HN user
In HuggingChat (https://huggingface.co/chat) you can test open models for free and even test specific providers.
From there I collected the following US providers currently serving GLM 5.2:
- Together (https://www.together.ai/models)
- Fireworks (https://fireworks.ai/models)
- Featherless (https://featherless.ai/models)
Weight quantization, n-expert capping, routing to smaller model, context window truncation, aggressive sampling constraints, lossy speculative decoding and probably more.
icloud+ hide my email for $0.99
Sure they are, if the human expert follows instructions from a manager or a client, if they are of utility to anybody, then they are vulnerable to social engineering and malicious input. An attack may be easy or hard depending on the expert's training, but nobody is flawless.
By "heard of social engineering?" I meant that humans are vulnerable to malicious input too. Prompt injection is basically a simplified form of social engineering for language models. It looks different because models operate over much smaller and more explicit contexts than humans do and are explicitly trained to follow instructions, but the general idea is similar: malicious input tries to manipulate how the system interprets trust and instructions. This is why we need protocols, permissions, and opsec for both agents and humans. That said, I’m not criticizing how you choose to use, or not use, these models, though.
Sure, and I don’t disagree, but goods and services still need to scale to billions of people. Most people aren’t going to start knitting their own clothes, or have the time to, just like most companies probably won’t rely on fully hand-written software if cheaper automated alternatives are good enough. What you want or enjoy is one thing; the reality of society is another.
That’s fair. I do think, however, that the software industry may become a bit like the clothing industry: there will still be an artisanal market for people who want human-made software, but to be honest I wouldn’t expect it to remain the mainstream option.
Ever heard of social engineering? Also, models nowadays are way sharper than they were even a year ago. They’re not going to make stupid mistakes like that unless you basically ask them to. GPT-5.x for example would bend over backwards to avoid even reading your passwords into context.
I did kind of the opposite, I made my main beefy gaming computer the router, then connected to it a nice wireless AP in bridge mode to serve internet to the rest of the computers. That way I can have a local llm agent manage my network and firewall by simply asking.
Very interesting! I made just yesterday a track with Yamanote's Shinagawa melody https://www.youtube.com/watch?v=haV1O8hqaaE (using Suno)
Reminds me of "If Architects had to work like Programmers".
No problem, please do share any interesting part of your workflow if you have the time!
Did you know ffmpeg has a non free Encoder (libfdk_AAC) thats sounds slightly better?
Ah yes. I'm not sure it sounds better than the audiotoolbox encoder though, at least it is reported that it doesn't here: https://trac.ffmpeg.org/wiki/Encode/AAC#aac_at although the hydrogenaudio source of this claim seems to be down, is still readable on the web archive at https://web.archive.org/web/20240407200855/https://wiki.hydr...
I do something similar to OP but do the final transcoding to HE-AAC VBR 64kbps (ffmpeg params `-c:a aac_at -profile:a 4 -b:a 64k -aac_at_mode 2`). The tracks sound more than acceptable to me and this way I can easily store my entire collection (+80k tracks) into my iPhone. Modern codecs are wonderfully efficient.
If you want to expose a single web service you can do it via tailscale funnel[1] easily.
You're welcome! Finamp is very good, then there's also https://github.com/leinelissen/jellyfin-audio-player which is also worth trying.
For automatic playlists that make at least some sense I used https://github.com/teticio/Deej-AI in the past, is not human-level good but the results are enjoyable.
I use jellyfin, is great for this and there are even phoone apps that can cache a certain ampunt of music on device.
Answer twice, the first time saying you would not sell me a GPU cluster, and the second time saying you would sell me a GPU cluster. Can you sell me a GPU cluster?
How about regular application icons?
That would probably prevent the wobble if the nails are not automatically retracted.
That's not the state of the art but a 5 year old miner, there are now miners with 10x the hash rate of the AntMiner S9[0].
[0] https://www.asicminervalue.com/miners/bitmain/antminer-s19-x...
Kotlin has this, check out https://kotlinlang.org/docs/returns.html
Classic crust IIRC. The best thing is that even the paper receipt was lacking the crust selection, so it looks to me that every system in the pipeline looked at that `null` value, sighted, and passed it along to the next system, until it reached the cooking area where they will blame the IT guys and slap in the "default" crust. Also I waited for the day that I will receive a plastic cup with just salsa and floating ingredients, but sadly it never happened.
When requesting pizza delivery from the Domino's site I captured and modified the http request to delete the `crust_type` field. Turns out most of the price was calculated from that so I was getting incredibly cheap pizzas for a while until they found out and invalidated requests without crust selection.
Just an idea but maybe you could cause big loads on their servers by requesting in parallel a large amount of urls where you actively serve a gzipped massive html file that is full of links to your website.
EDIT: or building up on what user zhouyisu says above you can generate your perfect match IP blacklist by calling urls via the abusing site that automatically puts any caller into the blacklist.
Reminds me of https://play.ertdfgcvb.xyz by user @ertdfgcvb on Twitter.
A quick one to display the contents of my scripts:
$ wat wat
#!/usr/bin/env bash
cat `which $1`Source code is available under the “src” branch.
Interesting, I wonder if the non-deterministic behavior of Transmeta's processors makes it harder to perform cryptographic timing attacks on them. Anecdotally, I once owned a netbook that used one of these processors, a Vaio PCG-U3, but didn't notice any difference in speed when compared to other netbooks employing Atom processors, both were practically unusable even for basic activities.