joel
@jmoonz.bsky.social
Pinned
Reposted by joel
NEW ANALYSIS: China's CO2 has now been 'flat or falling' for 21 months
* Down in 2025
* Still below Mar 2024
* Clean energy wave a key factor
If this is China's peak (TBC) it's the climate story of the century so far…
www.carbonbrief.org/...
* Down in 2025
* Still below Mar 2024
* Clean energy wave a key factor
If this is China's peak (TBC) it's the climate story of the century so far…
www.carbonbrief.org/...
February 12, 2026 at 7:38 AM
NEW ANALYSIS: China's CO2 has now been 'flat or falling' for 21 months
* Down in 2025
* Still below Mar 2024
* Clean energy wave a key factor
If this is China's peak (TBC) it's the climate story of the century so far…
www.carbonbrief.org/...
* Down in 2025
* Still below Mar 2024
* Clean energy wave a key factor
If this is China's peak (TBC) it's the climate story of the century so far…
www.carbonbrief.org/...
You forgot the end bit "... to prison"
February 12, 2026 at 6:10 AM
You forgot the end bit "... to prison"
well, Ian sure has gone insane since I last saw a post by him.
February 12, 2026 at 5:15 AM
well, Ian sure has gone insane since I last saw a post by him.
yeah, could be i just need to more aggressively prune context, but e.g. i asked codex to explain how several command line arguments interact in a script and it proceeded to make up an argument that didn't exist.
claude seems more reliable but i run out of usage pretty quickly, even w/o using opus.
claude seems more reliable but i run out of usage pretty quickly, even w/o using opus.
February 11, 2026 at 11:58 PM
yeah, could be i just need to more aggressively prune context, but e.g. i asked codex to explain how several command line arguments interact in a script and it proceeded to make up an argument that didn't exist.
claude seems more reliable but i run out of usage pretty quickly, even w/o using opus.
claude seems more reliable but i run out of usage pretty quickly, even w/o using opus.
I feel like i have days when AI coding is really good. Then there are days like today when the models seem completely moronic.
And unfortunately, I have no idea if that's a model issue or some opaque throttling for capacity reasons.
And unfortunately, I have no idea if that's a model issue or some opaque throttling for capacity reasons.
February 11, 2026 at 5:41 AM
I feel like i have days when AI coding is really good. Then there are days like today when the models seem completely moronic.
And unfortunately, I have no idea if that's a model issue or some opaque throttling for capacity reasons.
And unfortunately, I have no idea if that's a model issue or some opaque throttling for capacity reasons.
LinkedIn decided to start sending phone notifications to me about someone playing a game on LinkedIn, despite having all notifications from LinkedIn turned off.
They have no idea what they want the platform to be.
They have no idea what they want the platform to be.
February 10, 2026 at 10:33 PM
LinkedIn decided to start sending phone notifications to me about someone playing a game on LinkedIn, despite having all notifications from LinkedIn turned off.
They have no idea what they want the platform to be.
They have no idea what they want the platform to be.
chat - are 100 year corporate bonds a good idea? www.barrons.com/news/google-...
Google Turns To Century-long Debt To Build AI
Google-parent Alphabet will issue bonds maturing in 100 years as it continues to invest massively in infrastructure for artificial intelligence, according to data published Tuesday by Bloomberg.
www.barrons.com
February 10, 2026 at 8:16 PM
chat - are 100 year corporate bonds a good idea? www.barrons.com/news/google-...
Major pizza spike 😬 www.pizzint.watch
February 9, 2026 at 5:32 PM
Major pizza spike 😬 www.pizzint.watch
The real signal you are adult is you lock down discord privacy settings immediately.
February 9, 2026 at 4:51 PM
The real signal you are adult is you lock down discord privacy settings immediately.
I'm calling it - ai. com is the pets. com of our time.
"... Pets .com ... a defining cautionary tale of the dot-com bubble. Despite massive branding -including a famous sock puppet mascot and a 2000 Super Bowl ad - the company failed within nine months of its $82.5 million IPO"
"... Pets .com ... a defining cautionary tale of the dot-com bubble. Despite massive branding -including a famous sock puppet mascot and a 2000 Super Bowl ad - the company failed within nine months of its $82.5 million IPO"
February 9, 2026 at 7:43 AM
I'm calling it - ai. com is the pets. com of our time.
