Thomas Steinke
@stein.ke
Computer science, math, machine learning, (differential) privacy
Researcher at Google DeepMind
Kiwi🇳🇿 in California🇺🇸
http://stein.ke/
Researcher at Google DeepMind
Kiwi🇳🇿 in California🇺🇸
http://stein.ke/
Pinned
Thomas Steinke
@stein.ke
· Nov 14
I'm going to slowly repost my math notes from the other site🐦 here🦋; it's the only thing I posted over there that I think may have some long-term value & worth not deleting.
These started out as notes for myself, but people seem to appreciate them. 😅
I'll keep track of all of them in this thread.
These started out as notes for myself, but people seem to appreciate them. 😅
I'll keep track of all of them in this thread.
Dear X, Thank you for the kind invitation. I will decline it, but please give me a few days to think up a reason. Best, Thomas
January 6, 2026 at 10:31 PM
Dear X, Thank you for the kind invitation. I will decline it, but please give me a few days to think up a reason. Best, Thomas
Are you running your own PDS?
January 6, 2026 at 9:56 PM
Are you running your own PDS?
Finally: Happy New Year folks!
January 1, 2026 at 3:51 PM
Finally: Happy New Year folks!
Everyone is celebrating the new year already and I'm just sitting here UTC-08:00ing.
January 1, 2026 at 4:46 AM
Everyone is celebrating the new year already and I'm just sitting here UTC-08:00ing.
TIL that the etymology of "California" is plausibly related to "Caliphate".
December 31, 2025 at 4:09 AM
TIL that the etymology of "California" is plausibly related to "Caliphate".
The quote posts on this are pretty funny.
And, putting aside the name, this startup seems like it doesn't really solve the problem that burglar alarms just aren't very effective.
And, putting aside the name, this startup seems like it doesn't really solve the problem that burglar alarms just aren't very effective.
Sauron is appearing on the scene as concerns rise about crime among the most wealthy.
Sauron, the high-end home security startup for "super premium" customers, plucks a new CEO out of Sonos | TechCrunch
Sauron is appearing on the scene as concerns rise about crime among the most wealthy.
techcrunch.com
December 30, 2025 at 12:05 AM
The quote posts on this are pretty funny.
And, putting aside the name, this startup seems like it doesn't really solve the problem that burglar alarms just aren't very effective.
And, putting aside the name, this startup seems like it doesn't really solve the problem that burglar alarms just aren't very effective.
I have a competing startup called DeathStar Inc. We will make your home impenetrable!
Sauron is appearing on the scene as concerns rise about crime among the most wealthy.
Sauron, the high-end home security startup for "super premium" customers, plucks a new CEO out of Sonos | TechCrunch
Sauron is appearing on the scene as concerns rise about crime among the most wealthy.
techcrunch.com
December 30, 2025 at 12:05 AM
I have a competing startup called DeathStar Inc. We will make your home impenetrable!
Wikipedia should have a bot that, when the time comes, automatically changes "[subject] is" to "[subject] was" etc. This bot should be known as the killer app.
December 29, 2025 at 4:00 PM
Wikipedia should have a bot that, when the time comes, automatically changes "[subject] is" to "[subject] was" etc. This bot should be known as the killer app.
Imagine you are a trained next word prediction model and you see this.
December 29, 2025 at 5:59 AM
Imagine you are a trained next word prediction model and you see this.
Say we create a separate review process for papers that are 100% AI generated. Where do we draw the line? What if it's 95% AI? 80%? 50%? 25% – a lot of papers are already there?
(Obviously it's impossible to quantify the percentage of work that is AI, but I hope my point is clear nonetheless.)
(Obviously it's impossible to quantify the percentage of work that is AI, but I hope my point is clear nonetheless.)
December 28, 2025 at 3:18 AM
Say we create a separate review process for papers that are 100% AI generated. Where do we draw the line? What if it's 95% AI? 80%? 50%? 25% – a lot of papers are already there?
(Obviously it's impossible to quantify the percentage of work that is AI, but I hope my point is clear nonetheless.)
(Obviously it's impossible to quantify the percentage of work that is AI, but I hope my point is clear nonetheless.)
This out of sequence reply seems to have broken the thread. Main thread continues here:
bsky.app/profile/stei...
bsky.app/profile/stei...
4a. The supreme court will uphold birthright citizenship and block Trump's executive order.
4b. The decision will not be close - either 9-0 or 8-1.
4c. SCOTUS will dodge the constitutional question and instead cite 8 U.S.C. § 1401(a)
4b. The decision will not be close - either 9-0 or 8-1.
4c. SCOTUS will dodge the constitutional question and instead cite 8 U.S.C. § 1401(a)
December 23, 2025 at 4:58 PM
This out of sequence reply seems to have broken the thread. Main thread continues here:
bsky.app/profile/stei...
bsky.app/profile/stei...
In the interest of making prediction #5 quantifiable:
5d. NVDA will underperform SPY in 2026.
5d. NVDA will underperform SPY in 2026.
December 21, 2025 at 10:12 PM
In the interest of making prediction #5 quantifiable:
5d. NVDA will underperform SPY in 2026.
5d. NVDA will underperform SPY in 2026.
