Mark Pauley
banner
unsaturated.bsky.social
Mark Pauley
@unsaturated.bsky.social
120 followers 140 following 400 posts
Professional Software Engineer, amateur skateboarder. synthesizers, music, bad jokes and cats
Posts Media Videos Starter Packs
They’re really digging for some of these. “I’m not sad he died” does not seem like a fireable statement.
That’s not what he did though. He publicly trounced them in a mean way for his own personal profit. People who want to actually learn use tactics like those the Naval academy teach like presenting a strong form of your opponents argument. The truths he told were based on cherru-picking and bias.
Let’s be real: Charlie didn’t deserve to get killed.

But the guy built his career on traveling to college campuses to clown on teenagers with zero debate experience while he himself had 100’s or 1000’s of hours of debate experience.

He was like a teenager playing basketball against 9 year olds.
I admire him for really putting his money where is mouth is
At least Charlie died doing what he loved
Is it possible to watch the ledgers for activity on any of these recipient wallet addresses?
It’s exhausting to maintain standards when others don’t, but consider: your standards aren’t for them.

Your standards are for you, to remain someone you respect.
Ya know, as a matter of fact I wish atlas would shrug.
Reposted by Mark Pauley
After getting various answers from ChatGPT that obviously were based on Wikipedia, I asked it if Open AI supported Wikimedia. It found Apple, Google.org, plus individual donors like Omidyar Network did, but not OpenAI. So I asked it to prepare a call for social media. This is it. Let's use it:
Love the software, hate the long term unstable ABI
Stop making it exponentially bigger?
Here’s an older piece: arxiv.org/abs/2310.10402

Here is one relating healthcare applications where privacy is particularly sensitive: arxiv.org/html/2412.20...

Anyway I feel like ML needs to solve the technical part of the problem you are pointing out, and tech co.’s have a lot of trust to rebuild
Real-Fake: Effective Training Data Synthesis Through Distribution Matching
Synthetic training data has gained prominence in numerous learning tasks and scenarios, offering advantages such as dataset augmentation, generalization evaluation, and privacy preservation. Despite t...
arxiv.org
Diff privacy can be used to steer synthetic training data to resemble (or at least comprise the same sample distribution as) actual user data. I was looking for references to papers I’ve seen on this, but am drawing a blank.
I personally would be more at ease if the resulting dataset was
1: provably private and
2: open source, so that it benefits all similar systems going forward.

Anyway yeah I totally agree that companies snooping with zero privacy guarantees and likely zero benefit to anybody is bad
I used to think this way, and then I started playing with modern machine learning systems. The data is basically all of the spec.
So but there are means of getting approximate sets of data that can’t be used to snoop: en.wikipedia.org/wiki/Differe...

This presupposes trust though which is lacking
Differential privacy - Wikipedia
en.wikipedia.org
If you don’t trust the tech companies to handle your data with care, respect and privacy that’s 100% understandable. But the best you can hope for when you don’t give feedback is that someone near enough to your demographic is using the tech in a similar way as you and they are providing feedback.
In theory I agree but unfortunately that’s not how software works anymore.

The world is a squishy place with a huge long tail distribution and the software that best fits that is basically an input / output mapping that needs just ungodly amounts of data to fit your specific needs.
I like Bluesky except when people start talking about AI.
Like I get it, you don’t like it.
From where I was standing they didn’t which is why we made fun of them
They are trying to game Agentic AI crawlers
My paper had obscene pictures on it after I drew the with this pen!
Yes except we don’t know how to fix it
I mean it’s about as good as homies fantasy conspiracy and it only took you two prompts
I feel like I might try to get GPT into SCP mode just for entertainment purposes. Sounds wicked
The model is probably in some sci-fi / fiction pathway. This reads like something out of an SCP