Kevin Burns
@burnskp.com
Tech generalist working in the HPC, AI, and Security space.
My other computer is a Cray.
#linux #kubernetes #devops #infosec #pinball
My other computer is a Cray.
#linux #kubernetes #devops #infosec #pinball
Pinned
Kevin Burns
@burnskp.com
· Jul 9
Hi! I’m a long time Linux user who’s always in search of the perfect Linux laptop. I’m currently a Qubes user for the automation and ease of segmentation.
I consider myself an expert generalist with a focus on Linux, Kubernetes, security, HPC, AI, and developer workflows.
#introduction
I consider myself an expert generalist with a focus on Linux, Kubernetes, security, HPC, AI, and developer workflows.
#introduction
AI is a good smokescreen/marketing platform used to hide the real reason companies seem to be shedding devs. There’s not enough ROI to keep them onboard. Most likely due to interest rate increases and section 174 tax code changes.
August 9, 2025 at 9:52 PM
AI is a good smokescreen/marketing platform used to hide the real reason companies seem to be shedding devs. There’s not enough ROI to keep them onboard. Most likely due to interest rate increases and section 174 tax code changes.
I feel that AI won’t be a big game changer in a majority of large orgs, at least not for awhile.
The places that AI currently impacts aren’t the areas that cause issues getting features out the door. Like agile implementations, it focuses on a narrow part of the pipeline.
The places that AI currently impacts aren’t the areas that cause issues getting features out the door. Like agile implementations, it focuses on a narrow part of the pipeline.
August 9, 2025 at 9:47 PM
I feel that AI won’t be a big game changer in a majority of large orgs, at least not for awhile.
The places that AI currently impacts aren’t the areas that cause issues getting features out the door. Like agile implementations, it focuses on a narrow part of the pipeline.
The places that AI currently impacts aren’t the areas that cause issues getting features out the door. Like agile implementations, it focuses on a narrow part of the pipeline.
GPT-5 gives me an excuse to get a paid account to test out codex. Guessing it’ll be the default model for copilot, which’ll actually let me use it at work.
August 8, 2025 at 1:52 PM
GPT-5 gives me an excuse to get a paid account to test out codex. Guessing it’ll be the default model for copilot, which’ll actually let me use it at work.
Congrats @biascilab.bsky.social and cDc!
August 8, 2025 at 1:46 PM
Congrats @biascilab.bsky.social and cDc!
Reposted by Kevin Burns
GPT-5 is so amazing it can make 52.8 higher than 69.1 on a bar chart
It can even make 69.1 and 30.8 THE SAME SIZE
It can even make 69.1 and 30.8 THE SAME SIZE
August 7, 2025 at 5:08 PM
GPT-5 is so amazing it can make 52.8 higher than 69.1 on a bar chart
It can even make 69.1 and 30.8 THE SAME SIZE
It can even make 69.1 and 30.8 THE SAME SIZE
Obsessing over growth numbers of a social platform encourages bot spam accounts. If higher number better, who cares what makes the number higher? Moderation and banning bots becomes something to be avoid when possible. Everyone gonna game those metrics when they have to.
August 7, 2025 at 8:18 PM
Obsessing over growth numbers of a social platform encourages bot spam accounts. If higher number better, who cares what makes the number higher? Moderation and banning bots becomes something to be avoid when possible. Everyone gonna game those metrics when they have to.
It wonder if it would be feasible to have a cxl card loaded with ram and use it to swap out layers in a gpu for moe usage. You couldn’t batch it, but it’d be interesting to see the time trade off. No idea if the tech even supports it.
August 7, 2025 at 6:06 PM
It wonder if it would be feasible to have a cxl card loaded with ram and use it to swap out layers in a gpu for moe usage. You couldn’t batch it, but it’d be interesting to see the time trade off. No idea if the tech even supports it.
Reposted by Kevin Burns
I've had preview access to GPT-5 for a couple of weeks, so I have a lot to say about it. Here's my first post, focusing just on core characteristics, pricing (it's VERY competitively priced) and interesting details from the GPT-5 system card simonwillison.net/2025/Aug/7/g...
GPT-5: Key characteristics, pricing and model card
I’ve had preview access to the new GPT-5 model family for the past two weeks, and have been using GPT-5 as my daily-driver. It’s my new favorite model. It’s still …
simonwillison.net
August 7, 2025 at 5:44 PM
I've had preview access to GPT-5 for a couple of weeks, so I have a lot to say about it. Here's my first post, focusing just on core characteristics, pricing (it's VERY competitively priced) and interesting details from the GPT-5 system card simonwillison.net/2025/Aug/7/g...
