porridgeraisin

Mode-seeking is describing the way in which the distribution changes. RL is capable of picking out slightly lower probability trajectories and moving them toward the top of the distribution. However, …

by base distribution, I meant the base model's output distribution

https://socket.dev/blog/mini-shai-hulud-miasma-and-hades-wor... …

RLVR still does not expand beyond the base distribution though, it only mode-seeks within it. i.e, evaluation, retention yes. variation or "planning" no. That is not to say you cannot use LL…

RLVR still does not expand beyond the base distribution though, it only mode-seeks within it. i.e, evaluation, retention yes. variation or "planning" no. That is not to say you cannot use LL…

Yea, in many usecases the tooling space is increasingly sophisticated context management such as fine tuning domain specific mappings into the model so that it is able to work directly with a compress…

> even if the variable is declared static No, for static even padding bytes are zero. For automatic, yes it may effectively turn a = {} to a.member = 0, leaving the padding bytes uninitialised. Or …

> next time you come online Yeah that's the issue isn't it? I see in the UI it's sent. But actually it's sent only the next morning. To be fair. It's fine for an issue trac…

(curious) What if a user closes it before 4 seconds? Ctrl+enter, it optimistically locally updates within 1 second. I close ctrl+w. But my wifi goofed and it didn't reach the server.

Alt-drag works perfectly well on windows. I use a third party app called alt-drag to enable it. Has worked fine for years. It also allows you to use it with win-drag of course.

I am yet to do a "deep dive" into the results, but what a well written article. An LLM could _never_ write so crisply.

I would change it from `can` to `almost forced to`. Since I am seeing nuclearisation live, here are a few observations. Industrialisation is an inherently compounding event. Thus, it gets concentrated…

I would argue it is due to nuclearisation of families, that has always accompanied industrialisation. See my other comment on this post https://news.ycombinator.com/item?id=48416228 …

I have written about this before here: https://news.ycombinator.com/item?id=44098431 BTW, the fertility rate is _increasing_ now (granted from an existing base of 1.2TFR) in the riche…

The real lesson is to learn low level details once and use it forever. The advantage of abstractions that maintain backwards compatibility better than most (linux syscall, SQL, etc) is that they becom…

Yeah, I'm waiting for the panther lake (or similar) ones to become mainstream and I'll get a higher end laptop with those next. The latest framework once they open their parts marketplace in…

Another advantage recommending macs to my relatives... The apple branding and the sleek look makes them treat it much more carefully. An equally expensive less sleek laptop they treat like a fridge. T…

The price thing is really true. My current laptop i got at 48000 INR student discount (retailer at 65000) (the older HP Pavilion Aero 13inch). It's great 950g, 16gb ram, etc etc. still works well…

Memory is the main constraint. You have what, 8mb of psram. Compute wise you can manage. You can do quantisation and run a small 10-15 layer CNN perhaps. Image classification is possible. Keep in mind…

That's exactly what I said... and also not how wayland works. I cannot comprehend the way wayland folks think... quote from the xdg-pip discussion: > To not make PiP windows effectively "…

I don't get it, if you're on google meet, and you want to make one of many videos PiP. How can you ever do that in the window manager? It has to be done in the application! You right click o…

This is the issue with imposing semantics of the programming model on the behaviour. User behaviour is the only _real_ thing, it happens. Everything else is in your head. If people in the real world u…

I am not the greatest fan of most of systemd's features. I will always prefer it tho since I just view it as a "packaging format". The same way I view docker. It is just that it happens…

Urban poor of India

Yep, and there is (rightfully) general distrust in giving teachers that much authority over students. Parents already have that authority, which is why a family environment conducive towards education…

It's probably a threshold thing isn't it? You wouldn't get 20% of the effect at 20% of the traffic. There's a step function in there somewhere.

I'm not exactly a systemd shill. But what? How?