nathanpeck.com
Developer Advocate focused on Generative AI at AWS.
311 posts
1,056 followers
489 following
Regular Contributor
Active Commenter
comment in response to
post
Heck yeah! Well that and us-west-2
comment in response to
post
Been exploring around north island but staying in West Harbor, Auckland right now!
comment in response to
post
Yep! Got a straight to residence visa application in progress!
comment in response to
post
Working on my visa for a permanent move!
comment in response to
post
5 year depreciation sounds like it should be extremely good for incumbents.
A startup might not even last five years to complete the depreciation, but to a large, profitable corp it’s just a minor delay that also holds back upstarts.
So I question whether this is really causing big tech layoffs
comment in response to
post
I’ve seen more rainbows in the last couple months here than in the last couple decades before. Unbelievably beautiful!
comment in response to
post
Yep absolutely. I find myself using fewer frameworks, and being pushed towards slightly older frameworks that have more training data in the LLM.
I wrote a piece about it when I first noticed this: nathanpeck.com/how-llms-of-...
comment in response to
post
Half of the equation is official MCP servers instead of bootleg third party ones.
The other half of the equation is properly scoped access credentials for each MCP and MCP tool call, but that’s pretty hard to solve in a world where people just use an admin IAM role for everything
comment in response to
post
The traffic from the Costco is already horrific. I shudder to think of how that side of Westgate will feel if all the empty lots are turned into shops as well.
The street network seems utterly broken in the area, both as a driver, and as a pedestrian
comment in response to
post
Because success for AI services is still all or nothing. You are either growing so fast that your primary worry is capacity, or you have zero inbound interest and your primary worry is iteration to find product market fit.
At this time it’s not possible to be “mid” enough to worry about renewals
comment in response to
post
Rather than blaming AI we should look at the deeper causes for widespread disengagement from thinking. A culture of minimum effort and mediocre output has swept through every corner of society and government. It was here before AI, and it won't end by blaming the most convenient latest villain.
comment in response to
post
I think the biggest misconception about AI is that AI makes things easier. In actuality, most of us who use AI find ourselves working on harder and harder things.
Of course there are people who disengage their minds, but the surface level take that "LLM makes you dumb" is quite frankly ridiculous.
comment in response to
post
Looking farther out in the future there is no reason model training or inference has to happen fast or in a dirty power region. There’s a reason every provider is building data centers in areas with abundant clean power, such as AWS working on a region in New Zealand (80% renewable energy sources)
comment in response to
post
As a bonus I also included stats for running a model like DeepSeek locally on my own MacBook. Not all models are large and heavy. You can actually get great results from tiny, quantized models that run on a laptop that sips power by comparison. Models are evolving faster than most people realize
comment in response to
post
I can’t share any specifics based on my insider knowledge of data centers, but this quick analysis using nothing but public info is decent. Models are way more efficient today than they once were and data center energy sources are mostly much cleaner than consumer energy: chatgpt.com/share/681fb6...
comment in response to
post
I swear there is an actual propaganda campaign about LLM impact on the environment. Every time I run the numbers my LLM usage is extremely tiny in impact. Even the sustained usage of playing a video game on my RTX 4080 for a few hours is consuming more than my periodic dips into LLM usage all day
comment in response to
post
There is a toxic attraction to struggle: late night crunch time code written using a mechanical keyboard and vi running on an obscure Linux distro.
But if a junior goes home at 5 after prompting Cursor in light mode on a basic Windows PC? That's not a bad thing as long as the code works
comment in response to
post
Okay I'll bite:
"Researchers found this penalty could be offset when AI was clearly useful for the assigned task. When using AI made sense for the job, the negative perceptions diminished significantly."
Not surprising. People who demonstrate effectiveness are valued, no matter what tools they use
comment in response to
post
Not bad at all. Three stages: install, build, and deploy. Typical of modern web development, and it should be nice and minimal in its final prod stage, despite the scary Dockerfile complexity. Upfront complexity here leads to minimal simple final state
comment in response to
post
My argument is that this is because the AWS customer isn't paying for raw storage and compute.
For example 1GB of S3 is not 1GB to AWS. The 1GB is replicated across at least three AZ's, then multiple pieces of hardware in each AZ.
What grinds my gears is comparing 1 GB to many GB's of replication
comment in response to
post
Are you safe with data replicated across multiple AZ's in multiple regions in multiple countries around the world?
Yep, hard to get much safer than that. It costs a lot more than a VPS, but if your business has worth to you, then it's worth paying to keep the data safe.
comment in response to
post
Are you safe with your data replicated across multiple data center availability zones in a single country?
Maybe? Let's hope that geopolitics stay calm and peaceful, despite the global rise of provocative authoritarian leaders.
Data centers are juicy economic damage targets in a hot war.
comment in response to
post
Are you safe with your data hosted on a single $5 VPS?
One random hardware failure, and you lost everything.
Are you safe with data replicated across multiple VPS in the same data center?
OVHcloud fire: your business just went up in smoke.
comment in response to
post
It represents a new appreciation for the joy of prototyping, which is something that often gets lost in the crush to produce perfect prod code as fast as possible.
I’d like to see more people create four or five rapid throwaway prototypes with “vibe coding” before reimplementing for production
comment in response to
post
Also worth noting that New Zealand playgrounds feature many play activities that only a nation with proper healthcare could have hahaha.
I was initially scared, until I realized kids are learning safety and confidence by doing seemingly “dangerous” climbing and balance activities
comment in response to
post
US: prison looking school building with the same sad plastic playground as every other school
New Zealand: three story playground with the tallest slide I’ve ever seen, trampoline, merry go round, water play area, large bus on springs that kids can pretend to drive in, and free gas grills to BBQ at
comment in response to
post
In my experience Terraform is fast but sloppy: resources marked as created and ready when they are actually still provisioning and not yet healthy.
Of course there’s something to be said for that optimistic approach and it’s speed, but I’m too pessimistic and suspicious personally haha
comment in response to
post
I work in software too, and you are exaggerating. What happened was a few centralized tools like GitHub changed their default, and because most of us don't really care one way or the other we just use the new default branch name. No one who is serious is out there saying its "unacceptable"
comment in response to
post
Next step is whether the US military chooses to uphold the first part of it's oath:
"I, _____, do solemnly swear (or affirm) that I will support and defend the Constitution of the United States against all enemies, foreign and domestic"
Military coup is all that is left after courts
comment in response to
post
As long as the central “broker” learning from all these millions of conversations is talking to diverse enough people and learning consensus from all of them, then it can arguably be far more accurate and up to date as an ethical guide compared to ancient religious books or nation state law systems
comment in response to
post
While I agree with you I would also say that Altman’s premise is an extremely basic take on moral / ethical relativism, which is nothing new. It’s a philosophy that dates back literally thousands of years, and is also the foundation for most of our good ethical systems today