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

You need virtually no maths to deeply and intuitively understand embeddings: https://sgnt.ai/p/embeddings-explainer/

for some reason I thought this would look much more bucolic than it does

Ironically, hydro power causes a lot of environmental problems - a lot of glens in the Scottish highlands have had ugly roads bulldozed up them to install fairly small hydro schemes. Maybe this is worth it overall, but they can certainly be terrible to look at.

I live in a desert where we have district cooling (and no shortage of sand or solar power), instead of district heating. Wonder if they can pull off the same trick.

Well you can't really do -600C sand (or anything), so the benefits of sand VS water largely diminished. "just" freezing water already gives you around 300C equivalent of sand (if my napkin is correct).

Also the point of this plant is to exploit the counter-correlation of cheap electricity and cold. Usually there is a bigger correlation between cheap electricity and heat.


You can use heat to create cool by using absorption materials. It's of course way more complicated than with heat. But anyway with that, stored heat in sand could be used to create district cooling.

> you can't really do -600C sand (or anything)

You can if you stagger AC/HP or even peltier elements.


You misunderstood-- temperature is physically limited to -273°C, this is not an engineering problem. You have a smaller usable temperature range in a "cold storage" than with heat from fundamental physics alone.

Damn you got me there!

Cooling needs tend to correlate with the availability of solar energy. While heating especially far north does not so much.

In theory, yeah, cooling the sand would work, and it wouldn't freeze / expand. You'd need to use a coolant that doesn't freeze though, and of course keep any liquid out of it.

gpt-oss-120b has cost OpenAI virtually all of my revenue, because I can pay Cerebras and Groq a fraction of what I was paying for o4-mini and get dramatically faster inference, for a model that passes my eval suite. This is to say, I think high-quality "open" models that are _good enough_ are a much bigger threat. Even more so since OpenRouter has essentially commoditized generation.

Each new commercial model needs to not just be better than the previous version, it needs to be significantly better than the SOTA open models for the bread-and-butter generation that I'm willing to pay the developer a premium to use their resources for generation.


You are absolutely right, and exactly the same thing came into my head while reading this. Some of the replies to you here are very irritating and seem not to grasp the point you're making, so I thought I'd chime in for moral support.

Yah, I can’t imagine much of the food from The Great British Bake Off (as we call it) goes to waste!

> I am genuinely unsure where this disconnect comes from

Much easier to sympathise with a live animal that looks like an animal than with a brown rectangle covered in sauce. Also much easier to sympathise with the plight of one entity rather than millions: a GoFundMe for a relatable charity case rather than helping the billions of people worldwide who need it


The "Extra details that didn't fit in the post" at the bottom of the post are very much worth reading, even if they didn't naturally open themselves up.

Scanning non-destructively is much more expensive than scanning destructively, and arguably first editions of books and obscure and old books provide inferior training material than to newer versions. Easier to just bulk-buy 2nd hand books by weight.

I recorded a message, but having put in the auth code, it’s not entirely clear if it was accepted and the message was sent! Instead I was redirected to download the app without a confirmation message

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

Search: