someone's test run --

<1. For my first test, I asked Manus to create a biography on Rowan Cheung and deploy a website based on that biography. It was insanely impressive watching it go through my social channels, browse articles, and deploy the site. And it was 100% accurate, info up to date as of today.

2. Next, I decided to test something a bit more practical: Top rental spots in SF with

- Low crime

- Lots of AI activity

- High density of ambitious young entrepreneurs

I got a full detailed report with 4 options that were spot on.

3. Next I wanted to test how Manus would do very long research tasks. So, I asked it to create an entire course on AI for Content Creation It took nearly 2 hours to complete, but what I got was an impressive 8-chapter course with tools, use cases, and even prompt examples.

Manus is only available via invite code right now. This post was NOT sponsored in any way, but the cofounder @ Manus gave The Rundown 500 invite codes.

We just added it to The Rundown University's perks (which now has $1000+ worth of AI tool savings!)>