Neal Grantham 🆖

@alessio_joseph musicforprogramming.net

A model that honestly says it doesn’t know what will happen is not necessarily a bad model. Even with all the data, many things are hard to predict. We have to stop believing data is enough.

all of the questions you need to ask are downstream of "looking at the data once per week"

@just_cameron Hmm, yeah, haven’t tried with a variety of models. Do you notice any change in their response or do they simply seem to ignore it?

@nsgrantham mh, sort of okay. Doesn't seem to work well for cheaper models.

What's the best prompting tool to get a language model to stop saying "Sure, here you go: <ACTUAL RESPONSE I WANTED>"

remember data science

@Jake_Lawlor1 No good way that I’m aware of, ggbeeswarm is the closest we’ve got in the ggverse

Hey #dataviz tweeters, I've noticed a lot of really pleasing circle-pack-along-an-axis plots lately, especially in interactive news or "pop" data vis. Do we have a good way to replicate these in #RStats or #Shiny? ggbeeswarm gets close, but not quite #datavisualization #R

One thing nobody told me about getting older is how much time you spend thinking about databases.

personal files

When you find a 464-view YouTube video that explains the exact obscure sump pump configuration you're working on and the stock music soundtrack goes unbelievably hard:

An IBM slide from 1979.

@ftrain @aboard Might need to wait for next year but adventofcode.com could be a good fit

@Davallree you'll love it

about to watch Koyaanisqatsi for the first time

I asked chatGPT to create 8-bit/pixelated versions of ancient landmarks and cities in DALL·E 3. The results are insane. 1. The Hanging Gardens of Babylon

@mikpanko if you can't express what you want in words, no amount of clicking will get you the right answer if you can express it in words, it's not a big leap to express it in code

installing python

I'm so obsessed w/ this 1981 Apple manuals aesthetic, I want to put it on a t-shirt.

New on the blog: a framework for going from problem to decision with data, including my 'auditioning' method for figuring out the right data, which has saved me many hours! caitlinhudon.com/posts/framewor…

@Mayacelium all the time, I can't seem to get it to stop

…Is anyone else getting tagged in NFT spam or is that just a me problem? I’ve been ghosting Twitter cuz I hate it

@renegadesilicon keep up the good fight, the makers of GPTZeroX (and tools like it) have a responsibility to educate the public on the risks and limitations of their tool, and we as members of the ML community must continue to raise the alarm until they do so

@nsgrantham @jeremyphoward This is exactly right! There is not nearly enough discussion about the false positive pitfalls here. Any question I’ve asked to involved researchers or companies has had zero response. Hopefully your post will get some traction.

extremely irresponsible to release a tool like this without providing any information on its false positive rate how many students will be falsely accused of plagiarism because their educator used GPTZeroX and believed it would "Detect AI Plagiarism. Accurately" as it claims?