Edit: I meant what sequence of events would cause etcd to go into an inconsistent state when fsync is working this way

And also in prod, etcd recommends you run with SSDs to minimize variance of fsync/write latencies

https://archive.is/2022.02.18-161603/https://www.anandtech.c...

For an average day (incl. "non-working" hours) the brain uses far less at ~300Wh, and if you include the body the average person needs ~2.3 kWh.

In the rest of the article, they estimate inference electricity use is only ~8% of overall datacenter use, and if we think of the datacenter as the "body" in which the GPU / brain wouldn't work without, that's an overall median use of ~16 kWh for only 24 Claude Code requests.

I'm more impressed with the human brain's energy efficiency + multimodality + long term context + malleability than anything after using LLMs a bunch, even though I learnt a lot about that in a neuropsych course long time ago.

Conventionally programming software needed you to know your tools like language, framework, OS, etc. pretty well. There’s a divergent set of solutions dependent on your needs and the craftsmen (programmers/engineers) you went to. Many variables you needed to know to produce something useful. You need to know your raw materials, like your wood.

With LLMs it’s weirdly convergent. Now there’s so many ways to get the same thing because you just have to ask with language. It’s like mass produced furniture because it’s the most common patterns and solutions it’s been trained on. Like someone took all the wood in the world, ran it through some crazy processing, and now you are just the assembler of IKEA like pieces that mostly look the same.

There’s a lost in necessity in craft. It helps to know the underlying craft, but it’s been industrialized and most people would be happy enough with that convergent solution.

Was curious whether ads would cover the cost of inference, so a bit of napkin math.

They seem to display ~3 ads per minute, on tech products, presumably with pretty good signal and intent based on recent chat history. Not the most up to date on CPM but based on some basic searches we assume $30 per thousand impressions, that’s about 9c per minute and ~$5 per hour. Of course users aren’t always looking at the agent coding, but averaged out over say 3 hours of usage per day, that kinda covers the cost. The $10 per day limit is probably related to average daily session use.

On ChatGPT showing an ad per conversation with good signal and intent audiences could have pretty high CPM or CPC too, easily $0.01 to $0.10 per conversation? I think that’s easily sufficient to cover the API pricing for ChatGPT 5.2 instant or mini thinking for the majority of users queries.

Was it in a different nature to current VC funded FOSS though? It sounds like their contributions to FOSS was tangential and not the sold product?

Maybe a bit more like Google and Chrome?

One can certainly enjoy the laborious handcrafted process of building your own table, and yet go back to a shop that churns out cheap furniture that’s nonetheless useful for many others, and see the value in both.

Obviously there’s more degrees of freedom in software, but I’m trying to see it that way to rationalize how I’m feeling with the current state of things.

FOSS software that many rely on that has been around for a while were non-VC: VCS, Linux / GNU / BSD, web browsers, various programming languages, various databases...

Not bad as long as you know what you’re getting yourself into.

5-6 years ago, I remember trying to scale up to the 16/32GB tier was miserable and oversubscribed; moving to Hetzner at the same price point brought a huge performance boost (mostly on CPU and disk access speed).

Comments URL: https://news.ycombinator.com/item?id=46411153

Points: 3

# Comments: 0

Did a quick test, since before this I also used some very ad-heavy p2p solution, and I see similar issues there. Not sure if you're looking for feedback, but these were all issues I considered before settling on a server-based HTMX long-interrupted-polling approach, which if you think about having server + client + realtime-ish features in the context of "just htmx" + tiny LoC is pretty cool (well I think it's pretty cool :D)

In the WebRTC p2p approach, without some sort of sync protocol that validates the state of data:

- the host must be online / already there to join a room; the host leaving the room means everyone gets kicked!

- if you rejoin a room and don't receive updates, you get a partial view of the data

- if you have data connectivity issues, you get a partial view of the data

- you must have a WebRTC capable browser and Internet connection

App (can take a few seconds to spin up if dormant): https://estimate.work/

Source: https://github.com/weiliddat/estimate-work

Comments URL: https://news.ycombinator.com/item?id=46262785

Points: 1

# Comments: 0

Just to be pedantic it is integer scaled (from 1440x900 to 2880x1800 but then resampled down to the native resolution of the MBA 2560x1600 via something better than bilinear).