Posts
5794
Following
355
Followers
556
.

Jarkko Sakkinen

Edited 1 month ago
And you can see what happens when you "just prompt". This shows especially in how so called agent frameworks are designed. It's pretty much the same as with any process for which you want to minimize the amount of interruptions and context switches. What these frameworks do to the inference is essentially some degrade of the overall "cognitic performance" and accumulated time blows out of the roof.

Not thinking yourself can be fucking expensive ;-)
1
0
0

Jarkko Sakkinen

I guess since Anthropic has AI philosopher, it is starting to trend also among "AI natives" at career sites :-) Some of Anthropic's shit is so weird that I enjoy it almost.

Keynote speaker is last season.
1
1
0
@HuguesRoss @ljs Yeah, and I mean if this "make it work as PoC" does not really time that much with a talented team.

Almost never AI cost savings are not weighted with customer satisfication metrics, competitive advantage reached by using LLMs for when writing apps and other software to some highly compted market such as e.g., iOS App Store. When I buy stuff I don't care about at all how expensive it was produce or manufacture. Nobody does.
0
0
1
@cesarpose If I had evil plot tho, I'd prefer one that wouild actually slave the humanity under my control :-) This ain't gonna be one that will tbh...

Let alone slaves, they even don't have an actual base of non-AI customers. There's just bunch of shitty companies selling AI dev tools to other shitty companies.
0
0
1

Jarkko Sakkinen

Edited 1 month ago
Further, neither Anthropic nor OpenAI have a product that has any growth in their value.

DeepSeek pretty much does the stuff that they do with way lower amount of teraflops, and it's open source. Majority of customer base can probably do 95% of tasks locally within only few years. Buying AI will become like buying a printer or something.

I would never buy their stocks because it does not compute to a cash flow to have a worthless business.a

And their brand suck too. There are crazy people of course who think they are "builders" but most people are just scared of them and think that they are fucking weird (like they should tbh).

Everybody loosing their income and dignitty is not a business model. Tech is tech and we choose as a non-technical decision how we apply it in the end.
0
0
2

Jarkko Sakkinen

Believing in magic creatures is the professional standard of 2026. I wonder when this AGI bullshit will stop. It's sad and ridiculous.
0
0
1
And that is from second piece of TypeScript code I've looked at in my life so not exactly seasoned in that craft. Immediate WTF moment.
0
0
0

Jarkko Sakkinen

Talking about contradictory actions. I'm using Pi to do some QA, some research for sandboxing agents etc. Lots of unrelated on writing code. Actually more like providing more focus on just writing code would be better description. I was also happy to see that QEMU has AI guidelines that match pretty much how I see the world right now.

With some of the plugins I've found i've UNVIBE coded the implementations because there's like really weirdly ineffective patterns used.

For pulling a full WASM runtime to be able to call WASM version of xxhash is awkward :-) And that async routine. I replaced it with stock library SHA-256 and no bullshit function call. I guess LLM had made these ridiculous choices based on algorithm completely ignoring the context.
1
0
0

Jarkko Sakkinen

I'm glad that this is moving forward:

https://lore.kernel.org/linux-integrity/20260515211410.31440-1-ross.philipson@gmail.com/

By all means and purposes it does much the same as conifdential computing for in-premises.

That is sort of high-level use case difference between confidential computing and D-RTM.

CoC makes sense when renting from cloud, or when "untrust" is part of the product (e.g., as in Signal which uses SGX for contact delivery).
0
0
0
@ljs Long before AI, while i was doing some stuff for Nokia, we sometimes had to make PoC's look worse so that some random manager would not think that the product is complete. And now it's like... well, you know how it is like.

Further, I've never been in a project, worked on an upstream open source project feature, or anything, where initial the building of "the thing" would have taken more than a few weeks. After that follows countless months of tuning, adjusting and most importantly discussing with other people about the nuts and bolts.

IMHO, the bubble in AI is really related the misconception that building some functional version of software would be the main contributing cost factor for shipping quality software.
4
22
40

Jarkko Sakkinen

I wonder is there specific reason for NVIDIA driver in Buildroot being so old?

I've made update to the latest.

#buildroot
0
0
0

Jarkko Sakkinen

weekly ed zitron delivery: https://www.youtube.com/watch?v=TCeXwFWmv1U
0
0
1

Jarkko Sakkinen

1. Pre-2026: open source turning into slop thanks to unnecessary Rust rewrites.
2. Post-2026: open source turning into slop thanks to unnecessary Rust AI rewrites.
0
0
0
@ParadeGrotesque i better hope that dario won't sue me :-)
0
0
1

Jarkko Sakkinen

Every week requires at least one Ed Zitron clip to keep one's head sane in this business :-)
0
0
1

Jarkko Sakkinen

Fedora's "AI maximization" plan is like if NAS was a feature of the desktop. Desktop and compute workloads are not compatible. Just from plain technical perspective I don't understand it TBH.
0
0
0

Jarkko Sakkinen

😅

spotted from linkedin
2
5
12

Jarkko Sakkinen

Edited 1 month ago
The build tool called 'just' is actually pretty nice for orchestrating more complex builds.

The feature that makes it nice are target parameters and parametrized dependencies. They give it suprisingly increased amount of robustness over GNU Make.

I use it together with Buildroot in couple of occasions.
0
0
0

Jarkko Sakkinen

Vim 9.2 :----) let's go...

The lightline theme called (drumroll) monochrome is the only theme I've ever made.
0
0
0
Show older