peab

Look at this one: https://x.com/Zahri_10i/status/2078211494349537447?s=20

I've seen a few that have been shared - I'll link them here if I can find them.

I think right now most people here and on twitter have no taste, and post slop that were just sort of one shotted.

But there are people using these tools in a sort of hybrid way, which I think has incredible potential. You can use it to do CGI on existing footage, in a way that's orders of magnitudes faster and cheaper than current CGI methods.

I also think that the most viable models will be video-to-video, and audio-to-audio.

If you've read A Young Lady's Illustrated Primer, I think you might get what I'm saying. But essentially, you can capture a lot of emotion, expression, etc, using a cheap camera and a single actor, and then use AI to "stylize" it in a really cheap effective way, while keeping the emotion/expression. Adding lighting for example, upscaling the quality, changing the voice timbre while keeping the pacing, etc.

at this point, it's pretty easy to create evals/benchmarks, and then run the latest model on them.

LLMs are so easy to swap out, so having good benchmarks/evals are pretty useful.

Even then, a lot of the time the model improvements are so obvious that you don't even need an eval.

No, not really. My comments are organic - they come from a place of frustration, seeing such trivially wrong arguments against data centers.

All the arguments I see against data centers are arguments against industrialization. There are also arguments against capitalism and wealth inequality, that I sympathise with.

exactly. All of the arguments I see against data centers are simply arguments against industrialization.

The dirty water bit is the most ridiculous thing. It's meant to paint the picture that the data centers themselves are using the water and dirtying it, when in reality the cause is the development of the actual buildings and infrastructure, which would happen with any industrial development!

yeah I had this happen to me. Except when I go to maintain it, now cursor/claude are good enough to essentially handle it on their own, so it turns out to be very low effort to maintain.

there's an unnatural amount of doomerism against datacenters, of exactly this kind. It's pretty obviously astroturfed.

i noticed a thing with headlines like these: "x may cause y". Whenever it's "may" or "might", it's almost always meaningless

This is great! I disagree with some of the commenters here that the originals sound better.

The originals sound very much like demos, and from a producer's perspective are very low quality (no offense). They're definitely more raw, but objectively not as good - i.e harmonies aren't tight, the levels are not well balanced, etc.

It's funny that people hate that AI can improve this, because even without AI, modern music uses a ton of digital tools to mix and master - and true musicians don't care whether it's digital or not.

These commenters would be the same people who boo-ed bob dylan when he went electric.

Look at John Mayer - he uses AI to model amps, instead of lugging around giant heavy tube amps.

Question for you - what was the workflow exactly? I've been wanting to test out some AI tools to do similar things with my music.

I don't think that's true anymore

Where are you getting that from, that they're ok with CSAM?

I think they've been clear that they want to follow the law.

Every image gen provider struggles with this. I worked for an image gen app years before it became popular (Wombo dream) - it's a hard problem to solve, there are sick people out there.

The best feature is that it can delegate questions out to GPT-5.5 in the background, so you're no longer restricted to a voice model that's several years behind the frontier.

Ahh, this makes sense. I was wondering when they would start doing this. I stopped using voice mode all together because it was frustrating talking to a dumb AI, when most of the time I discuss things with Opus 4.8 or gpt 5.5.

I was working on a phone call agent recently, and thought about doing this. It makes sense

I've been building multiple products with LLMs, and they are in fact interchangeable for the most part.

In fact, most benchmarks show this! Most benchmarks have similar performance for the same classes of models.

On top of this, there are tools like open router, or even the openai SDK which trivially allows you to swap endpoints for the LLM!

If you're using the agents SDK from openai or something, then yeah it's not interchangeable but that's you doing it wrong

Enterprises switched from openai to anthropic this year - anthropic overtook openai for the first time. I don't see why they wouldn't switch again.

There's barely any moat. All the data is with connectors, memory is near useless

It really depends on what you're doing, but most LLM usage and agentic runs are pretty interchangeable in my experience, and it's usually trivial to switch.

If anything, you're better off supporting multiple LLMs as backup because most model providers have been so inconsistent with working all the time

This is awesome

not true. multimodality is still far from being solved

I agree - and I'm also confused why other's haven't simply copied every feature. It really doesn't seem that complicated.

Anecdotally At the moment, about half of our inbound applicants are fake profiles

right - so imagine how much worse it would be with a paper map or printed instructions

Who hasn't been annoyed with "AI customer service", who hasn't been annoyed with "customer service", period

nah, i distinctly remember family road trips as a kid. Driving was super stressful with my mom yelling about missing an exit as she reads the map, getting lost in a not so nice neighborhood, etc.

you're romanticizing the past

Oh wow, this is great!

but would the scale stay the same?

Take a hotel for example - it's nice to have a butler, someone at the front desk, and a waiter, perhaps. But you don't need the cleaning crew, the kitchen staff, etc, that run behind the scenes. These you could replace with robots, no problem.

I mean, yeah, but it's not like it's done in a deceiving way. Nobody forces anybody to be a first hire

So how much should they be rewarded? You suggest a cap? What prevents someone that's more poor than you deciding that you nobody needs to make more money than them, so now you should make less money?

no, it's not. It's math. If you make a rule that you lower the person at the top of the pile, over time, there will be nobody left to lower - everybody will be at the top of the pile.

The fact people say 1 billion is too much money, or 1 trillion is too much is something to dig into - the number is relative! 1 billion is only a lot because not many people have a billion dollars. If inflation keeps up, one day everybody will be a billionaire.

yeah, but lot's of bad parents exist. Which is why there are laws around kids having to be in school, etc.

Maybe fine the parents if the kids get caught. Teaches them to teach their children better.

incredible to see this opinion on hacker news.

He can borrow against it which gets around taxes and that should probably be addressed, but he like the hypothetical fresh billionaire startup founder don’t have that money. And the mega rich on paper can’t access more than a small percentage of that money without reducing their control of the company they built or are building.

Yes, this is a good take. I wish more people understood this. Things like sales taxes could address this. Land value tax, with single homestead exemptions are another.