Paul Rietschka
@prietschka.bsky.social
Data science/machine learning in the Pacific Northwest. In Minneapolis through early 2025.
Sigh. This place.
November 30, 2025 at 1:47 AM
Sigh. This place.
Materials costs are insane, the amount of concrete being poured on the new part of the Victorian we're renovating here in Whittier is costing a f*cking fortune. So. Much. Concrete.
November 29, 2025 at 6:52 PM
Materials costs are insane, the amount of concrete being poured on the new part of the Victorian we're renovating here in Whittier is costing a f*cking fortune. So. Much. Concrete.
I’m so sick of the “cheap Hardieboard slathered all over everything” look that seems to be everywhere.
November 29, 2025 at 5:51 PM
I’m so sick of the “cheap Hardieboard slathered all over everything” look that seems to be everywhere.
Oh definitely. Those sweet terms? Gone.
As it should be! The UK, if it were to rejoin the EU, should be **prohibited** from leaving and joining the eurozone should be compulsory.
As it should be! The UK, if it were to rejoin the EU, should be **prohibited** from leaving and joining the eurozone should be compulsory.
November 29, 2025 at 4:26 PM
Oh definitely. Those sweet terms? Gone.
As it should be! The UK, if it were to rejoin the EU, should be **prohibited** from leaving and joining the eurozone should be compulsory.
As it should be! The UK, if it were to rejoin the EU, should be **prohibited** from leaving and joining the eurozone should be compulsory.
Omg, dude. No. Please leave this thread.
November 29, 2025 at 4:07 PM
Omg, dude. No. Please leave this thread.
Dude, the dismissiveness you're showing here is really something.
I suggest you log off Bluesky and maybe log into your Coinbase or Kalshi account and place some bets? Seems you're better off there than in my replies here.
I suggest you log off Bluesky and maybe log into your Coinbase or Kalshi account and place some bets? Seems you're better off there than in my replies here.
November 29, 2025 at 4:02 PM
Dude, the dismissiveness you're showing here is really something.
I suggest you log off Bluesky and maybe log into your Coinbase or Kalshi account and place some bets? Seems you're better off there than in my replies here.
I suggest you log off Bluesky and maybe log into your Coinbase or Kalshi account and place some bets? Seems you're better off there than in my replies here.
I don't respond well to "debate me, bro" nonsense, and your dismissal of Zitron's well-researched reporting means we have nothing to talk about.
Byeeee.
Byeeee.
November 29, 2025 at 3:57 PM
I don't respond well to "debate me, bro" nonsense, and your dismissal of Zitron's well-researched reporting means we have nothing to talk about.
Byeeee.
Byeeee.
Same, I would always buy the largest Macbook Pro possible and max out the specs, as I could just justify it as a business cost.
But, at least as someone whose work is solidly DS stuff, that just doesn't hold in 2025.
But, at least as someone whose work is solidly DS stuff, that just doesn't hold in 2025.
November 29, 2025 at 3:42 PM
Same, I would always buy the largest Macbook Pro possible and max out the specs, as I could just justify it as a business cost.
But, at least as someone whose work is solidly DS stuff, that just doesn't hold in 2025.
But, at least as someone whose work is solidly DS stuff, that just doesn't hold in 2025.
Inference costs are increasing, my friend. neoclouds are financial engineering, and will implode in the very near future.
Read Zitron's reporting on this.
Read Zitron's reporting on this.
November 29, 2025 at 3:38 PM
Inference costs are increasing, my friend. neoclouds are financial engineering, and will implode in the very near future.
Read Zitron's reporting on this.
Read Zitron's reporting on this.
I think that workflow, where more and more moves to remote compute, is specifically good for DS stuff, but also holds for more typical SWE stuff, too Which means that OSS GPT on a local machine isn't going to truly, if you ask me, be a thing.
Just too much work for too little payoff.
Just too much work for too little payoff.
November 29, 2025 at 3:35 PM
I think that workflow, where more and more moves to remote compute, is specifically good for DS stuff, but also holds for more typical SWE stuff, too Which means that OSS GPT on a local machine isn't going to truly, if you ask me, be a thing.
Just too much work for too little payoff.
Just too much work for too little payoff.
I mean, Workbench is great for just the typical EDA I do, what I call "f*cking around;" I can just use a small instance and it costs basically nothing. The fact you're inside GCP means I don't have to think about security and such.
It's just great.
It's just great.
November 29, 2025 at 3:35 PM
I mean, Workbench is great for just the typical EDA I do, what I call "f*cking around;" I can just use a small instance and it costs basically nothing. The fact you're inside GCP means I don't have to think about security and such.
It's just great.
It's just great.
This is where we diverge in hardware use. I have two 15-in. Macbook Airs that I use for work, the blue/black version, and I won't lie it's the best machine I've ever owned.
But having them means I've migrated to using Vertex for 99.9% of what I do, which is frankly what all DSs should be doing.
But having them means I've migrated to using Vertex for 99.9% of what I do, which is frankly what all DSs should be doing.
November 29, 2025 at 3:35 PM
This is where we diverge in hardware use. I have two 15-in. Macbook Airs that I use for work, the blue/black version, and I won't lie it's the best machine I've ever owned.
