Launching The State of AI in College Admissions - a survey of 166 American Universities on their stated policies for how much AI is allowed in college admissions, what should be disclosed and what enforcements exist.
Applying to college in 2025 cant get more stressful. Of the 166 top colleges:
55 have explicit guidelines of AI use.
40 require disclosure of AI use.
106 enforce policies by using tools and systems to detect AI use.
Built for Students, Parents, Counselors and Educators dealing with college admissions this Fall. Use our report to quickly search through policies or generate a disclosure in seconds using our free tool.
I am building GradPilot (www.gradpilot.com) to help students with their graduate school application essays. We believe that AI can help many students review their own college applications, especially students who don't have access to counselors.
One of my findings with fine tuning models is that it relies on the inherent strengths of LLMs to learn from the latent space and having great datasets is super valuable to getting the outputs you desire. Aligned Labs was able to get the best creative writers, educators and people who read college applications for us to build a dataset that we fine tuned with GPT4o , GPT4o-mini, llama3-70B and llama3-8B. I highly recommend them!
At first glance this sounds like genius but its not
YC or any successful VC doesn't fund ideas. There is something else the founders have figured out which is key to success and funding. This could be a distribution trick, a key insight, a user growth metric.
Its not hard to imagine YC or a VC coming across the same idea pitched by different founders. Why they picked the team they picked cannot be obvious from trawling the startup directory, that insight is hidden, its literally a secret that should be guarded
bitcoin doesnt suffer this because proof of work provides an independent method for verification. the energy utilization argument that comes up should be evaluated in the context of the comparison of this independent humanless verification vs proof-of-stake or other shared or trusted mechanisms
Do some people have specific ideas or visions of what this could turn out to be? I've tried to play this out in my head but wasn't able to arrive at any new paradigms but then again I've only scratched the surface of plan 9 grid
P2P grid compute + storage that is actually cheap and can drive down the cost of computation significantly.
Can be achieved with unikernels, homomorphic encryption, ipfs. There's a lot of compute and storage available all over the world, idle, waiting to be used. A system like this can utilize it
Thanks for all the feedback HN, I do agree the service is a risky tool so right now I've stubbed it out, effectively disabling its use while I figure out how to address the comments here
Applying to college in 2025 cant get more stressful. Of the 166 top colleges:
55 have explicit guidelines of AI use.
40 require disclosure of AI use.
106 enforce policies by using tools and systems to detect AI use.
Built for Students, Parents, Counselors and Educators dealing with college admissions this Fall. Use our report to quickly search through policies or generate a disclosure in seconds using our free tool.