Hacker Newsnew | past | comments | ask | show | jobs | submit | mandeepj's commentslogin

I call the orange guy many things! I believe he's an accidental president. DNC screwed up big time both times. The stakes were higher than ever, so they could have played it safe by looking at past elections, but nope. They wanted to write history, but got the other guy to do it.

Bush (reminder: a republican) screwed things so bad that the country opened to something that had never happened before - A black President.

Now, orange guy (again, a republican, see the pattern) has screwed, and I'm not sure where his bottom is, will set the country to accept again something that hasn't happened before - A Woman President; maybe a black one. There's still time until the 2028 general election.

Also, what do conservatives conserve? They conserve their brains by not using them. Don't take my word; just look at the history, what they have done so far! They are the same everywhere - be it the US or India - same hate mongering lunatics!


> I had 15,000 hours of audio data

do you really need that much data for fine-tuning?


More data -> better, faster on-device models

The actual plan was to distill Gemini 2.5 Pro into the best on-device voice dictation model.

Pretty sure it would have worked. Alas.


Reasons for running local aside...

What is the practical latency difference you see between on-device and, say, whisper, in streaming mode, over the internet? Comparable? Seems that internet latency would be mostly negligible (assuming reasonable internet/cell coverage), or at least compensated for by the higher end hardware on the other side?


depends on the model!

If you run a smaller whisper-distil variant AND you optimize the decoder to run on Apple Neural Engine, you can get latency down to ~300ms without any backend infra.

The issue is that the smaller models tend to suck, which is why the fine-tuning is valuable.

My hypothesis is that you can distill a giant model like Gemini into a tiny distilled whisper model.

but it depends on the machina you are running, which is why local AI is a PITA.


> Payment of compensation to Iran

Fox News is still singing in chorus about the billion dollars payment to Iran by Obama.


> effectively $20/$200 in credits for codex

So, 1.3ish million tokens for Codex? Following the token limit from here https://openai.com/api/pricing/


It's not a product, but enablement or a feature! Just like a 'Pro' label :-)

> America is at near full employment [2]

That can’t be further from the truth


> even a minimal setup (1 vCPU, 2 GiB RAM, 5-minute session lifetime) would put us north of $70,000 a year based on Daytona's per-second sandbox pricing ($0.0504/h per vCPU, $0.0162/h per GiB RAM)

$70k?

how about if we round off one zero? Give us $7000.

That number still seems to be very high.


Hm. I think a dedicated 16-core box with 64 ram can be had for under $1000/year.

It being dedicated there are no limits on session lifetime and it'd run 16 those sessions no problem, so the real price should be around ~$70/year for that load.


It looks like, to me, that someone spent a long back-and-forth with an LLM refining a design - everything they wrote screams "over-engineered, lots of moving parts, creating tiny little sub-problems that need to then be solved".

I find it very hard to believe that a human designed their process around a "Daytona Sandbox" (whatever the fuck that is) at 100x markup over simply renting a VPS (a DO droplet is what, $6/m? $5/m?) and either containerising it or using FreeBSD with jails.

I'm looking at their entire design and thinking that, if I needed to do some stuff like this, I'd either go with a FUSE-based design or (more flexible) perform interceptions using LD_PRELOAD to catch exec, spawn, open, etc.

What sort of human engineer comes up with this sort of approach?


> What sort of human engineer comes up with this sort of approach?

I don't know. There is that "just-bash" thing in typescript which they call "a reimplementation of bash that supports cat and cd".

The problem they solve I think is translating one query language (of find and ripgrep) into one of their existing "db". The approach is hilarious of course.

It's "beyond engineering" :)


At that point I would buy an old mini PC off of ebay and just put it on my desk.

Could OpenAI have released a local paid version, instead of shutting down Sora? Maybe. A lot of users have beefy machines.

You probably need $100K of hardware to run Sora.

Full quality Sora yeah probably needs serious hardware. distilled version on a 4090 though? maybe. danjl earlier in this thread made a solid case for just distributing the weights and letting people run locally. the SD/ComfyUI crowd already does this daily. OpenAI won't because deepfakes. they already had a mess WITH server-side moderation. open weights with zero moderation, good luck with that PR.

> OpenAI won't because deepfakes.

Do you think someone would spend 5 or 6 figures on a license and hardware to create deepfakes?


People pay for OnlyFans accounts; why not this?


If larry hadn't sucked dRump's c*ck that hard, he'd not be in the free downfall that he is in today. He's dismantling America brick by brick. I hope he goes bankrupt soon.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: