switchbak

If it's not enabled by default, "disabled" is the natural state, no?

Or is this more intrusive than I imagine?

You're in-between tiers, and thus don't exist. Please wait 4 days between washing and try again.

This kind of heuristic is actually extremely useful - and mostly what I try to develop over a couple weeks as the new models come out. I would love if we could get a fairly standardized take like this on the regular.

Of course we have far too many benchmarks (of questionable utility), but it would be helpful to get OpenAI's take on where it slots in, and how the landscape changes as these get rolled out.

I have found Fireworks to not be so stellar myself lately - I'm having a better time on BaseTen, but I only just started dabbling around there. Definitely the supposed speeds of Fireworks (that I never saw) are actually available there.

I listen to a niche green energy podcast, and the lady has such a creaky voice that I can barely stand it.

I think there's a generational judgement part of it, and some hearing damage that makes it harder to understand their words on the very low/creaky end.

Not to mention, it can cause real problems for your vocal chords. A little bit is entirely natural, and I don't really notice it in most men, but a lot of young(er) women seem to go overboard on it (in my opinion).

Lucky for us, the embedded folks don't take security too seriously! You don't need to be a garden variety hacker when your LLM can handle the drudgery for you.

But this is precisely it - when you combine some kind of fidelity with software development (so you can guide the LLM well), with some kind of depth of knowledge in a domain (probably a niche domain) you can do amazing things.

I'm currently experimenting with some custom hardware + software, composed in a way that seems novel, for a niche domain. This is with a new kid coming soon. There's just no way I would have crossed the activation threshold to do that without having LLMs provide assistance.

Wait, are you asking "Where's the disruption" when we have a couple folks vibe coding replacements for Adobe's whole stack, as well as things like Autodesk Fusion 360? Are you serious? Then you say things are being "vibe coded" away ... well, what is it? Sounds to be like the moats of the large corps are being dissolved, and that those who can ship working software - will.

"software is getting cheaper" - and easier to make, but it hasn't lifted all of the other constraints. Guess who's well positioned to understand what those constraints are and to work well in this new landscape? Answer: they didn't go to business school.

I use a terrible investment platform that purposefully nerfs the web app so you are forced to use the mobile app.

For something like a bank, I think they ought to be mandated to offer a functional website, yes. A slick app is nice too, but not if it’s the only option, specifically designed to steal my data and spy on me.

I'm absolutely going to hate talking to these things as customer service agents, with fake vocal fry and odd amounts of intonation and emotion peppered in.

For sure there's gonna be people falling in love with them though! I find all the anthropomorphization creepy though.

I've been playing with the fast mode on BaseTen - clearly a different cost bracket, but WOW - the speed is phenomenal! A real game changer to be able to stay in flow.

That definitely matches my experience, I have found Astra to be more capable than many other benchmarks show, have been very impressed with Argon for text, and generally like Sol. I find Haiku and Sonnet to be not worth my time as compared to the speed and cheapness of something like DeepSeek.

The only exception on the chart is DeepSeek - it scores exceptionally low there, which doesn't exactly track my experience. It usually does what you tell it to, and it's not exceptionally intelligent, but I do find it quite good at what it does (the bulk of code, scripting, etc). I know a lot of folks like it too ... curious if I'm just not exploring the simpler models enough?

I recently jammed all my stuff behind Tailscale. I sure hope that doesn't burn me.

But it's a lot better than having things live on the open internet!

Also looking at those old C++ services a whole lot differently now.

So many questions!

What exactly is an LLM sellout? Someone who sees the world moving toward it?

Wouldn't Rust be a sellout too?

Why is Scala an LLM sellout?

I pay API rates via OpenRouter / Fireworks, and I find it challenging to hit the $20 mark.

If I was to do anywhere near that kind of work on Anthropic, I'd find myself running out of tokens, waiting days to continue working, and leaning on their el-crap models to try to avoid running out again. This happens all the time.

This way I get to leverage DS for nearly all my work, benefit from its far superior latencies, and I still get to lean on OpenAPI / Anthropic or whatever SOTA models I want when I need them.

The only time that really falls down is if I'm doing huge amounts of real novel work, which hasn't been the case for a while.

I was doing exactly that some 27 years ago for my first internship. It has been a disappointment ever since seeing how little many other environments have prioritized tight feedback loops. Outside of dynamic stuff, mainly.

That line of reasoning ... everyone in this thread is now dumber for having listened to it.