I’ve been eating gluten free for almost 16 years now and wouldn’t wish it on anyone. I can’t think of another diet that has as much processed food if you’re not careful and all you do is substitute gluten products in a normal American diet.
August 7, 2025 at 5:03 PM
I’ve been eating gluten free for almost 16 years now and wouldn’t wish it on anyone. I can’t think of another diet that has as much processed food if you’re not careful and all you do is substitute gluten products in a normal American diet.
Information wants to be free!
New from 404 Media: more than 130,000 Claude, Grok, ChatGPT, and other LLM chats are readable on Archive.org. It's similar to the Google indexing issue, but shows it impacts many more LLMs than just ChatGPT. Some chats contain API keys.
www.404media.co/more-than-13...
www.404media.co/more-than-13...
More than 130,000 Claude, Grok, ChatGPT, and Other LLM Chats Readable on Archive.org
The issue of publicly saving shared LLM chats is bigger than just Google.
www.404media.co
August 7, 2025 at 4:56 PM
Information wants to be free!
I feel like a lot of the arguments about LLM reasoning are really arguments around the definition of the word reason. It seems to be an overloaded term in LLMs which is essentially “talk this through before answering”. However, most people feel it’s this magical thing brains do.
August 7, 2025 at 4:43 PM
I feel like a lot of the arguments about LLM reasoning are really arguments around the definition of the word reason. It seems to be an overloaded term in LLMs which is essentially “talk this through before answering”. However, most people feel it’s this magical thing brains do.
Taylorism and the alignment of LLM incentive structures.
In psychology we try to figure out how people actually tick. With LLMs we get to figure out how much the data on incorrect assumptions of human psychology impacted the LLMs training and token generation.
In psychology we try to figure out how people actually tick. With LLMs we get to figure out how much the data on incorrect assumptions of human psychology impacted the LLMs training and token generation.
August 7, 2025 at 4:41 PM
Taylorism and the alignment of LLM incentive structures.
In psychology we try to figure out how people actually tick. With LLMs we get to figure out how much the data on incorrect assumptions of human psychology impacted the LLMs training and token generation.
In psychology we try to figure out how people actually tick. With LLMs we get to figure out how much the data on incorrect assumptions of human psychology impacted the LLMs training and token generation.
Poking around slack channels and I found one for the ultimate question.
#is-devops-jira
I think it is.
#is-devops-jira
I think it is.
August 4, 2025 at 8:26 PM
Poking around slack channels and I found one for the ultimate question.
#is-devops-jira
I think it is.
#is-devops-jira
I think it is.
One nice thing about getting moved to a new team is running a script and unassigned all these Jiras on my backlog
July 31, 2025 at 7:44 PM
One nice thing about getting moved to a new team is running a script and unassigned all these Jiras on my backlog
The idea of an upgradable laptop like the Framework 13 is great on paper, but leaves a bit to be desired. For instance, it pretty much shackles the main board to the first cooling solution they ever shipped.
July 31, 2025 at 7:37 PM
The idea of an upgradable laptop like the Framework 13 is great on paper, but leaves a bit to be desired. For instance, it pretty much shackles the main board to the first cooling solution they ever shipped.
I’ve spent the last 3 years trying out dozens of laptops to find the perfect one. Latest one I’m testing out is the HP Elitebook X g1i. Currently on sale at Costco. Besides the screen quality/light bleed issues it may be the first to hit my requirements in temp and battery life.
why are there still no good laptops except macbooks
July 31, 2025 at 7:36 PM
I’ve spent the last 3 years trying out dozens of laptops to find the perfect one. Latest one I’m testing out is the HP Elitebook X g1i. Currently on sale at Costco. Besides the screen quality/light bleed issues it may be the first to hit my requirements in temp and battery life.
wlroots uses lcms2 as the back end and I may have better luck with a icc version 2.4 than the 2.2. However, I’m not aware of a way to generate one in Linux. AFAIK I have to reinstall windows to create one.
July 31, 2025 at 7:29 PM
wlroots uses lcms2 as the back end and I may have better luck with a icc version 2.4 than the 2.2. However, I’m not aware of a way to generate one in Linux. AFAIK I have to reinstall windows to create one.
I’ve been messing with sway for way too many hours trying to get color profiles to work properly. Turns out they partially worked, but against dark color. Didn’t notice because I use a light background. Now to figure out if I can create an icc that works.
July 31, 2025 at 7:27 PM
I’ve been messing with sway for way too many hours trying to get color profiles to work properly. Turns out they partially worked, but against dark color. Didn’t notice because I use a light background. Now to figure out if I can create an icc that works.