To be clear, I count bombing Iran as him having gone to war in 2025. I expect something similar in 2026 e.g. against Venezuela.
December 21, 2025 at 8:38 PM
To be clear, I count bombing Iran as him having gone to war in 2025. I expect something similar in 2026 e.g. against Venezuela.
5a. Frontier AI progress will continue, but at a slower pace than previously.
5b. Smaller & open-source models will narrow the gap with frontier models.
5c. Demand for GPUs will start to slow down or, at least, expected future demand will come down.
[Yes, these predictions are hard to quantify.]
5b. Smaller & open-source models will narrow the gap with frontier models.
5c. Demand for GPUs will start to slow down or, at least, expected future demand will come down.
[Yes, these predictions are hard to quantify.]
December 21, 2025 at 7:56 PM
5a. Frontier AI progress will continue, but at a slower pace than previously.
5b. Smaller & open-source models will narrow the gap with frontier models.
5c. Demand for GPUs will start to slow down or, at least, expected future demand will come down.
[Yes, these predictions are hard to quantify.]
5b. Smaller & open-source models will narrow the gap with frontier models.
5c. Demand for GPUs will start to slow down or, at least, expected future demand will come down.
[Yes, these predictions are hard to quantify.]
4a. The supreme court will uphold birthright citizenship and block Trump's executive order.
4b. The decision will not be close - either 9-0 or 8-1.
4c. SCOTUS will dodge the constitutional question and instead cite 8 U.S.C. § 1401(a)
4b. The decision will not be close - either 9-0 or 8-1.
4c. SCOTUS will dodge the constitutional question and instead cite 8 U.S.C. § 1401(a)
December 21, 2025 at 7:56 PM
4a. The supreme court will uphold birthright citizenship and block Trump's executive order.
4b. The decision will not be close - either 9-0 or 8-1.
4c. SCOTUS will dodge the constitutional question and instead cite 8 U.S.C. § 1401(a)
4b. The decision will not be close - either 9-0 or 8-1.
4c. SCOTUS will dodge the constitutional question and instead cite 8 U.S.C. § 1401(a)
3a. Trump will go to war with another country. (Although war won't officially be declared.)
3b. Trump will not win the Nobel Peace Prize.
3b. Trump will not win the Nobel Peace Prize.
December 21, 2025 at 7:56 PM
3a. Trump will go to war with another country. (Although war won't officially be declared.)
3b. Trump will not win the Nobel Peace Prize.
3b. Trump will not win the Nobel Peace Prize.
2a. Democrats will eke out a win in the US House of Representatives.
2b. But Democrats won't get a majority in the Senate.
2b. But Democrats won't get a majority in the Senate.
December 21, 2025 at 7:56 PM
2a. Democrats will eke out a win in the US House of Representatives.
2b. But Democrats won't get a majority in the Senate.
2b. But Democrats won't get a majority in the Senate.
1. New Zealand's governing coalition (Nat+ACT+NZF) will win a majority. (I wouldn't bet on them sticking together though.)
December 21, 2025 at 7:56 PM
1. New Zealand's governing coalition (Nat+ACT+NZF) will win a majority. (I wouldn't bet on them sticking together though.)
In the interest of future i-told-you-sos, I'm going to post some predictions for 2026.
I'm hoping these are adventurous enough to not all be right. 😜
I'm hoping these are adventurous enough to not all be right. 😜
December 21, 2025 at 7:56 PM
In the interest of future i-told-you-sos, I'm going to post some predictions for 2026.
I'm hoping these are adventurous enough to not all be right. 😜
I'm hoping these are adventurous enough to not all be right. 😜
Yes, orbiting data centers don't make sense economically, but people do crazy things in the name of latency.
Space-based AI data centers are Big Tech’s latest fad and Silicon Valley’s newest investable venture.
Yet, scientists who study space remain skeptical of the idea.
Read more from @elissawelle.bsky.social: www.theverge.com/ai-artificia...
Yet, scientists who study space remain skeptical of the idea.
Read more from @elissawelle.bsky.social: www.theverge.com/ai-artificia...
December 19, 2025 at 12:19 AM
Yes, orbiting data centers don't make sense economically, but people do crazy things in the name of latency.
Is there any plausible threat model that justifies sending a login link and the password as two separate emails?
December 12, 2025 at 11:27 PM
Is there any plausible threat model that justifies sending a login link and the password as two separate emails?
It's also about family size. The calculus of having your travel paid for is very different if you still have to pay for your spouse+kids.
December 6, 2025 at 5:47 PM
It's also about family size. The calculus of having your travel paid for is very different if you still have to pay for your spouse+kids.
Sisyphus got off easy with that boulder.
December 6, 2025 at 5:23 PM
Sisyphus got off easy with that boulder.
Searle's "Chinese room" thought experiment is actually describing an AI-powered hell in which the damned are condemned to manually evaluate an LLM weight-by-weight. Their souls must labor for millennia to compute a single inscrutable token.
December 6, 2025 at 4:08 AM
Searle's "Chinese room" thought experiment is actually describing an AI-powered hell in which the damned are condemned to manually evaluate an LLM weight-by-weight. Their souls must labor for millennia to compute a single inscrutable token.