Admission Hacks is a way to generate your own college rankings based on any criteria important to you. Standard rankings like US News can have colleges move by 50 or more places based on criteria you don't care about and are really a better measure of "what does an uninformed person think about the reputation of a college?"

It's a set of unique data with a custom AI engine on top of it to allow you to navigate the rankings in plain language.

The ratings, which were a great innovation when they launched in the 1980's, ultimately had a terrible effect on incentives for the colleges and universities and the experience for the student. Suddenly students filling out 8 to 10 (or even more!) applications, every college is striving for the same subset of students that boost their rankings, and colleges are stuck trying to modify their actual education with an eye towards how they interact with the rankings.

We saw the amazing amount of data that was starting to become public about colleges - ranging from the data the US government released, to the colleges publicly releasing the data used for rankings to academic studies like Raj Chetty's opportunity atlas. In the end,we pulled together over 15 different data sources that gave real insights into the differences between colleges.

AI hit an inflection point in early 2023 with the broader availability of GPT4. We realized we could put a lot of my knowledge about how to read the data into GPT and allow people to interact directly with the data. The rest, as they say, is history.

In some ways, we don't know. Probably not everybody; it really takes some sophistication to understand why this data is better than other sites (and not just because it's broad), or how to use the data to cut through the marketing noise and make good decisions.

I suspect it will be STEM-savvy parents and high schoolers, along with a range of college counselors who aren't threatened by making available more broadly the ability to navigate this data.

There are over 15 data sources pulled together into a single database, with over 4 thousand different fields of data per college; it's far too much data for a human to navigate. As you ask various questions, you can hover over the header to any field and see the source for that particular column.

These are good categories to think about when selecting colleges. These days, most kids apply to 8-10 colleges with a good foundation of 2 Likely schools and 2-3 Target schools that they would love, and then the remainder a mix of Reach and Unlikely. Note that it's not an insult to say a school is Unlikely for you; most Unlikely schools are Unlikely for nearly all students just given the sheer number of applicants.

The specific categories are defined as follows on Admission Hacks:

Here's a quick video to walk you through using it , or read on...

Basically three steps:

1. Tell it about you so that it can estimate your admissions chances well enough to help you determine your categories. This includes all areas that can impact admissions, including your GPA, test scores, hometown, gender, and ethnicity (yes, still!) as well as data that can affect specific colleges like where your parents attended school (legacy) or if you're the first in family to attend.
2. Tell it what you're looking for in a college. Go ahead and be broad, there are lots of ways to decide - what do you want to major in? Do you care about how many students there are? Where it's located? Safety? Sexual orientation of other students? Gender balance? Any considerations on religion? Does cost matter? (In which case, also give it your family income so it can estimate how much aid people like you tend to get at each college), and so forth. Often things just come up while you're visiting a college that leads you to add criteria (or potentially subtract them!)
3. Refine your list depending on what you're trying to do with it - e.g., &qout;Show me target and likely schools only&qout; or &qout;I'm visiting Philadelphia next month, can you show me colleges I should visit in that area&qout;. Or even as you research or visit various colleges, you can say things like &qout;I like Wesleyan and Amherst, can you show me other colleges like that&qout;.

You can always get more details about the college (including what they say they value in admissions and how much) by clicking on the eye icon next to any item in the list. At the bottom of that window, you can find other colleges that are similar that both (a) meet your criteria and (b) are sorted by your likelihood of admission so you can pick similar colleges that are either higher or lower likelihood.

On the UI, everything can be done via the chat window. If you want, you can click on any criteria you've identified - they will appear in the window above your college list. That will let you change the criteria, change its importance in your personal ranking, or remove it altogether.

Finally, on the "Advanced Usage"...you can also use the "Filter" text to filter by entering any category (capitalized) - e.g., "Likely" - or any institution name - e.g., "ucla" or "penn state" - to see where it ranks in your list and why.

ChatGPT and other large language models are basically programmed to return plausible-sounding answers. Often that's because its answer is right, but often it's wrong but looks like it could be right (what the AI experts call "hallucinations").

In this case, we did not train the AI on this data and let it answer the questions. Instead, we are using the AI for things it does very well: translating your questions into "computer-ese" so we can pick the right fields to show you from the data, and summarizing the data to specifically answer your questions. That's how we avoid hallucinations that typically plague an experience like this.

I don't want to overstate this - you should still double-check an institution's website to make sure the data is correct (on deadlines, specific programs, etc.). But that's for the old-fashioned reason that there can be discrepancies between what universities report to the government or ratings services and what they later change on their site.

NOTE: That said, it's still a GPT model at the back-end. So if you ask it questions not covered by our data, you'll get typical GPT answers which may include hallucinations.

Good question. AI is expensive. This can't just run for free on advertising $. I would like to keep it free for students who don't have access to personalized college counseling. I suspect there are three ways over the long run:

1. Charge schools and college counselors (directly, or indirectly by integrating with services that they are already paying for)
2. Add services that are worth paying for that are particularly wanted by families that can pay for them - customized advice and help navigating for international applicants, or more detailed and accurate admissions estimates for selective schools (along with advice for how to tailor what you do to their expectations) for parents and kids who really do want to target them.
3. Ideally, I'd like schools themselves to game customized rankings, not generic rankings like US news. To do that, they'd need data, which we can sell them. Which leads to the next question…

The generic rankings systems like US News lead colleges to do unnatural acts to try to climb the rankings: boosting tuition and spending the money on things that improve your ranking, for example. Or encouraging all students to apply to reduce your admit rate. Or introducing early decision, forcing students to commit to coming to your college even though their chance of getting in is small, because that boosts your yield rate. Or even just lying about your own statistics.

All of those actions are feeding the system that makes it harder on applicants and harder on colleges. If you were to try to improve your rankings on this system, it leads to a very different set of actions: since every search is different, colleges would instead be rewarded for finding niches that many students need that are currently underserved.

We'd like to help this process along, and I suspect ultimately if this succeeds that's where the business model lies: colleges pay to give us demographics of students they want to target, and we can tell them what kinds of things students like the ones they want to target are searching for in a college. What is the average rank they have in those student searches, and most importantly, what characteristics should they improve to show up in more of those searches. The colleges will benefit by finding their niche, rather than all competing based on the same criteria.

Plus we can help colleges target students in ways that help the students and don't cost the college valuable financial aid $. After all, if you're looking for physics students, or students from rural areas, and are willing to give them a boost in their admissions chances, that by itself should help draw their applications *but only if those students know about it*. It's a benefit to that physics student you're trying to draw to your school to build your physics program to know about that, and to you as a school.