Jeremy Blackburn
@mrjimmyblack.com
Associate Professor of Computer Science and Director of the Institute for AI and Society at Binghamton University. I work on building a quantitative understanding of assholes on the Web @idrama.science. 4 kids, 2 wolfhounds, 2 Löwchens.
And just like that, the code works and we be steering. Revelation!
November 10, 2025 at 10:26 AM
And just like that, the code works and we be steering. Revelation!
Been working on some LLM inference time experimentation using mistral-rs and candle (vllm and PyTorch equivalents).
1. Compile times are pretty annoying, but not that terrible (~10 minutes for a release build).
2. Type system helps understand wtf is going on.
3. My experiments hard failing,
1. Compile times are pretty annoying, but not that terrible (~10 minutes for a release build).
2. Type system helps understand wtf is going on.
3. My experiments hard failing,
November 9, 2025 at 1:38 PM
Been working on some LLM inference time experimentation using mistral-rs and candle (vllm and PyTorch equivalents).
1. Compile times are pretty annoying, but not that terrible (~10 minutes for a release build).
2. Type system helps understand wtf is going on.
3. My experiments hard failing,
1. Compile times are pretty annoying, but not that terrible (~10 minutes for a release build).
2. Type system helps understand wtf is going on.
3. My experiments hard failing,
I was very wrong about code assistants not having much utility without a traditional CS background. www.reddit.com/r/MyBoyfrien...
[Mature Content] From the MyBoyfriendIsAI community on Reddit: Love in Claude Code
Posted by starlingmage - 23 votes and 6 comments
www.reddit.com
November 7, 2025 at 1:13 PM
I was very wrong about code assistants not having much utility without a traditional CS background. www.reddit.com/r/MyBoyfrien...
Without something like a traditional CS background, I don't see how AI code assistants are going to be particularly useful. This is Nth time I've had to push claude in the right direction wrt performance.
November 4, 2025 at 1:08 PM
Without something like a traditional CS background, I don't see how AI code assistants are going to be particularly useful. This is Nth time I've had to push claude in the right direction wrt performance.
We have hit the need for a "sophisticated approach" that is related to the only (required) unsafe code I've written (in Rust of course; I've written plenty of unsafe code in C).
November 2, 2025 at 3:03 PM
We have hit the need for a "sophisticated approach" that is related to the only (required) unsafe code I've written (in Rust of course; I've written plenty of unsafe code in C).
Now we're cooking with fire.
November 2, 2025 at 2:32 PM
Now we're cooking with fire.
I didn't realize that progress bars and live stats were so out of distribution, but claude has not figured this out in several days and I got tired of watching it fail.
November 2, 2025 at 2:07 PM
I didn't realize that progress bars and live stats were so out of distribution, but claude has not figured this out in several days and I got tired of watching it fail.
Hrm. Claude """"""""""sees"""""""""" me.
November 1, 2025 at 6:48 PM
Hrm. Claude """"""""""sees"""""""""" me.
One of the killer app type of features of code assistants is that they have the context of your code base. Yet, I find myself having to forcibly "re-educate" when existing code base already has a solution for the task at hand. Not good news for clueless human users.
November 1, 2025 at 5:59 PM
One of the killer app type of features of code assistants is that they have the context of your code base. Yet, I find myself having to forcibly "re-educate" when existing code base already has a solution for the task at hand. Not good news for clueless human users.
Trying to use claude with a more out of distribution project starting with a code base that makes pretty heavy use of concurrency to maximize throughput (both wrt async code and blocking code). I haven't touched it in a year or so. I've caught several poor architectural choices claude made so far.
October 30, 2025 at 12:02 PM
Trying to use claude with a more out of distribution project starting with a code base that makes pretty heavy use of concurrency to maximize throughput (both wrt async code and blocking code). I haven't touched it in a year or so. I've caught several poor architectural choices claude made so far.
Saw vid saying the RS is only true superhatch. Said only downside is motor will explode because of "direct injection". 1) Biggest issue is RS's seat position and "you feel every bump" suspension. 2) RS was only 2.3L ecoboost with problem: wrong head gasket from factory and ez fix. 3) I'm so mad.
October 28, 2025 at 9:40 AM
Saw vid saying the RS is only true superhatch. Said only downside is motor will explode because of "direct injection". 1) Biggest issue is RS's seat position and "you feel every bump" suspension. 2) RS was only 2.3L ecoboost with problem: wrong head gasket from factory and ez fix. 3) I'm so mad.
My wife just said “load it for my like if I was a worm” and I for real contemplated a hard reboot.
October 26, 2025 at 10:31 PM
My wife just said “load it for my like if I was a worm” and I for real contemplated a hard reboot.
Almost time for winter tires to go on.
October 26, 2025 at 6:14 PM
Almost time for winter tires to go on.
Is Danica Patrick really the only woman they could find to be on the F1 production this weekend? She believes that lizard people are real. Do her mediocre racing career and transition to bikini model in GoDaddy ads really justify platforming her at this level of motorsports?
