Gradient Assent (@GradientAssent) header
Gradient Assent (@GradientAssent) profile picture

Crew member, ship of Theseus. Supply-side futurist. Building new tools for thought with NLP. Latest project: https://teapawt.net/t.co/hgGqdE8HEi

San Francisco, CA
Joined April 2020
Photo by Gradient Assent (@GradientAssent)Photo by Gradient Assent (@GradientAssent)Photo by Gradient Assent (@GradientAssent)Photo by Gradient Assent (@GradientAssent)Photo by Gradient Assent (@GradientAssent)Photo by Gradient Assent (@GradientAssent)Photo by Gradient Assent (@GradientAssent)Photo by Gradient Assent (@GradientAssent)Photo by Gradient Assent (@GradientAssent)Photo by Gradient Assent (@GradientAssent)
The only way technology is good is if it gets implemented by the proper procedure, i.e. shitposting my chosen politicians into power and paying for it with taxes on the specific people I want to tax :D
4
2
108
I think this is because he actually lives in it, doing much of his thinking in tweets. Many others who write about the memetisphere do so from above, looking down into it.
1
So we end up with overly formal and abstract analyses based on "network science" or blunt extensions of the top level tribal warfare narratives (the alt-right and their frog memes!)
What's the best explanation for why we don't see non-GPU ML accelerators deployed at scale anywhere except Google?
1
There are only two things to do: generate or discriminate.
1
3
Like a fox, I'm always trying to sniff out new things to assimilate into my grand unifying theory of everything.
2
Hey @Scholars_Stage and @palladiummag - did you notice you were cited in a State Dept. document? state.gov/wp-content/uploads…
2
17
Start of the fastest growing Substack subscriber base yet? astralcodexten.substack.com/…
1
An extremely important conversation. Does the current trajectory of information and communications technology inevitably lead to China-style digital totalitarianism? Thought-provoking arguments on both sides.
This quoted post is unavailable.
1
Trying out a @threadapalooza on my favorite thing to think about: The future of machine learning and language understanding technology and how it might change the world. 🧵👇
1
1
3
17/ And in the meantime, these systems are being deployed anyway, despite their shortcomings relative to human moderators. Users, employees advertisers, and even legislation demand that the moderation happen, whether or not it's technically or economically viable to do it well.
1
18/ So for now we just lower the bar. Precision of 50% - one false positive for every true positive - is often considered adequate in social media moderation models. And sometimes much lower. As long as we get those false negatives down as low as we can.