Seriously - appeal to authority is a fallacy, unless it's an appeal to an authority that I happen to agree with? It's hard to take you seriously.

So then we shouldn’t be talking about AI either? No contentious topics for HN?

That has ALWAYS been a huge problem in our industry. But HR is the blessed role, and by God, we must not violate the org chart, no matter the cost. And oh the cost!

I do find I’m getting dumber in particular ways. Does that matter? Well probably - thinking logically is important, and writing code is a sure way to exercise that. I might just have a day a week I force myself to code something hard. But I can’t deny I like operating above all the drudgery.

And holy shit is AI ever a potential for harm when wielded by someone that doesn’t show sufficient restraint (ie: think). I think that’s my biggest problem the last year or so.

Unfortunately I know too well the kinds of crap involved (!) with those labour jobs. I’ll hold on until I have to jump ship. Perhaps a bad idea, but it’s where I’m at.

I dunno - the abstraction level is going up, that's all. You still have to know all the techy bits, if you don't you're going to have a hard time and get eclipsed by someone that does.

At least that's how it is today. But last year it was barely even worth using. Next year? Really hard to say, but it feels like it's plateauing to me. Depends on the day you ask though.

Will we be entirely replaced? Maybe. Or it'll get close enough to the "no code" wet dream of a few years back, and you'll have enough low cost LLM kiddies that the whole notion of the highly paid software engineer is just a dead end.

And then what? ... I kind of hate whatever this career is now anyway, I wouldn't mind doing something else, but I'm addicted to that sweet sweet cheese!

You can do what all our applicants did - just have AI answer all of our interview questions for you, and collect a couple paycheques until we finally fire you. Rinse and repeat.

I mean ... don't do that. But we have been BOMBARDED with this, literally 100% of our pipeline was fake until we rejigged it a bunch. I wonder how much of peoples challenges are just that the scammers are making the job hunting environment a market for lemons?

Thanks for underlining the last paragraph so eloquently.

The restraint, rigour and responsibility you showed prior to LLMs relates directly to how much of that you apply now that they’re here.

Given how rare it was previously, and how little it was cultivated by shortsighted management, I’m not surprised with all the slop we see today.

The truth is there’s nothing stopping someone from using these tools to remove dead code, simplify project structure, remove dependencies, etc. but I don’t see it happening very often. I have done it a bunch myself though so it’s absolutely possible, it’s all a matter of investment and prioritization.

There's always some anxiety around job security, that's nothing new. I've been around enough rounds of that to know it well. This isn't that.

We're talking about a fundamental disruption of knowledge work, to know that you might not even be able to pivot to something else. It's kind of like when people would callously tell the coal miners to "learn to code", except it's now "learn to be an electrician".

But I don't feel you're really acting in good faith here, with your negative stereotypes and derisive language.

My experience more closely matches yours, but I've also done the multiple rounds of painful interviews.

I do find that the latter can be explained by care and concern, and also by organization chaos and poor boundaries. It's hard to tell until you're in them, but I try to avoid the latter whenever I can.

As another commenter mentioned - I now respect companies that take their time in hiring. The last thing I want is to jump on with some company that will hire anyone. But there is an element of respect - if you're just duplicating questions and wasting my time? GTFO here.

I have a subscription at work, and still I find myself wishing I could use a fast Chinese model. Something wired up to really fast inference - that rapidity of feedback is a feature in itself.

4.1 Flash seems to be in that sweet spot of very decent, really fast and really cheap. Even omitting the cost, it’s still compelling for staying in flow.

That promo video … it’s so cringe it feels like satire. But it’s not.

It’s PERFECT!

I tried fireworks.ai, drawn in by their supposed blazing speed. Yawn. Mostly worse than vanilla Deepseek.

Is Lithos actually fast for common usage?

If it truly was "insanely difficult", we would have almost no products. Clearly that's not the case. Of course most products are highly imperfect, but here we are.

And not all code is the same. To a huge degree - most things are pretty straightforward, and the hard stuff is usually a minority of the whole.

Correct code might not be sufficiently tested, it might waste memory, it might do dumb things with the wrong data structures, it might be terribly factored, it might be commented in sanskrit - and yet it can still be "correct". Don't muddy the waters by bringing in all these other factors. Those are not correctness, even though it might be something that you (and I) strongly desire, or things that are necessary to ship with. Correctness is not "done".

And I agree that any code released into the wild should be as secure as you can make it. Though a scientist running a sim on their desktop might not be remotely concerned at how secure their fortran code is.