Found 9 bookmarks
Custom sorting
(1) † lucia scarlet 🩸 on X: ""ok but how do we know Private Cloud Compute is actually private. do we just take Apple's word for it" Wrong 1. they will regularly release images of the entire custom OS powering these servers for public inspection, along with a virtualisation environment, so security https://t.co/NUvmLYYAdh" / X
(1) † lucia scarlet 🩸 on X: ""ok but how do we know Private Cloud Compute is actually private. do we just take Apple's word for it" Wrong 1. they will regularly release images of the entire custom OS powering these servers for public inspection, along with a virtualisation environment, so security https://t.co/NUvmLYYAdh" / X
1. they will regularly release images of the entire custom OS powering these servers for public inspection, along with a virtualisation environment, so security… — † lucia scarlet 🩸 (@luciascarlet)
#reference#chips#privacy#security#open source#research#academic#companies/apple#apple-silicon#ai#servers#discourse/counterclaims
·x.com·
(1) † lucia scarlet 🩸 on X: ""ok but how do we know Private Cloud Compute is actually private. do we just take Apple's word for it" Wrong 1. they will regularly release images of the entire custom OS powering these servers for public inspection, along with a virtualisation environment, so security https://t.co/NUvmLYYAdh" / X
swyx 🔜 ai.engineer on X: "IMO @AnthropicAI is very close to making a breakthrough in productizable interpretability. For ~4 years all we've had to really control LLMs is temperature/top_p and logit bias. We recently got `seed` and constrained structured output, with `interactive=false` on the way. But https://t.co/Z8C28evxgN" / X
swyx 🔜 ai.engineer on X: "IMO @AnthropicAI is very close to making a breakthrough in productizable interpretability. For ~4 years all we've had to really control LLMs is temperature/top_p and logit bias. We recently got `seed` and constrained structured output, with `interactive=false` on the way. But https://t.co/Z8C28evxgN" / X
For ~4 years all we've had to really control LLMs is temperature/top_p and logit bias. We recently got `seed` and constrained structured output, with `interactive=false` on the way. But… — swyx 🔜 ai.engineer (@swyx)
#ai#research#academic#threads#companies/Anthropic
·x.com·
swyx 🔜 ai.engineer on X: "IMO @AnthropicAI is very close to making a breakthrough in productizable interpretability. For ~4 years all we've had to really control LLMs is temperature/top_p and logit bias. We recently got `seed` and constrained structured output, with `interactive=false` on the way. But https://t.co/Z8C28evxgN" / X
Ethan Mollick on Twitter
Ethan Mollick on Twitter
A paper that really illustrates both the unexpected power, and unexpected risks, that come from LLMs.Given text of anonymous posts on Reddit, GPT-4 can infer things like income, gender & location with 85%+ accuracy at 1% of the cost required by humans. https://t.co/qcrodZsgUQ pic.twitter.com/at8NfwLxjr— Ethan Mollick (@emollick) October 20, 2023
#research#pdf#stats#LLMs#privacy
·twitter.com·
Ethan Mollick on Twitter
Crémieux on Twitter
Crémieux on Twitter
This is a pretty large effect!TL;DR: People whose work is graded later tend to be graded more harshly.If your grading order is alphabetical, that'll unduly penalize people with alphabetically-later surnames. https://t.co/Af7prAV7rx— Crémieux (@cremieuxrecueil) October 16, 2023
#research#stats#education#learning#testing#psychology
·x.com·
Crémieux on Twitter
Scott Santens on Twitter
Scott Santens on Twitter
Sources:1. https://t.co/4x5lVCIwZ42. https://t.co/Bc4oFxTnd03. https://t.co/tabNU56LEc4. https://t.co/0qk93bEv0S5. https://t.co/yRh517CHbe6. https://t.co/m17g7VWvhI7. https://t.co/xtnR3PAXKM8. https://t.co/9P9aZCNWx79. https://t.co/Kce9FEWDiO#BasicIncome pic.twitter.com/zBpqVZOw3K— Scott Santens (@scottsantens) April 12, 2023
#stats#research#ubi#finance#threads#govt#public-life
·twitter.com·
Scott Santens on Twitter