HNHacker News
TopNewBestAskShowJobs

Twirrim

11,673 karma · joined July 28, 2010

submissionscomments
Twirrim··on OpenAI withdraws three mathematical results
There's this amazing rhetorical technique called hyperbole. Compared to what I'm getting from co-workers, and in code bases I'm subjected to these days, what I used to have to deal with was a masterpiece.

I've waded through unfamiliar code at 3am trying to figure out just why the hell things broke this time more times than I can care to remember, particularly trying to tease apart the ways that code has organically grown compared to the original intent. If I'm subjected to one more piece of "spooky at a distance" injected behaviour code I'll probably scream loud enough to be heard half way across the country.

It's still just leaps and bounds more readable than what people are slopping together (I do have some co-workers that have been extremely tightly focused on avoiding "slop" with their AI and doing a lot of tuning, and it's certainly preferable to the ones that haven't)

Twirrim··on OpenAI withdraws three mathematical results
It's really depressing watching brilliant software developers and engineers producing the worst, unmaintainable code imaginable, and being okay with shipping it because it's passing the test suite. Assuming the test suite is any good anyway (because holy crap, the nonsensical tests AI writes...)

Your code used to be a masterpiece, so well crafted it's easy for AI to tweak and modify because you've got everything so logically organised and scoped... and now this is what you're producing? Hard to debug monstrosities that only an LLM can realistically bolt new features or tweaks onto, because it can do the kinds of refactoring necessary each time.

We use to talk about the fact that code should be readable because you spend more time reading it than writing it, but I think that misses the key part that readable code is also typically easier to debug. If you can read and understand the code, you can follow the logic when things are wrong in production, and you can more easily reason about the emergent properties of interactions between the complex systems that are involved.

Twirrim··on Clef: Open-weight decision models, and new RL fine-tuning platform
Agreed, could be something as dumb as a simple javascript snippet running in the browser. Would be nuts to use something like an LLM just for that.
Twirrim··on Clef: Open-weight decision models, and new RL fine-tuning platform
We already have examples of LLMs running 16k+ tokens a second using custom ASICs.

It's down at the moment (Not sure if it'll return?) but Chat Jimmy[0] produced by Taalas[1] was powered by an ASIC running Llama 3.1 8B, and hitting 17,000 tokens/sec. It was amazing to use, you'd no sooner have hit enter than you had a full response back. I actually found its speed to be a problem for interactive stuff, as every answer got several paragraphs I'd then wade through, vs a model populating text closer to my reading speed.

I appreciate, there are differences between an 8 billion parameter model and something GPT-4-ish, but we're currently in the middle of a race between a half dozen or so companies to produce the next best frontier model, which requires their infrastructure to be dynamic.

We really don't always need newer better faster stronger models, there's quite a lot of room for "good enough" where getting 17kt/s at significantly lower power would be amazing.

[0] https://chatjimmy.ai/ [1] https://taalas.com/

Twirrim··on Nissan's third generation e-POWER powertrain
The volt was never popular enough, unfortunately. Right around the time we were looking to buy a new car, and had the Volt high on our list, the retired the product line.
Twirrim··on What Sun got wrong
One of the things that Larry Ellison found utterly baffling when he acquired Sun was the way that sales commission was calculated. It was all on the value of the contract, with no requirement for the contract to be profitable.

At least in the later stages of life Sun sales folks were literally selling the company into bankruptcy. They'd find out what competitors were offering, undercut it to the degree they'd make a loss, just to get the contract and their sweet commission.

To not realise you've got this going on really speaks to how little they were interested in the actual running of the business.

Twirrim··on AWS says it can't restore some data from mideast facilities struck by Iran
Long time no see.

2013-2014 would be about right, they were still in SEA when I joined in 2013 but were actively working on decommissioning it.

Twirrim··on AWS says it can't restore some data from mideast facilities struck by Iran
That's absolutely something you can configure your S3 bucket to do if you want (I have one of mine replicating elsewhere).

The amount of data S3 stores "automatically replicating" to other geographical locations would make things prohibitively expensive, especially when you consider the daily delta, and how much of that is ephemeral or frequently mutated data that is stored. The bandwidth costs alone would be eye-watering, let alone the storage costs.

S3 cannot make any automated decision about whether data is, or isn't important, and if they did they'd only open themselves up to lawsuits if they guessed wrong. That's why it's made an option for the end user to enable replication if they want to, or choose to replicate their own data.

Twirrim··on People who can't picture anything are rewriting the science of imagination
We don't... at least not in our minds anyway.
Twirrim··on Operation Smart Kettle – Börzels Blog
I miss the days when my tech feeds used to routinely include these style of "Wonder what's going on here?", and "I did sometimes ridiculous with <something I'd never expect to have complicated electronics>" posts.

Feels all too often like it has been replaced with "We switched from $A to $B, the reasons will shock you!" (Followed inevitably by "We switched from $B to $C").

