conradev
The model output is fairly uniform because the models are fairly uniform, but the input is how I understand the prompter’s theory of mind for the LLM.
The higher fidelity the theory of mind, the more productive the resulting conversation is. Basic things, like knowing what the model is even aware of.
It’s okay at subjective product judgements with limited context, and so the best engineers make the most important decisions themselves, constraining the model’s solution space to something looking like success.
Maintaining consistent progress towards a common goal with a bunch of different perspectives is the job.
If a model does something wrong, I don't delete it and rephrase. I explain where it went wrong. I get better results this way!
You will need to turn off your Bluetooth if you want to stop advertising a rotating random number.
The only difference with the contact tracing is that a key, on the device, can correlate the ciphertext. I have it off because the utility was never demonstrated.
You can extract meaning from them! Find your current computer, find the target computer, calculate the percentage difference. 1000 is 1% better than 990, so it indeed doesn't mean anything at all.
On social media in China there is an oft-repeated joke that goes something like this: In other countries, governments intervene to prevent anti-competitive behaviour; here (in China), they intervene to curb competition.
You have SATA or SAS to pick from. The CPU requirements for a storage server are not high. On a typical ATX board you have motherboard SATA and can put SAS controllers in the spare PCIe slots.
My first "NAS" was two 22TB hard drives in a ZFS pool on my motherboard SATA
Once the hacker got in, they enabled PGP with a random key to prevent the account recovery process from working. It took many, many months to get the account back after the attacker used the account to max out advertising spend. Meta did and does not care.
I realize now: why would they change anything? They made money off of the interaction
Solar panels are made in factories where energy and inputs are cheaper and then shipped on diesel cargo ships to the rest of the world.
Our cars run on gas, so we need new electric ones, but the ships and car carriers are themselves… not electric yet.
Because energy is more expensive, everything is more expensive, including electrification.
My current wallet doesn't give me any affordances: https://grifiti.com/products/grifiti-band-joes-3-25-x-1-25-i...
…but some other push-based IPC mechanism would be a lot more battery friendly
Ma's voting rights were reduced from 50% to 6%. Teams, Organizations, Insights, Workflows, SSO/SAML, and other EE-only features have been removed
I use Namespace (https://namespace.so) and I hook it up both to my personal GitHub as well as my personal Forgejo. I’m in the process of moving from the former to the latter!
Coding agents actually help with a lot of the ergonomic issues. If you have an evaluation issue, it can be annoying to climb into nixpkgs to diagnose it. But codex will do that for you.
I’ve found agenix in particular to be really great addition for agents: secrets you can copy around without risk of accidental disclosure.
In a day I can now deploy Caddy, Authentik, Fleet, Headscale, Stalwart, jmap-webmail, Forgejo, SFTPGo, Immich, Grafana, Jaeger, PostHog, etc. and have them all work together. I can do this on a tiny VPS, and codex can actually estimate and test performance to minimize cost.
The equivalent Kubernetes setup wastes so much on isolation and a scheduler that is overkill for anything small.
You can run a codex instance on machine A and connect the TUI to it from machine B. The same open source core and protocol is shared between the Codex app, VS Code and Xcode.
$ echo 'test' | xxd -i -n foo
unsigned char foo[] = {
0x74, 0x65, 0x73, 0x74, 0x0a
};
unsigned int foo_len = 5;