Greg Faletto
@gregoryfaletto.com
Statistician, Data Scientist.
Hell right now we're living through "you made me feel oppressed, I'm using the power of the state to oppress you" and conservatives are going to be *shocked* in four years when we don't see it as "even"
December 8, 2025 at 8:29 PM
Hell right now we're living through "you made me feel oppressed, I'm using the power of the state to oppress you" and conservatives are going to be *shocked* in four years when we don't see it as "even"
Oh man if "I materially hurt you, you made me feel bad about it, now we're even" isn't the defining conservative ideology of our time
December 8, 2025 at 8:23 PM
Oh man if "I materially hurt you, you made me feel bad about it, now we're even" isn't the defining conservative ideology of our time
"you shouldn't just do things"
December 8, 2025 at 8:06 PM
"you shouldn't just do things"
whoa major heel turn
December 7, 2025 at 7:56 PM
whoa major heel turn
National lab would be comparably paid, wouldn't be shocking if the private research lab didn't pay life-changingly more
December 7, 2025 at 3:31 PM
National lab would be comparably paid, wouldn't be shocking if the private research lab didn't pay life-changingly more
It's crazy how many people think Biden failed to cancel any student loan debt. $190 billion! www.nasfaa.org/news-item/35...
December 6, 2025 at 5:52 PM
It's crazy how many people think Biden failed to cancel any student loan debt. $190 billion! www.nasfaa.org/news-item/35...
are you shorting Palantir harder than ever? bsky.app/profile/ronf...
December 5, 2025 at 9:22 PM
are you shorting Palantir harder than ever? bsky.app/profile/ronf...
One thing about Donald Trump was that he has always genuinely acted like the 2020 election was stolen from him and his millions of voters, with all the gravity that playing that part entails
December 5, 2025 at 8:53 PM
One thing about Donald Trump was that he has always genuinely acted like the 2020 election was stolen from him and his millions of voters, with all the gravity that playing that part entails
A lot of voters would respect Democrats and their beliefs more if they were more willing to stand by it. Part of why the median voter didn't believe one history's most wicked Americans was trying to be president again was that his opponents with the power to do something about it didn't act like it
December 5, 2025 at 8:53 PM
A lot of voters would respect Democrats and their beliefs more if they were more willing to stand by it. Part of why the median voter didn't believe one history's most wicked Americans was trying to be president again was that his opponents with the power to do something about it didn't act like it
yes or "you don't deserve good things older generations provided me, or our generation provided each other, because now there's iphones so you have it easy"
December 4, 2025 at 8:54 PM
yes or "you don't deserve good things older generations provided me, or our generation provided each other, because now there's iphones so you have it easy"
Yes the lenses of "how much did DOGE trim the federal budget?" vs "how much did DOGE affect headcount?" vs "what were the costs and benefits of DOGE" lead to very different answers to the question "how big of a deal was DOGE?"
December 3, 2025 at 4:19 PM
Yes the lenses of "how much did DOGE trim the federal budget?" vs "how much did DOGE affect headcount?" vs "what were the costs and benefits of DOGE" lead to very different answers to the question "how big of a deal was DOGE?"
Nice! On forcing the LLM to read--often I'll start a conversation by providing a file and asking it to summarize it before I ask it anything else.
I like using backticks too--unless my prompt is something I would type into Google ("tallest building in SF?") I usually write prompts in Markdown
I like using backticks too--unless my prompt is something I would type into Google ("tallest building in SF?") I usually write prompts in Markdown
December 2, 2025 at 8:40 PM
Nice! On forcing the LLM to read--often I'll start a conversation by providing a file and asking it to summarize it before I ask it anything else.
