BillFranklin

I reckon Sony get a few more years of even more profitable rent seeking before the EU regulates them like Apple’s App Store and forces a game purchase to be valid on all platforms it’s playable on.

Quicker to type and scan! Though I admit this is preference, delimiters would work fine too.

Parsing works the same but is based on a simple regex rather than splitting on a hyphen.

euc=eu central; 1=zone/dc; p=production; wkr=worker; 1=node id

Thank you! The cloud servers are sufficiently cheap for us that we could afford the extra flexibility we get from them. Hetzner can move around VMs without us noticing but in contrast they are rebooting a number of metal machines for maintenance now and for the last little while, which would have been disruptive especially during the migration. I might have another look next year at metal but I’m happy with the cloud VMs currently.

I didn’t touch on that in the article, but essentially it’s a one line change to add a worker node (or nodes) to the cluster, then it’s automatically enrolled.

We don’t have such bursty requirements fortunately so I have not needed to automate this.

Thanks, hadn’t heard of pigsty. As you say, I had to use nvme ssds for the dbs, the performance is pretty good so I didn’t look to get metal nodes.

I read about 30 LLM papers a couple months ago dated from 2018-2024. Mostly folks are publishing on the “how do we prompt better” problem, and you can kind of get the gist in about a day by reading a few blogs (RAG, fine tuning, tool use, etc). There is also more progress being made for model capabilities, like multi modality, and each company seems to be pushing in only slightly different directions, but essentially they are still black boxes.

It depends what you are looking for honestly “the latest things happening” is pretty vague. I’d say the place to look is probably just the blogs of OpenAI/Anthropic/Genini, since they are the only teams with inside information and novel findings to report. Everyone else is just using the tools we are given.

I stopped using GitHub copilot. I really didn’t like my train of thought being interrupted.

I have heard more comprehensive “leave the thinking to us” tools like cursor give better results.

Personally I think if your tools or projects are so dull that you require an AI to use them, perhaps you need sharper tools or more interesting projects.

To be fair to them, they have done a significant amount of work to design the network to be open to competing servers, and I think it would be quite tricky to unpick that. In comparison to successful networks like TikTok, Twitter, Facebook, LinkedIn, ATP gives a far fairer playing field and Bluesky hasn't done anything (aside from taking funding) to suggest they're not going to run with it.

You are right that they could change their architecture so that their Relay is more trusted or blocks others in some way, once they capture the market. I am not sure what guarantee they could give with the current ATP arch - perhaps a committee would help, but other social networks have no incentive to support ATP.

It looked like a PR stunt