"... Pets .com ... a defining cautionary tale of the dot-com bubble. Despite massive branding -including a famous sock puppet mascot and a 2000 Super Bowl ad - the company failed within nine months of its $82.5 million IPO"
"... Pets .com ... a defining cautionary tale of the dot-com bubble. Despite massive branding -including a famous sock puppet mascot and a 2000 Super Bowl ad - the company failed within nine months of its $82.5 million IPO"
Reposted by joel
After making a big deal out of the roadside drug testing rollout, government, police, and media have all gone strangely quiet.
I wonder if it's not going as well as they envisioned?
I've become part of the organised resistance to it, which you can check out here: rrc.org.nz
I wonder if it's not going as well as they envisioned?
I've become part of the organised resistance to it, which you can check out here: rrc.org.nz
Roadside Rights Collective
rrc.org.nz
February 8, 2026 at 11:09 PM
After making a big deal out of the roadside drug testing rollout, government, police, and media have all gone strangely quiet.
I wonder if it's not going as well as they envisioned?
I've become part of the organised resistance to it, which you can check out here: rrc.org.nz
I wonder if it's not going as well as they envisioned?
I've become part of the organised resistance to it, which you can check out here: rrc.org.nz
Those statements have now been updated and also removed reference to FPR, since we don't really have the data to claim a true FPR.
(We wrote the website before we had access to the lab test results, so had to guess what MoT/Police/AG meant by 10% probability of a false positive).
(We wrote the website before we had access to the lab test results, so had to guess what MoT/Police/AG meant by 10% probability of a false positive).
February 9, 2026 at 1:51 AM
Those statements have now been updated and also removed reference to FPR, since we don't really have the data to claim a true FPR.
(We wrote the website before we had access to the lab test results, so had to guess what MoT/Police/AG meant by 10% probability of a false positive).
(We wrote the website before we had access to the lab test results, so had to guess what MoT/Police/AG meant by 10% probability of a false positive).
Now their testing of the device is based on some samples that have substances present 50% below the limit for detection, and they only did 2 "clear" samples.
But they also didn't test for cross-reactivity with other medicines or how individual saliva differences impact sensitivity.
But they also didn't test for cross-reactivity with other medicines or how individual saliva differences impact sensitivity.
February 9, 2026 at 1:09 AM
Now their testing of the device is based on some samples that have substances present 50% below the limit for detection, and they only did 2 "clear" samples.
But they also didn't test for cross-reactivity with other medicines or how individual saliva differences impact sensitivity.
But they also didn't test for cross-reactivity with other medicines or how individual saliva differences impact sensitivity.
But if we are doing something non-standard happy to be corrected.
The tests are well known to have poor accuracy though. That is why they need to do 2 in a row, *as well* as a lab test for an actual infringement to be issued.
The tests are well known to have poor accuracy though. That is why they need to do 2 in a row, *as well* as a lab test for an actual infringement to be issued.
February 9, 2026 at 1:04 AM
But if we are doing something non-standard happy to be corrected.
The tests are well known to have poor accuracy though. That is why they need to do 2 in a row, *as well* as a lab test for an actual infringement to be issued.
The tests are well known to have poor accuracy though. That is why they need to do 2 in a row, *as well* as a lab test for an actual infringement to be issued.
We will consider how to reword now that we have obtained access to lab results - but 8 to 16% of samples that should have tested negative came up positive depending on the substance.
We are using standard FPR = (False Positives / (False Positives + True Negatives))
We are using standard FPR = (False Positives / (False Positives + True Negatives))
February 9, 2026 at 1:01 AM
We will consider how to reword now that we have obtained access to lab results - but 8 to 16% of samples that should have tested negative came up positive depending on the substance.
We are using standard FPR = (False Positives / (False Positives + True Negatives))
We are using standard FPR = (False Positives / (False Positives + True Negatives))
Which model/hardware combo is that?
(I am using Nvidia hardware, so while I can get fast output, it is toasty)
(I am using Nvidia hardware, so while I can get fast output, it is toasty)
February 8, 2026 at 5:16 PM
Which model/hardware combo is that?
