If you’re meta and you have to defend the AI by admitting “it’s not really intelligent and everything it says is bullshit”, that’s not a position of strength.
Particularly when it's likely that despite the AI bullshitting in the absence of data on its creators, it's also incidentally true that the "black queer momma" persona is a few lines of behavioural stereotypes composed by white male programmers without any particular experience of black queer mommas.
I dont think that necessarily applies when you could easily make a training set from some actual black American people’s writings on the internet or book, or even an individual that self identified that way and train on all their writings around the internet, and result in those same stereotypes when you ask an AI to create such a profile
You dont need a black American engineer or product manager to say “I approve this message” or “halt! my representation is more important here and this product would never fly” as they are just not person the data set was trained on, even if you asked an AI to just create the byline on the profile for such a character
its weirder, and more racially insensitive, for people to be vicariously offended on behalf of the group and still not understand the group. In this case, the engineer or product manager or other decision maker wouldnt have the same background as the person that would call themselves “momma”, let alone it not mattering at all, if you can regurgitate that from a training set
I mean, sure, some programmers with very little experience of queer black mommas could, hypothetically, be so good at curating information from the internet and carrying out training exercises that they created a persona that convincingly represents a queer black momma. Do we think this is what happened in this instance?
In which, ironically, the bot called it a "glaring omission"
the bot is echoing sentiments of comment sections it was trained on and had no idea of its origins
its acting aware and sensitive but only has information about the tech sector as a whole
my critique is about how the standards being applied are dumb all the way down. the standards are not actually that enlightened even if there was more representation congruent with the race/identity being caricatured as representation. nullifying the whole criticism.
the training set is the only thing thats important with a language model. and its just a symptom of dead internet theory, as even the persona’s byline was probably generated by a different language model.
well, yeah, I acknowledged the bot has no idea of its actual origins in my first post. the point is that at some point some actual product manager thought that creating this persona (probably a generic training set plus a few lines of human-authored stereotypes as prompt) and releasing it to the public as an authentic, representative personality was a good idea. Unlike the product manager, the bot's bullshitting was context aware enough to express the sentiment that this was a bit embarrassing
If the intent is to make a recognizable caricature and apply labels to it (cough stereotype), you don't have someone draw themselves. And it's really looking like stereotyping is their intent.
That is something I’d like to see but I don’t want to wade into the already very complicated discussion around arrow strings in pandas. If a Pandas developer wanted to take this on I think that would make things easier since there’s so much complexity around strings in Pandas.
That said there is a branch that gets most of the way there: https://github.com/pandas-dev/pandas/pull/58578. The remaining challenges are mostly around getting consensus around how to introduce this change.
If NumPy had StringDType in 2019 instead of 2024 I think Pandas might have had an easier time. Sadly the timing didn’t quite work out.
I'm looking at the very first example, and I'm a little confused.
The function `home()` displays a list of messages, but they aren't passed into `home()`. Instead, `messages` is basically a global variable, and some other functions can append messages to it. Then I went looking at some more examples, and I see this pattern repeated. Is this how you're supposed to build webapps with this package? How does it isolate the list of messages for different users?
Which example? I see global vars in a couple of the demos. The game state for the Game of Life makes sense, since it is intended to be shared. The `messages` list in the Chatbot demo is definitely NOT how you'd build a multi-user application, that's mainly showing the styling aspect.
In general, you'd have an actual database and make it so users can only see their own data! See https://github.com/AnswerDotAI/fasthtml/blob/main/examples/a... which adds a filter to queries and DDL statements to ensure that the user can only see/edit their own todos.
those nested if - for - for - if loops are horrendously difficult to understand.
take the fn starting at line 387. they comment why they do certain imports, but this function is comparatively underdocumented. it's not easy to wrap my head around the control flow. some bits are nested about 6 levels too deep for comfort, there are too many positions from which it can return or raise, and the function is about 3x too long
I guess it's the difference between an ensemble and a mixture of experts, i.e. aggregating outputs from (a) model(s) trained on the same data vs different data (GPT-4). Though GPT-4 presumably does not aggregate, but it routes.
I understand the intension and reference you're making. I bet the implementation of GPT-4 is probably something along those lines. However, spreading speculation in definitive language like that when the truth is unknown is dishonest, wouldn't you agree?
Sure, I could it put it less definitively, but realistically, what else can it be? The transformer won't change much and all of the models, at the core use it. It's a closely guarded secret because it's easy to replicate.
That was me on Reddit. I emailed him asking about some of his work on invariant means in the 1970s. He said “I had no taste then”, and told me what a waste of time it was.
At that point, I decided to go into data science instead of trying to get a post doc…