Twirrim··on LRU is harder to beat than the KV-cache papers suggest
It'll just be telling you what the data it ingested claims. Not what is necessarily true. It's as subject to garbage in, garbage out as anything, and you don't know what it's actually trained on.
Twirrim··on NTSB issues investigative update on B-767 runway excursion accident in Miami
The NTSB report will detail what can be done, and they will have full access to every bit of information involved in the incident.

They have a well deserved reputation for the quality and accuracy of their incident reports, and they don't pull their punches.

Twirrim··on NTSB issues investigative update on B-767 runway excursion accident in Miami
The branding on the plane is Prime Air, but it's not actually Amazon owned and run. Instead it's a cargo airline (21 Air) that they contract with to operate cargo flights with Prime Air livery on the planes.

It'll be interesting to see the outcome of the NTSB investigation, they're usually quite thorough.

Twirrim··on Neki – Sharded Postgres
It's very effective, I'm certainly persuaded to never touch their products.

A remarkable own goal given that if they'd literally held their tongue, I'd be at worst indifferent.

Twirrim··on I have a theory that software drives people insane
I've had more than one half-serious conversation with SVPs here that every staff/architect level engineer should have to spend at least 6 months working on the engineering team that directly interacts with the largest customers.

I flat out guarantee they'll come away with a wildly different perspective of the things that need done, and the ways their service operates.

Ideally I'd want them to do a rotation in customer support so they get the small customer perspective too, and at least learn what kind of paper cuts they're inflicting on customers.

Twirrim··on JEP 544: Ahead-of-Time Code Compilation
The slight drawback is that AOT compiled Java code's performance isn't as good as the JIT compiled form running on the JVM.

There's a reason why folks keep coming back to JIT compilation. It's hard to beat the value that can be gained from actually gathering data on how the code is actually used, which leads to a whole set of potential optimisations (which is the problem Profile Guided Optimisation attempts to solve for AOT compiled code.) The JVM is arguably one of the most advanced and capable JITing runtimes.

As with anything it's a trade off. Fast start times, pretty fast running (I think maybe less memory usage?); vs the full JIT speed you can get on the JVM at the cost of start-up speed and memory consumption. All depends on what you want to use the application for.

If you're talking something like a serverless function, go native. If you're talking production server where you're measuring runtime in more than dozens of minutes, probably better to stick to the JVM & JIT.

Twirrim··on No Man's Sky Cosmos
I played it about a year ago for the first significant time after having bought it some point a long time back.

It's... fine? It kept me engaged until it didn't. Part of the "didn't" came about because I was being paper-cut to death by bugs that googling showed me had been reported for a long time, if not since the very beginning.

I kept feeling like it was a game on the verge of happening throughout, and it never quite made it. Lots of things to go do to serve.. no purpose. Nothing was that unique or memorable, and I often had very little clue where I was in relation to anything. At one stage I was overseeing a town but that seemed to do nothing for me other than make me go back and make a snap decision for no discernible reason, occasionally fight hostile creatures, and also act as a sink for resources I was gathering.

I kept thinking "Why am I even doing this" until it really sunk in "Why am I even doing this?" and quit.

It also exasperated me that even though I was using a pretty fast gaming system, on a regular basis assets would take ages to appear, including buildings. "Fly to this place on the planet, land at this archaeological dig, wait 3 minutes for the dig to bother to appear before you can do anything." (saw identical behaviour on my steamdeck too, so doesn't seem "my machine" per se?)

Twirrim··on Bill Gates tries to install MovieMaker (2003)
There was nothing surprising in that email chain at all, looking at it as someone with lots of experience in large companies.

In a twisted kind of way, the replies were reasonable and logical. They're not dumb, they can see it was crap. No one org can actually own this fix, so in that kind of situation it realistically requires someone like Bill to choose someone to drive it, and demand updates to ensure that everyone gets aligned. Telling his execs, especially in a company so notoriously political and "empire" oriented as it was, isn't going to work.

I actually liked that in Amazon, Jeff Bezos used to send out the infamous "question mark" emails on such a regular basis that they built an exec commitee specifically for the purposes of driving direct action on problems. Didn't matter where you were, or what you were working on, if that question mark email came you dropped everything to work on it.

Twirrim··on We could save petabytes of cache storage with Zstandard and Pingora
I'd be curious whether block level de-duping would add value too in their case. You effectively achieved that to some degree with storing the deltas.
Twirrim··on A CVE Dispute
> If one person can spend 2 days fighting to prevent 1h of useless work for all of the people running curl, that person has saved the software industry millions of dollars worth of developer time.

Working in a large company, that used to be one of the ways I'd measure my annual business impact. "Look, through my reviews I reduced the work of 70 tasks that were going to go out to most engineering teams for an average reduction from 2 hours to 30 minutes. With 300 teams in scope, that saved us approximately of 31,500 hours of headcount, the equivalent of 15 full time employees"

One thing that has proven remarkably effective is making SVPs sign off on tasks that impact more than a certain number of teams, and/or take more than a certain length of time. Yet to see a single one argue the point when they see the task being asked.