(I am using Nvidia hardware, so while I can get fast output, it is toasty)
(I am using Nvidia hardware, so while I can get fast output, it is toasty)
I find i have no patience to do things tediously by hand now if there is no reason to.
February 8, 2026 at 5:11 PM
I find i have no patience to do things tediously by hand now if there is no reason to.
Like any tool, if I make a mistake by not explaining myself well enough, or the nail doesn't go in with a single hammer strike, I don't expect the tool to apologize for it.
February 8, 2026 at 5:09 PM
Like any tool, if I make a mistake by not explaining myself well enough, or the nail doesn't go in with a single hammer strike, I don't expect the tool to apologize for it.
I haven't sorry - I have too many things I want to do that I that I don't have any spare tokens 😅
But it would be valuable to do some comparisons.
But it would be valuable to do some comparisons.
February 7, 2026 at 6:03 PM
I haven't sorry - I have too many things I want to do that I that I don't have any spare tokens 😅
But it would be valuable to do some comparisons.
But it would be valuable to do some comparisons.
I strongly resonate with everything your wrote.
My only way to slightly have more control is to host more of my stuff locally and use open source software.
Which has its own frustration, but it's nice when I can just play the media I have instead of worrying about a company's rights licensing
My only way to slightly have more control is to host more of my stuff locally and use open source software.
Which has its own frustration, but it's nice when I can just play the media I have instead of worrying about a company's rights licensing
February 7, 2026 at 5:54 PM
I strongly resonate with everything your wrote.
My only way to slightly have more control is to host more of my stuff locally and use open source software.
Which has its own frustration, but it's nice when I can just play the media I have instead of worrying about a company's rights licensing
My only way to slightly have more control is to host more of my stuff locally and use open source software.
Which has its own frustration, but it's nice when I can just play the media I have instead of worrying about a company's rights licensing
But I do appreciate how easy it is to start on a new/existing project now.
I didn't know how to setup dotnet, and chatgpt guided me through what I need to do to build and deploy the plugin.
(I'm using both OpenAI/Anthropic currently)
I didn't know how to setup dotnet, and chatgpt guided me through what I need to do to build and deploy the plugin.
(I'm using both OpenAI/Anthropic currently)
February 7, 2026 at 1:29 AM
But I do appreciate how easy it is to start on a new/existing project now.
I didn't know how to setup dotnet, and chatgpt guided me through what I need to do to build and deploy the plugin.
(I'm using both OpenAI/Anthropic currently)
I didn't know how to setup dotnet, and chatgpt guided me through what I need to do to build and deploy the plugin.
(I'm using both OpenAI/Anthropic currently)
I built and deployed it locally and it's working, but at the same time I found that the previous plugin update nuked my settings - so I'm not confident enough in the answer to submit it as a PR.
February 7, 2026 at 1:26 AM
I built and deployed it locally and it's working, but at the same time I found that the previous plugin update nuked my settings - so I'm not confident enough in the answer to submit it as a PR.
It indicated that was as simple as upgrading the plugin build dependencies to the newest version of Jellyfin.
February 7, 2026 at 1:26 AM
It indicated that was as simple as upgrading the plugin build dependencies to the newest version of Jellyfin.
Today's test of Claude Opus 4.6 -
Ask it to fix a plugin I use with Jellyfin which broke on the last update. I know nothing about Jellyfin internals or the plugin, and the latest Jellyfin update broke it.
Ask it to fix a plugin I use with Jellyfin which broke on the last update. I know nothing about Jellyfin internals or the plugin, and the latest Jellyfin update broke it.
February 6, 2026 at 9:07 PM
Today's test of Claude Opus 4.6 -
Ask it to fix a plugin I use with Jellyfin which broke on the last update. I know nothing about Jellyfin internals or the plugin, and the latest Jellyfin update broke it.
Ask it to fix a plugin I use with Jellyfin which broke on the last update. I know nothing about Jellyfin internals or the plugin, and the latest Jellyfin update broke it.
What I like about people complaining about Bluesky on Bluesky, is that they too are Bluesky, exhibiting the traits of Bluesky.
February 6, 2026 at 8:58 AM
What I like about people complaining about Bluesky on Bluesky, is that they too are Bluesky, exhibiting the traits of Bluesky.