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

would diffusion models benefit from things like Cerebras hardware?

of course not, but it can often give a plausible answer, and it's possible that answer will actually happen to be correct - not because it did any - or is capable of any - introspection, but because it's token outputs in response to the question might semi-coincidentally be a token input that changes the future outputs in the same way.

> Do you get updates to the proprietary software stack if you go without a subscription license?

what proprietary software stack? they just publish it all on https://github.com/oxidecomputer/ .


it's simpler than that - making it faster means it becomes less of an asynchronous task.

current speeds are "ask it to do a thing and then you the human need find something else to do for minutes (or more!) while it works". at a certain point at it being faster you just sit there and tell it to do a thing and it does and you just constantly work on the one thing.

cerebras is just about fast enough for that already, with the downside of being more expensive and worse at coding than claude code.

it feels like absolute magic to use though.

so, depends how you price your own context switches, really.


which won't be surprising if you think about a little bit - 8.8.0/24 is anycasted, which just means that multiple independent locations around the world announce the IP range, so that your requests broadly go to a nearby instance of it. that's great for your inbound requests, but if it originated it's own DNS queries from that IP, then the replies would also get attracted to whatever instance of 8.8.8.0/24 is near the authoritative DNS server it just queried, not the instance of 8.8.8.0/24 that sent them.

also, even aside from that, having some large fraction of the entire world's DNS requests coming from one IP address would trigger everyone's anti-DOS filters and probably lead to some extremely funny router catastrophes as all traffic hashes to one bucket or whatever.


yes, OPs one has strictly less listings (diskprices.com does multiple countries).


Even terabytedeals.com does, you have a drawer in the top-right corner!


why would you look at the "front page" if you only wanted to see things you subscribed to? that's what the "latest" and whatever the other one is for.

they have definitely made reddit far worse in lots of ways, but not this one.


> why would you look at the "front page" if you only wanted to see things you subscribed to?

"Latest" ignores score and only sorts by submission time, which means you see a lot of junk if you follow any large subreddits.

The default home-page algorithm used to sort by a composite of score, recency, and a modifier for subreddit size, so that posts from smaller subreddits don't get drowned out. It worked pretty well, and users could manage what showed up by following/unfollowing subreddits.


The front page when I used reddit only contained posts from your subscribed subreddits, sorted by the upvote ranking algorithm.


> I have ADHD. I think.

then think about talking to a medical professional, and a therapist, and coming up with your own coping strategies.

> How do you manage the constant stream of thoughts and ideas?

take notes of ideas and come back to them later when you have time.


My ideal is document everything. Every idea. All thoughts. Links. Ides I have. Then thoughtfully come back to them.


My idea is document everything. Every idea. All thoughts. Links. Ides I have. Then thoughtfully come back to them.


yes


Someone should IARC the internal one.


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

Search: