selfhoster11

7 ms·

61.
by selfhoster11 11mo ago
Yes, except unironically. A market that cannot efficiently serve the vast majority of the population is a failed market.
62.
by selfhoster11 11mo ago
Planning ahead is the #1 problem for people with executive dysfunction, many of whom work in IT.
63.
by selfhoster11 11mo ago
Yes, exactly. There's a lot of poor people.
64.
by selfhoster11 11mo ago
It's still a good primer for those of us that haven't been keeping up with de-Googled Android for a while.
65.
by selfhoster11 11mo ago
From a corporation with a proven track record in those things? Let's not pretend like they should be given the benefit of the doubt.
66.
by selfhoster11 11mo ago
o3 and GPT-5 will unthinkingly default to the "exposing a reasoning model's raw CoT means that the model is malfunctioning" stance, because it's in OpenAI's interest to de-normalise providing this information in API
67.
by selfhoster11 11mo ago
The more options you have, the better IMO.
68.
by selfhoster11 11mo ago
MDF/MDS isn't open, but could possibly be reverse-engineered enough to read the DPMS data.
69.
by selfhoster11 11mo ago
That actually sounds like a good match for LLMs' ability to do fuzzy intuition and pattern-matching.
70.
by selfhoster11 11mo ago
> It feels like wrestling with 1998 in trying to use VLC these days. It's got a real WinAmp feel to it. What do you mean by that? To me, it's like any other application - you double-click the file (or run `vlc file.mp4`), and i
71.
by selfhoster11 11mo ago
I can see it being useful. Lots of people who don't know English are out there.
72.
by selfhoster11 11mo ago
If 20B reasoning models are the goal, we can do that a whole lot cheaper than for $1T.
73.
by selfhoster11 11mo ago
SSH access over a slow network connection?
74.
by selfhoster11 11mo ago
I don't see how this makes the "JS availability should be the baseline" assumption any more legitimate. We make it possible to function in a society for those 6% of people. Low percentage still works out to a whole lot of peo
75.
by selfhoster11 11mo ago
I don't see the problem here. Nobody should be able to install or disallow installation of software on a device except for the end user.
76.
by selfhoster11 11mo ago
You can navigate to the files you downloaded with NewPipe and tap on them. They will open in the default video player.
77.
by selfhoster11 11mo ago
We don't need more KYC, no.
78.
by selfhoster11 1y ago
Huh? Don't you have a file manager on your phone?
79.
by selfhoster11 1y ago
Hard open source with zero network connectivity of any kind (other than to a llama-server hosted on my own network, or maybe an optional paid provider) is the only way I would trust this. Otherwise, this is literally spyware.
80.
by selfhoster11 1y ago
You're much better off accessing Llama 3 through a third party hoster. Some have a web UI if you don't want to deal with API calls. It's much more transparent this way, since you know that the only moderation layer/syste
81.
by selfhoster11 1y ago
What code is needed to make a decision to go with a smaller provider instead of AWS?
82.
by selfhoster11 1y ago
That doesn't fully explain why this doesn't happen with non-profits.
83.
by selfhoster11 1y ago
The guy in question has a contact link or email. I'd reach out and just ask.
84.
by selfhoster11 1y ago
It's time to switch to the API, then. Not to be glib about it, but an API powered experience where you control the front-end is far more controllable and stable than ChatGPT, which you have no control over.
85.
by selfhoster11 1y ago
Yes, that is exactly right. Those exact companies aren't serious companies.
86.
by selfhoster11 1y ago
Not everything needs to be a race, and certainly not art.
87.
by selfhoster11 1y ago
There is a third-party downloader that uses an API to grab binaries for all games in your library. I just download the individual files one by one, though.
88.
by selfhoster11 1y ago
We've got plenty of servers, electricity, network hardware and people who code. We are missing the oxygen in the room, which American services all collectively sucked out. Banning those services will open some potential for innovation.
89.
by selfhoster11 1y ago
There are LoRAs for large language models, but they seem to be less popular than for image models.
90.
by selfhoster11 1y ago
If it's a smart enough LLM with a bearable personality distinct from a ChatGPT customer service assistant, then why not?

More ›