Twirrim··on A CVE Dispute
> Lots of teams that are supposed to be in charge of security don't ask "does this CVE affect us", but simply shift the burden of patching downward and outward

Unfortunately, I've had many a frustrating session with compliance auditors who do not care that it couldn't affect you, you're required to meet the PCI-DSS deadlines of 90 days for low, etc, and I've seen security groups force to accept that this is what they have to do, and then get the unfortunate task of dealing with engineers pissed off that them for forcing them to what security is being forced to do. Everyone's a whole big bundle of happy joy joy, all because of some stupid auditor following a stupid (but well intentioned) checklist.

Twirrim··on Flyway for freeloaders, rollback implemented in Python
>... A single command to roll back applied migrations with Flyway Community Edition, no license, no undo command, no crying in the club.

Oh here we go...

> ... Here’s how it works, buckle up girls.

sigh. Yeah, no.

Twirrim··on Apple introduces M6 and M5 Ultra
I'm always a little cautious about "tried it in the store" experience. If I nuke and rebuild my machine (either Linux, or Windows), it's always significantly faster than it was before. Over time stuff just tends to accumulate that you don't notice. Like the old days of needing to defrag hard disks.
Twirrim··on An Engineer's Old Cooking Trick Is Going Viral, Divides the Internet
We occasionally (maybe once a year) use a meal service for a couple of weeks, especially if it's a busy week.

The recipes sheets they provide drive me absolutely nuts working from them, to the point where I feel like I have to re-write them just to get them into a logical order. They're done entirely differently from any recipe book I've ever worked from. For example I constantly find I'm missing steps from how they're mushed together under photos, so I get later through the recipe and it talks about using something I supposedly did something with earlier and.. crap, I have to go back up through the recipe sheet trying to find what I missed the first time through. I've never experienced this with any other recipe book I've read.

Twirrim··on Linux 7.3 improves performance when running out of vRAM
Right, and when it gets filled, the OOM killer comes along and does its thing.
Twirrim··on Ask HN: I've quit six systems for tracking my illness. What works?
Where? I don't see a single one from clicking around a bit on the site.

The landing page doesn't even say what it is, just some audio thing somehow tied to Dr Oz (can't actually exercise it, because I run firefox). Why is the first thing you're expecting me to do is talk to my browser?

What if I'm deaf, mute, or blind and using accessibility tools? What am I even supposed to be talking about?

In the menu, projects doesn't do anything (and what's a project anyway in this context?)

There's 94 programs... for what? If I go open a program, https://medicare.dev/program/71152-general-enrollment-period... there's still no disclaimer, and still no explanation about the reason someone should be using the site. I guess this is to build.. AI something? Maybe for use in medicare.gov? Is this entire site meant to be used by folks working for medicare (and if so, should I even be able to see this much?)

Messages takes me to a login screen with no sign-up.

Forms and Invest take me to the same page, giving me a chance to invest in foreign markets(?), but not what the investments are in? (codify.city seems like it's created by you as well, similarly says sweet nothing about what purpose it's serving)

"How it works"... doesn't? "The walkthrough isn’t available yet."

"Wiki" seems to be providing some examples of what seems to be instructions to AI agents, but no indication how you'd even use them. (also if you scroll the box, there's an error "Syntax error in text mermaid version 11.16.0"

This whole site fails to nail the very basics, and seems like slop-coded nonsense. I'd encourage you to stop building features, and start doing the boring groundwork. If you're actually expecting anyone to use it, explain it.

Also yet to see any of the disclaimers you claim are all over the place. They're really not.

Twirrim··on AI;DR (AI; Didn't Read)
I've been wondering lately if this is about the end of the road for me for a career in tech (I've had a good long one, so.. maybe?).

I'm so tired of AI Slopped architecture documents, code, messages in slack, emails etc. I'm close to flat out rejecting any architecture doc I get that's written by an LLM, because I'm not sure the engineers even understand what they're proposing. The code is often a nightmare under the hood, even if it works. I've got smart coworkers who now just copy paste everything via an LLM giving themselves something approaching a lobotomy in the process.

Twirrim··on AI-Generated GitHub Copilot “Autofix” Allowed Compromise of Snowflake's Jira
You can't rely on people spotting the significance of such changes
Twirrim··on Stop sending me huge PRs; a rant
I've had _some_ success by asking Codex write a full plan, broken up in to logical phases, complete with git commit points along the way.

It's probably not quite how I would approach doing git commits, but they're at least logical boundaries, and make narrative sense for a reviewer.

Twirrim··on RISC-V: They Should Have Known Better
> So you write software for a platform you know nothing about?

That's how a sizeable chunk of software is written and shipped.

Runtime detection of CPU features is very much a thing, and is in fact used extensively in software you use or interact with every single day.

Just as a quick example, OpenSSL's approach for x86_64 is OPENSSL_ia32cap

https://docs.openssl.org/master/man3/OPENSSL_ia32cap/

This ensures (in theory, at least) that even if you're using your linux distribution's openssl library which is more generically targeted, you will get optimal/native runtime performance for your actual CPU.

← PreviousPage 2 of 34Next →