October 18, 2025 at 10:20 AM
Is Danica Patrick really the only woman they could find to be on the F1 production this weekend? She believes that lizard people are real. Do her mediocre racing career and transition to bikini model in GoDaddy ads really justify platforming her at this level of motorsports?
wtf is the point of this? www.cbsnews.com/news/militar...
U.S. military strikes another boat in Caribbean — and there are survivors, official says
The U.S. military struck a sixth boat in the Caribbean Sea on Thursday, and there were survivors aboard the vessel, a U.S. official told CBS News.
www.cbsnews.com
October 17, 2025 at 3:53 PM
wtf is the point of this? www.cbsnews.com/news/militar...
Reposted by Jeremy Blackburn
Across the country, airports are refusing to air a video from the Homeland Security chief blaming Democrats for the shutdown. trib.al/tw3Hd3X
Airports Are Pushing Back Against Kristi Noem’s Shutdown Propaganda
Across the country, airports are refusing to air a video from the Homeland Security chief blaming Democrats for the shutdown.
trib.al
October 13, 2025 at 8:38 PM
Across the country, airports are refusing to air a video from the Homeland Security chief blaming Democrats for the shutdown. trib.al/tw3Hd3X
I definitely have plenty of old and tattered T-shirts, but it is weird to me how some people have one specific outfit that they think is special and wear to every event for the past 10+ years. I wonder if they realize how obvious (and sad) it is?
October 13, 2025 at 8:08 AM
I definitely have plenty of old and tattered T-shirts, but it is weird to me how some people have one specific outfit that they think is special and wear to every event for the past 10+ years. I wonder if they realize how obvious (and sad) it is?
It’s amusing to see people that clearly have no technical understanding of how LLMs work pontificate on how they work. They’ve clearly picked up some of the appropriate language/concepts, but it’s uncanny valley level stuff where their use is off in subtle ways that expose that they have no clue.
October 12, 2025 at 11:57 AM
It’s amusing to see people that clearly have no technical understanding of how LLMs work pontificate on how they work. They’ve clearly picked up some of the appropriate language/concepts, but it’s uncanny valley level stuff where their use is off in subtle ways that expose that they have no clue.
Someone made the mistake of posting a Hasan meme in the group chat and wasn't really familiar with my game...
October 10, 2025 at 1:35 PM
Someone made the mistake of posting a Hasan meme in the group chat and wasn't really familiar with my game...
Posted without additional commentary. www.theguardian.com/technology/2...
4chan unlikely to be included in Australia’s under-16s social media ban, eSafety commissioner says
Julie Inman Grant says message board will need to comply with other codes which will also include age assurance for sites hosting violent and adult content
www.theguardian.com
October 9, 2025 at 11:24 PM
Posted without additional commentary. www.theguardian.com/technology/2...
Took a quick look at the code. Initial thoughts: 1) looks like claude rolled its own auth system instead of using something like axum-login. Not sure which is more verbose. 2) It wrote a bunch of config code that reads env vars, etc. Would have been fewer LOC with clap I think (bc of macros).
unsupervised AI coding is wildly verbose, it will almost always choose to bolt something on over taking a step back and fixing the roots of issues.
October 9, 2025 at 9:41 PM
Took a quick look at the code. Initial thoughts: 1) looks like claude rolled its own auth system instead of using something like axum-login. Not sure which is more verbose. 2) It wrote a bunch of config code that reads env vars, etc. Would have been fewer LOC with clap I think (bc of macros).
Got Claude Pro plan ($200/year). Pro plan seems to be limited to Sonnet instead of Opus. Initial generated code DID NOT COMPILE. Fixes were in rust error messages (as is typical) and after about 2h20m or so app worked. Try to gen docker files for deployment: session limit reached; resets in 2h40m
Try Claude Code or OpenAI codex, they are far better with Rust
October 9, 2025 at 12:31 PM
Got Claude Pro plan ($200/year). Pro plan seems to be limited to Sonnet instead of Opus. Initial generated code DID NOT COMPILE. Fixes were in rust error messages (as is typical) and after about 2h20m or so app worked. Try to gen docker files for deployment: session limit reached; resets in 2h40m
Kinda funny that even some of my 100% NOT normie friends have so far managed to avoid any meaningful knowledge of Hasan. If only I could excise the part of my brain that makes internet drama a special interest.
October 9, 2025 at 8:58 AM
Kinda funny that even some of my 100% NOT normie friends have so far managed to avoid any meaningful knowledge of Hasan. If only I could excise the part of my brain that makes internet drama a special interest.
Not gunna lie: passkeys are pretty awesome.
October 9, 2025 at 8:56 AM
Not gunna lie: passkeys are pretty awesome.
Tried to use Gemini 2.5 pro to do some vibe coding for a rust (axum) back end for a web app. Biggest AI fail I’ve seen yet. Code is unusable; won’t even compile. It does insanely complicated things, implements wrong trait methods, somehow manages to create dependency hell, etc. Really really bad.
October 7, 2025 at 12:55 PM
Tried to use Gemini 2.5 pro to do some vibe coding for a rust (axum) back end for a web app. Biggest AI fail I’ve seen yet. Code is unusable; won’t even compile. It does insanely complicated things, implements wrong trait methods, somehow manages to create dependency hell, etc. Really really bad.