But having them means I've migrated to using Vertex for 99.9% of what I do, which is frankly what all DSs should be doing.
But having them means I've migrated to using Vertex for 99.9% of what I do, which is frankly what all DSs should be doing.
Yeah, but even that requires a large machine. I know myself and a lot of DSs use less powerful machines these days and rely on remote compute.
It's certainly possible to use a local LM, but it's not ideal and performance will be substantially reduced. I just don't see it as the long term answer.
It's certainly possible to use a local LM, but it's not ideal and performance will be substantially reduced. I just don't see it as the long term answer.
November 29, 2025 at 3:25 PM
Yeah, but even that requires a large machine. I know myself and a lot of DSs use less powerful machines these days and rely on remote compute.
It's certainly possible to use a local LM, but it's not ideal and performance will be substantially reduced. I just don't see it as the long term answer.
It's certainly possible to use a local LM, but it's not ideal and performance will be substantially reduced. I just don't see it as the long term answer.
So what will happen is the age of subsidized compute will end, cos. will receive a **single** billing from AWS or Google Cloud that makes accounting have a seizure, and a directive will be issued for **all** MCP servers to be shut down, immediately.
Then it's all over.
Then it's all over.
November 29, 2025 at 3:22 PM
So what will happen is the age of subsidized compute will end, cos. will receive a **single** billing from AWS or Google Cloud that makes accounting have a seizure, and a directive will be issued for **all** MCP servers to be shut down, immediately.
Then it's all over.
Then it's all over.
Yeah, there's no way this pencils out.
This is like at automaker building cars for $250,000/ea., selling them for $40,000, and pretending things will eventually pencil out via price increases.
But once you've created demand at a price point, increasing costs exponentially doesn't work.
This is like at automaker building cars for $250,000/ea., selling them for $40,000, and pretending things will eventually pencil out via price increases.
But once you've created demand at a price point, increasing costs exponentially doesn't work.
November 29, 2025 at 3:22 PM
Yeah, there's no way this pencils out.
This is like at automaker building cars for $250,000/ea., selling them for $40,000, and pretending things will eventually pencil out via price increases.
But once you've created demand at a price point, increasing costs exponentially doesn't work.
This is like at automaker building cars for $250,000/ea., selling them for $40,000, and pretending things will eventually pencil out via price increases.
But once you've created demand at a price point, increasing costs exponentially doesn't work.
No way local models can support this given hardware reqs, and the subsidization will soon go away.
What we'll see is a wave of MCP servers being shut down everywhere.
What we'll see is a wave of MCP servers being shut down everywhere.
November 29, 2025 at 3:12 PM
No way local models can support this given hardware reqs, and the subsidization will soon go away.
What we'll see is a wave of MCP servers being shut down everywhere.
What we'll see is a wave of MCP servers being shut down everywhere.
Again: I’m so over people defending MCP this way. Now I’m being a little ungenerous to a friendly face so I’ll rescind the use of “desperate” as I’m just being my typical f^*&y b*^!h here, so apologies.
But my god man, the humanity, the compute use!
But my god man, the humanity, the compute use!
November 29, 2025 at 3:05 PM
Again: I’m so over people defending MCP this way. Now I’m being a little ungenerous to a friendly face so I’ll rescind the use of “desperate” as I’m just being my typical f^*&y b*^!h here, so apologies.
But my god man, the humanity, the compute use!
But my god man, the humanity, the compute use!
Can someone tell me why there isn’t a grassroots movement to re-join the EU?
November 29, 2025 at 3:02 PM
Can someone tell me why there isn’t a grassroots movement to re-join the EU?
Sounds complicated, brittle, and compute intensive.
Because it is.
I know we disagree on the efficacy of this garbage technology, I know you **desperately** want to justify/wave away the compute use here, and the brittleness.
So…I think we’ll end this discussion here my dear Canary.
Because it is.
I know we disagree on the efficacy of this garbage technology, I know you **desperately** want to justify/wave away the compute use here, and the brittleness.
So…I think we’ll end this discussion here my dear Canary.
November 29, 2025 at 2:59 PM
Sounds complicated, brittle, and compute intensive.
Because it is.
I know we disagree on the efficacy of this garbage technology, I know you **desperately** want to justify/wave away the compute use here, and the brittleness.
So…I think we’ll end this discussion here my dear Canary.
Because it is.
I know we disagree on the efficacy of this garbage technology, I know you **desperately** want to justify/wave away the compute use here, and the brittleness.
So…I think we’ll end this discussion here my dear Canary.
But do you think any Democratic president has the stomach for this?
Or to do what will be needed if this happens, which most likely involved arresting and holding the GOP justices on nat security grounds?
Good luck with that….
Or to do what will be needed if this happens, which most likely involved arresting and holding the GOP justices on nat security grounds?
Good luck with that….
November 29, 2025 at 2:55 PM
But do you think any Democratic president has the stomach for this?
Or to do what will be needed if this happens, which most likely involved arresting and holding the GOP justices on nat security grounds?
Good luck with that….
Or to do what will be needed if this happens, which most likely involved arresting and holding the GOP justices on nat security grounds?
Good luck with that….