bradchattergoon.bsky.social
Data Science SM student @harvard. Previously: Yale MBA Caltech BS, Econ research at HBS, Yale SOM. I try to think about things from the beginning.
124 posts
114 followers
180 following
Active Commenter
comment in response to
post
This is a serious productivity improvement for making tailored graphics. Great application of the tech for communicating research/information
comment in response to
post
True true
comment in response to
post
I like the face to face vs windshield framing. It’s easy to extrapolate to an us vs them frame and I think that is part of why auto first cities aren’t great
comment in response to
post
of output over time which is good, but to be exceptional requires more effort investment.
I suppose it boils down to recommendations for being within a std or 2 of the mean or being an outlier. 3/3
comment in response to
post
those who are primarily content creators and thus are selling the idea of productivity but not necessarily doing that much deep work. That is to say, the hardcore hustle narrative isn’t representative. Another layer is that consistent progress over time will produce a reasonable amount… 2/
comment in response to
post
I think there are potentially a couple layers to this. One layer is that content creation selects for a type of person; perhaps the much higher productivity type possibly, if they are the ones that have energy to both do their usual deep work + content creation. Another type could be… 1/
comment in response to
post
Good point! Tough to go back!
comment in response to
post
Is this a statistics joke?
comment in response to
post
My guess is more mathematical strategies will remain intact. For instance if a team comes up with a great pricing algo and trades mispriced assets this would be immune. HFT and market makers are of course also immune. 2/2
comment in response to
post
I think the best way to incorporate these tools would be to build them so that they grade in contexts where they are confident but pass the submission to a human in contexts where they are not confident. This is technically feasible and this is what I’d want to see in such a tool.
comment in response to
post
In my own experience, having ChatGPT as a personal language tutor has been an amazing complement to taking the intro Korean course at school this semester. Being able to ask and receive immediate coaching was invaluable.
comment in response to
post
When did you last use it?
comment in response to
post
Ah I see. I’m not sure it’s trained to be good at languages at than English so I’m not too surprised if this is the case. A model trained specifically for that use case might work much better. I’ve found it very useful for beginner Korean.
comment in response to
post
Can you share a screenshot of your prompt? It’d be helpful to know what sorts of things it is failing at.
comment in response to
post
Can you share what models you used? I’ve have found the latest Open AI model, o1, to be quite good.
comment in response to
post
Fully support these ideas if they can be enacted.
comment in response to
post
Sure, it's not necessary to measure quality or output in terms of patents filed. Does the 44% increase in materials discovered or 17% increase in downstream product innovation also not qualify as an improvement in quality or output? It qualifies for me.
comment in response to
post
This sounds like it will make college tuition more expensive than it is right now, perhaps accelerating the rate of cost increase. In this case college will be less accessible to people and therefore less people will enroll. Does that affect your suggestion?
comment in response to
post
Since it is meant to be based off of the specific course's content which means there is quality control on the content. 2/2
comment in response to
post
Agreed. There is no check on whether wrong or right things can be published and labelled as a textbook even before this tech so I think that's just a fundamental problem of a free press, which comes along with the benefits. With such a concern the original tool, AllDayTA, seems like a good fit. 1/
comment in response to
post
Can you share an example?
comment in response to
post
You might be interested to check out this paper.
aidantr.github.io/files/AI_inn...
comment in response to
post
Are there changes you can suggest that would reverse it? I haven't seen anyone propose changes that could reverse it yet.
comment in response to
post
"courses in measure theory and convergence, then come back to this data analysis". These sorts of practitioners want simple rules to follow so they can submit their paper to a journal. I mention this to say that the problem you raise isn't about the tech, but about how we teach statistics. 2/2
comment in response to
post
As I understand it, the n > 30 thing is a means so simplify statistics for practitioners. The true "math" reality of convergence is of course more complicated but when a biologist wants to evaluate some data, it's probably not a very useful response to say "go take a few semester long..." 1/
comment in response to
post
They didn't outright try to stop use of the technology. I think this is a good strategy because it allows for use of the tech while minimizing the possible harms. This is just my opinion of course. Anyone is free to response how they would like to. 2/2
comment in response to
post
I think fighting against a tsunami isn't a good use of effort because tsunamis won't stop advancing. I think trying to guide that tsunami so it does minimum damage is a better strategy. For example the writers/actors strike a while back wanted safeguards against use of actors' likeness. 1/
comment in response to
post
Thanks for clarifying! Looks like we may indeed be getting close to these models not having the problem you described if not there already.
comment in response to
post
I expect that's something that will improve with time, or with a more tailored model to the specific use case. The latter is what I would expect would be used in educational settings so it ideally wouldn't be adopted until it has that level of performance consistency.
comment in response to
post
A great example of what I think the benefit of these types of tools can be.
comment in response to
post
I tried your prompt in the latest ChatGPT model, o1, and it gets it correct. Were you using an older model?
comment in response to
post
Ok thanks. Will incorporate in the future.
comment in response to
post
What's the best way to signal non-seriousness via emoji? I thought that was the way.
comment in response to
post
Hi I'm a tangerine. I'm not very good at poems. As a tangerine I can't even speak engl... *Narrator: tangerines can't talk*
comment in response to
post
I was going for friendly and disarming ☹️
comment in response to
post
You are forgiven 🙂😛
comment in response to
post
My expectation is that AI technology won't replace humans in research. Research is inherently about new things, AI is trained on existing things. I think AI can help improve research efficiency, but it will likely be complementary to humans, not a substitute. This is just my opinion of course.
comment in response to
post
That's a tough reality to be in but if it was already happening before the introduction of these technologies I don't think lobbying against these technologies will stop that trend from continuing.
comment in response to
post
Starting to realize this too. Twitter/X algorithm is absolutely unusable, but some of the same types of toxic attitudes that plagued Twitter are here as well. A polarized extreme on a platform isn't a good one, even if the polarization is in the direction one aligns with.
comment in response to
post
They could do research for professors instead of grading. You can reference the discussion for this suggestion.
comment in response to
post
It would allow me to instead focus on uncommon mistakes that need more attention. This is based on my experience in STEM courses. It may be different in others, in which case those departments are free to make decisions which best align with their needs. 2/2
comment in response to
post
I was thinking more about grading. From my personal experience most grading exercises pertain to correcting common mistakes students make with respect to the material. If I could have a technology that I train to replicate my responses to these common mistakes that would save me a lot of time. 1/
comment in response to
post
I think the purpose of HE is to help students learn first and foremost, so if this technology has the ability to improve that outcome then I think attention should be paid to that aspect in this discussion in a meaningful way as well. Do you agree or disagree? 3/3
comment in response to
post
You've been focused on the impact on HE employment but you haven't commented at all on the potential upside of such technologies. Specifically, I think the value of a customized tutor on demand 24/7 with immediate responses is likely to be a huge value add to the learning experience of students. 2/
comment in response to
post
To be clear, I didn't roll out any products. I am commenting on what I think will eventually happen.
Also, I strongly disagree with your premise. We can't know everything; this does not mean no one should ever take any action. 1/
comment in response to
post
I don't have an answer for that question at this time. The answer would depend in part on what you believe increasing access to the halls of academia means, and what you believe the benefit of such access is. Note that I'm restricting comments to PhD programs, not other academic programs.