Reposted by Kevin Burns
The longer conversation on measuring the impact of AI on software engineering:
• YouTube: www.youtube.com/watch?v=xHHl...
• Spotify: open.spotify.com/episode/4hDo...
• Apple: podcasts.apple.com/us/podcast/m...
• Summary and transcript: newsletter.pragmaticengineer.com/p/measuring-...
• YouTube: www.youtube.com/watch?v=xHHl...
• Spotify: open.spotify.com/episode/4hDo...
• Apple: podcasts.apple.com/us/podcast/m...
• Summary and transcript: newsletter.pragmaticengineer.com/p/measuring-...
Measuring the impact of AI on software engineering – with Laura Tacho
YouTube video by The Pragmatic Engineer
www.youtube.com
July 27, 2025 at 3:06 AM
The longer conversation on measuring the impact of AI on software engineering:
• YouTube: www.youtube.com/watch?v=xHHl...
• Spotify: open.spotify.com/episode/4hDo...
• Apple: podcasts.apple.com/us/podcast/m...
• Summary and transcript: newsletter.pragmaticengineer.com/p/measuring-...
• YouTube: www.youtube.com/watch?v=xHHl...
• Spotify: open.spotify.com/episode/4hDo...
• Apple: podcasts.apple.com/us/podcast/m...
• Summary and transcript: newsletter.pragmaticengineer.com/p/measuring-...
SRE is also a fun one, ranging from senior/staff/principal engineer level experience in developing and operating systems to offshored level 1 and 2 support.
July 24, 2025 at 6:18 PM
SRE is also a fun one, ranging from senior/staff/principal engineer level experience in developing and operating systems to offshored level 1 and 2 support.
I can talk about where DevOps came from and what it was about for way longer than people would be willing to listen. However, the majority of companies I’ve seen equate DevOps to the “stuff the devs don’t want to do/is seen as too low value”. QA, Builds, operations, etc. whatever is seen as a cost.
July 24, 2025 at 6:16 PM
I can talk about where DevOps came from and what it was about for way longer than people would be willing to listen. However, the majority of companies I’ve seen equate DevOps to the “stuff the devs don’t want to do/is seen as too low value”. QA, Builds, operations, etc. whatever is seen as a cost.
One of my favorite bureaucratic org patterns is providing the illusion of change by appointing the person most opposed to the change to be the head of said change. The cloud architect that hates anything cloud. The head of DevOps who loves ticketing systems and siloing the workforce.
July 24, 2025 at 6:10 PM
One of my favorite bureaucratic org patterns is providing the illusion of change by appointing the person most opposed to the change to be the head of said change. The cloud architect that hates anything cloud. The head of DevOps who loves ticketing systems and siloing the workforce.
My favorite operations pattern is IaC as shell scripts. This is where you take a tool like Puppet or Ansible and use it to run non idempotent bash scripts and hope it doesn’t break. I’ve worked with people who have a year+ experience with these tools and legit thought that was how they were used.
July 24, 2025 at 6:04 PM
My favorite operations pattern is IaC as shell scripts. This is where you take a tool like Puppet or Ansible and use it to run non idempotent bash scripts and hope it doesn’t break. I’ve worked with people who have a year+ experience with these tools and legit thought that was how they were used.
Honestly with the incentive structure I’m not sure why you would turn on the stuff. If your choice is between spending money and resources on an outcome that only has downsides why wouldn’t you choose to keep your head in the sand? It’s the sane choice from a cost/benefit analysis.
July 24, 2025 at 5:24 PM
Honestly with the incentive structure I’m not sure why you would turn on the stuff. If your choice is between spending money and resources on an outcome that only has downsides why wouldn’t you choose to keep your head in the sand? It’s the sane choice from a cost/benefit analysis.
It’s cool. It’s totally fine. We have no proof they did anything besides hack it and go home. What? No, we don’t log or monitor or have the ability to show proof. That would be unethical and privacy invasive… and provide a way to establish fault. Which is totally not why we turn all that off.
New: DHS confirms that it was hacked through the SharePoint vulnerability but says "there is no evidence of data exfiltration at DHS or any of its components at this time." At the same time, it notes that "the investigation to identify potential exposure remains ongoing."
July 24, 2025 at 5:22 PM
It’s cool. It’s totally fine. We have no proof they did anything besides hack it and go home. What? No, we don’t log or monitor or have the ability to show proof. That would be unethical and privacy invasive… and provide a way to establish fault. Which is totally not why we turn all that off.