banana_giraffe

Joined 12 years ago

Karma

4,153

Submissions

930

Member since

Oct 7

If you want to email me for some odd reason:

    a=(b,c)=>'p90m0ch24dfK8WQpoCyYzZSgjmYvLhzQEgKkV5nCgGgZ'.charCodeAt(b*2+c);
    [...Array(22)].map((_,d)=>String.fromCharCode(a(d,0)^a(d,1)^(d+59))).join``

    ⠀

⠀ ⠀

View on Hacker News

Recent submissions

Comment · 15 hours ago

It's not sponsorblock quality, but I have trained a model for podcasts. Turns out, at least for podcasts, you can get away with a surprisingly small model, small enough it can run on device directly.

No idea if it's good enough for all podcasts that are out there. It works for me and the podcasts I listen to, and I have tried testing it against a larger corpus, but I'm one developer and I just wanted to solve a problem for me.

https://github.com/Q726kbXuN/podcast-ad-tagger

Comment · 6 months ago

DuckDB can do exactly this, once you get the API working in your system, it becomes something simple like

    SELECT \* from read_csv('example.csv');

Writing generally involves reading to an in-memory database, making whatever changes you want, then something like

    COPY new_table TO 'example.csv' (HEADER true, DELIMITER ',');

Comment · 7 months ago

A quick benchmark using float32 copies using torch cuda->cuda copies, comparing some random machines:

    Raptor Lake + 5080: 380.63 GB/s
    Raptor Lake (CPU for reference): 20.41 GB/s
    GB10 (DGX Spark): 116.14 GB/s
    GH200: 1697.39 GB/s

This is a "eh, it works" benchmarks, but should give you a feel for the relative performance of the different systems.

In practice, this means I can get something like 55 tokens a sec running a larger model like gpt-oss-120b-Q8_0 on the DGX Spark.

Comment · 8 months ago

And for those that want to get the feed URL, this little Python snippet can get the feed URL from a channel's page, either from the channel's URL itself or a video or playlist page:

https://gist.github.com/Q726kbXuN/834882f59bc921a3865272a66d...

Comment · 10 months ago

They actually added some new question categories a while ago [1]

"Troubleshooting / Debugging" is meant for the traditional questions, "Tooling recommendation", "Best practices", and "General advice / Other" are meant for the soft sort of questions.

I have no clue what the engagement is on these sort of categories, though. It feels like a fix for a problem that started years ago, and by this point, I don't really know if there's much hope in bringing back the community they've worked so hard to scare away. It's pretty telling just how much the people that are left hate this new feature.

[1] https://meta.stackoverflow.com/questions/435293/opinion-base...

Comment · 11 months ago

https://docs.aws.amazon.com/lambda/latest/dg/gettingstarted-...

The default Lambda quota for all accounts is 10240 MB. I've never seen it below that (in recent memory, at least), even on fresh accounts not connected to a big org.

I know I routinely use 10gb of RAM for my account that's never talked to support for the related CPU allocation.

Comment · 1 year ago

Reminds me of Apple's AWDL, a similar workaround for crappy networks when the devices need a high speed low latency network. I do wonder if the headset here will do similar channel hoping tricks to join both the dongle's network and the normal wifi network.

Comment · 1 year ago

Gift link:

https://www.nytimes.com/2025/09/11/world/asia/nepal-protest-...

Comment · 1 year ago

Gift link:

https://www.nytimes.com/2025/09/05/technology/anthropic-sett...

Comment · 1 year ago

https://gist.github.com/Q726kbXuN/c80028aa1714032546d3b442c1...

Nope, they don't eat up space. One of the fun side effects of sparse files on NTFS is calling WriteFile on the sparse file can lead to an ERROR_DISK_QUOTA_EXCEEDED error, which applications tend to not expect.

Comment · 1 year ago

It was briefly suspended:

https://imgur.com/a/NZ4aK60

Comment · 1 year ago

Not quite. The Game Genie presented a UI before the game started where you could enter codes. These codes were basically patches to the game's ROM. When the game ran, the Game Genie would intercept the requests for the bytes of ROM based off your code and replace them with a different value.

There was a prototype of a more advanced model that would be closer to what you're describing, but it never made it to market.

Comment · 1 year ago

Gift link: https://www.nytimes.com/2025/07/09/technology/linda-yaccarin...

Comment · 1 year ago

The site uses "font-size: calc(1.5vw + 1.5vh);" or some variant, for most of the sizes, which has the effect to lock the size for everything to the size of the viewport, regardless of zoom.

Comment · 1 year ago

Last time I ran this at scale was a couple of months ago, so my information is no doubt out of date, but in my experience, YouTube seems less concerned about this than they are when you're grabbing lots of videos.

But that was a few months ago, so for all I know they've tightened down more hatches since then.