I like using backticks too--unless my prompt is something I would type into Google ("tallest building in SF?") I usually write prompts in Markdown
I like using backticks too--unless my prompt is something I would type into Google ("tallest building in SF?") I usually write prompts in Markdown
yeah I think if you're classifying techniques you've basically got "deep learning" (mostly transformers these days) and "regularized m-estimators" (where most of classic ML, matrix completion/recommenders, PageRank, etc. goes). But I think that has little to do with what people like and don't like
December 2, 2025 at 1:05 AM
yeah I think if you're classifying techniques you've basically got "deep learning" (mostly transformers these days) and "regularized m-estimators" (where most of classic ML, matrix completion/recommenders, PageRank, etc. goes). But I think that has little to do with what people like and don't like
for example quite a lot more people worked in textile production after the industrial revolution than before. Bank teller employment also increased throughout the adoption of ATMs
December 1, 2025 at 9:57 PM
for example quite a lot more people worked in textile production after the industrial revolution than before. Bank teller employment also increased throughout the adoption of ATMs
often, but not always, true! reduced cost of producing expands the market that can be served profitably, increasing demand for labor output. Sometimes the increase doesn't make up the productivity gain and net employment goes down, but sometimes it more than makes up for it and employment goes up
December 1, 2025 at 9:57 PM
often, but not always, true! reduced cost of producing expands the market that can be served profitably, increasing demand for labor output. Sometimes the increase doesn't make up the productivity gain and net employment goes down, but sometimes it more than makes up for it and employment goes up
"We are doing GATTACA from the classic sci-fi movie 'Don't do GATTACA'"
December 1, 2025 at 9:35 PM
"We are doing GATTACA from the classic sci-fi movie 'Don't do GATTACA'"
Also there's this existing solution we can fall back on, so we've ruled out any outcome worse than "rely on desalination way more"
December 1, 2025 at 4:42 AM
Also there's this existing solution we can fall back on, so we've ruled out any outcome worse than "rely on desalination way more"
Maybe I'm wrong but I'm pretty sure every month where compute costs go down exponentially and token usage goes up exponentially is a month that AI companies (and their investors) are quite pleased about. More entrenched demand and decreasing costs make it easier to find a sustainable business model
December 1, 2025 at 12:41 AM
Maybe I'm wrong but I'm pretty sure every month where compute costs go down exponentially and token usage goes up exponentially is a month that AI companies (and their investors) are quite pleased about. More entrenched demand and decreasing costs make it easier to find a sustainable business model
Yes it's been an ongoing tug of war as the companies have been searching for pricing models that work for both them and customers. I'd argue Jan 2025 was a long time ago by AI standards, Claude Code didn't even exist yet.
December 1, 2025 at 12:41 AM
Yes it's been an ongoing tug of war as the companies have been searching for pricing models that work for both them and customers. I'd argue Jan 2025 was a long time ago by AI standards, Claude Code didn't even exist yet.
My understanding is that monthly subscribers are generally more profitable than API customers who pay per token, because the flat rate ends up being higher. They can always (and do in at least some cases) rate limit subscription customers who hit a token limit
December 1, 2025 at 12:12 AM
My understanding is that monthly subscribers are generally more profitable than API customers who pay per token, because the flat rate ends up being higher. They can always (and do in at least some cases) rate limit subscription customers who hit a token limit
Yeah I believe each color line represents roughly equal capability (but not necessarily the same model). Not totally clear to me which data points correspond to exactly which model
November 30, 2025 at 8:56 PM
Yeah I believe each color line represents roughly equal capability (but not necessarily the same model). Not totally clear to me which data points correspond to exactly which model
Like, if tomorrow McDonald's sold 100 times as many hamburgers and hamburger prices went down by 90%, they wouldn't be wringing their hands about the fact that their marginal costs went up 10x.
November 30, 2025 at 7:55 PM
Like, if tomorrow McDonald's sold 100 times as many hamburgers and hamburger prices went down by 90%, they wouldn't be wringing their hands about the fact that their marginal costs went up 10x.
Total inference cost is [cost per token] times [tokens used]. AI companies want the first number to go down (ideally a lot) and the second number to go up (ideally a lot). So total inference cost isn't a great metric.
First number is in fact going down dramatically (the y-axis is on a log scale).
First number is in fact going down dramatically (the y-axis is on a log scale).
November 30, 2025 at 7:54 PM
Total inference cost is [cost per token] times [tokens used]. AI companies want the first number to go down (ideally a lot) and the second number to go up (ideally a lot). So total inference cost isn't a great metric.
First number is in fact going down dramatically (the y-axis is on a log scale).
First number is in fact going down dramatically (the y-axis is on a log scale).
Consider being honest with him that he's acting like a loser. People wake up every day and choose to act contemptibly for no valid reason, the rest of us don't have to forever tolerate their subsequent whining about being treated with contempt
November 28, 2025 at 11:09 PM
Consider being honest with him that he's acting like a loser. People wake up every day and choose to act contemptibly for no valid reason, the rest of us don't have to forever tolerate their subsequent whining about being treated with contempt