@w8153328 @SnazzyCipher @guoah79 @ronaldmannak @jzellis Look dude I and seemingly the other guy can tell you have no experience with any of this. When you accept a government research contract, and the contract explicitly requests a solution that works without an internet connection, tell me how you feel about this then

@w8153328 @SnazzyCipher @guoah79 @ronaldmannak @jzellis Can you explain to me how you would access a gov cloud from an air gapped network or machine? Like one that might actually be used in the field?

@SnazzyCipher @guoah79 @w8153328 @ronaldmannak @jzellis The government is pretty liberal about using external tools in certain circumstances, but they only ever run on air gapped networks, ruling out the cloud

@w8153328 @SnazzyCipher @ronaldmannak @jzellis Yeah sure but my point is that if you replace “LLMs” with “Deep learning models” in both of the two original tweets, they’re right, we do use machine learning models for that, they do convert PDFs to markdown

@guoah79 @w8153328 @SnazzyCipher @ronaldmannak @jzellis Those work too, but those are all cloud solutions and not suitable for a lot of the places that still haven’t digitized these old documents out of confidentiality concerns

@GeorgePunished @GuyInSF2 @ronaldmannak @jzellis Out of curiosity, why not just use

@w8153328 @SnazzyCipher @ronaldmannak @jzellis

@StrongManWeak @vikhyatk You’re wrong. In all creative works, the copyright is implicit. And it was established in NEC Corporation v. Intel Corporation that code does have the same protections as other creative works

@StrongManWeak @vikhyatk I’m going to urge you to just Google a little bit, this is “top result” level stuff

@StrongManWeak @vikhyatk Here’s a great example. When I log into Twitter, I download their web client source to my browser, unlicensed for reuse. They put it out there publicly, but I can’t just use it to make a Twitter mirror on my domain. Hope that helps

@robokoteg @HSVSphere @joseph_h_garvin I know it’s tractable, but that doesn’t mean most people should leap to use it unless their API has been stable for quite a while and they’re truly just optimizing

@MTFS111 @joseph_h_garvin @HSVSphere What took a lot of data was probably the pandas dataframe of 2 billion datapoints. I’m going to assume the memory used for the sliding window of 30 datapoints became negligible at that point. In fact, we ended up just using df.rolling(). Feel free to look into how that works

@MTFS111 @joseph_h_garvin @HSVSphere I’m not gonna lie to you, this is something I used for a timeseries based data science project once, and I truly didn’t think about it that deep. It only moved forward though

@HSVSphere @robokoteg @joseph_h_garvin To be clear, there is a place for performance optimized code. The people reading this are probably not writing that code. 90% of the people reading this are web devs who are better off thinking about how they can iterate on their software the fastest, not how fast it goes

I am here to say that in 99.9% of use cases no one cares how fast your code is, they care if it works in an intuitive way or not