cristoperb

> One has to trust a pretty big company...

Or a medium-sized (~50 employee) nonprofit, anyway.

Ollama is a ycombinator startup, so I guess they have to find some roi at some point.[1]

I personally found Ollama to be an easy way to try out local LLMs and appreciate them for that (and I still use it to download small models on my laptop and phone (via termux)), but I've long switched to llama.cpp + llama-swap[2] on my dev desktop. I download whatever ggufs I want from hugging face and just do `git pull` and `cmake --build build --config Release` from my llama.cpp directory whenever I want to update.

1: https://www.ycombinator.com/companies/ollama 2: https://github.com/mostlygeek/llama-swap

For your target audience it might not matter, but for the hn crowd this is potentially a very confusing naming collision. Especially since the X logo already looks so much like the X11 Window System logo.

Haha, the fact that an LLM tries to offload NLP jobs to Python libraries is funny. But I just tried this prompt (with "Don't use python") and uploaded a text file and it worked well (it immediately started outputting entities without running Python):

"NER task: extract all named entities from this text. Don't use python. JSON format"

I tried Gemini too, and it works without needing to tell it to not use Python.

The API is still better, of course, because you can constrain output to a JSON schema of your choosing... but all models now seem quite good at outputting valid JSON when asked and can follow 1-shot schema examples

It starts a webserver to serve its UI, which is what your comment parent meant. It doesn't provide its own openai-style API, which I guess is what you meant.

My simplified understanding: The target model can validate the draft tokens all at once, in a single forward pass. The output of that forward pass is a list of probabilities for each draft token which are compared to the probabilities produced by the draft model. If the target model's probabilities are the same or greater than the draft model, the tokens are accepted. Worst case none of the draft tokens are accepted and instead the target model selects the single next token as usual.

No, they're text only

That's inexpensive enough to be the basis for a modern recreation of the CueCat:

https://en.wikipedia.org/wiki/CueCat

Even slightly better is to first check both

    if not API_KEY and not CHANNEL_ID:
        print("Missing both YOUTUBE_API_KEY and YOUTUBE_CHANNEL_ID.")
        exit(1)
    if not API_KEY:
        print("Missing YOUTUBE_API_KEY.")
        exit(1)
    if not CHANNEL_ID:
        print("Missing YOUTUBE_CHANNEL_ID.")
        exit(1)

That way you don't end up fixing one just come back and be told you're also missing another requirement

I don't know, but as someone who likes using em-dashes in my writing it is disappointing that they have become a marker of LLM slop.

I hope this sentiment isn't too widespread... I use .xyz for my personal blog and primary email :shrug:

I recently found this youtuber. He has a playlist for 0.22 (which is the dev version for what will be 1.0):

https://www.youtube.com/playlist?list=PLWuyJLVUNtc3UYXXfSglV...

He also wrote uIP, the TCP/IP stack for the Contiki OS which is even smaller than LwIP

https://en.wikipedia.org/wiki/UIP_(software)

See also David Graeber's "On the Phenomenon of Bullshit Jobs":

https://strikemag.org/bullshit-jobs/

sqlite does recognize the SMALLINT keyword and treats it as an INTEGER (sqlite only has two number types, integer and real). See here under "Type Affinity": https://www.sqlite.org/datatype3.html

I think my messages sent over e2e encrypted chat with expiring session keys (like Signal) is more secure than broadcasting on radios with known weaknesses. Can you explain briefly why I'm wrong/"delusional"?

I don't see anything about Putin on that page

Your site looks nice and is quite readable! The thing I most dislike about sites that just use raw HTML is the lack of `max-width` on the text containers (which makes using reader mode necessary), so thanks for including that

Is there a way to refresh from the front page? or do you just quit and start it again?

yeah the landing page is pretty broken on Firefox for me, but the docs site works: https://docs.amber-lang.com/

But that article reports the same numbers as currently on the UN site: "7,797 children (32%), and 4,959 women (20%.)" That's most of the fatalities.

> They updated the ratio of civilians. Turns out the majority of those are adult male combatants.

As of yesterday, the UN is reporting that 60% of deaths are women, children, and the elderly. So even if you want to pretend that all adult men in Gaza are combatants, your minimization is verging on genocide denial. (And the total figures are "not including more than 10,000 reported missing or under the rubble" so the death stats are almost certainly low.)

https://www.ochaopt.org/content/hostilities-gaza-strip-and-i...

> it targets only Israel without laying any blame or responsibility on the Hamas

The reason students are asking their schools to divest from Israel but not from Hamas is because no western schools are funding Hamas or corporations that sell weapons to Hamas. You want protest camps set up to demand schools keep not doing that?

> the UN just updated its casualty statistics

I just looked them up (https://www.ochaopt.org/). 34,904+ killed, 70,000 homes destroyed, and 1.7 million people displaced in Gaza. I agree Hamas is probably worse than the Israeli government in some sort of doctrinal or deontological sense, but in terms of actual atrocities committed, it is not even close at this point.

That reminds me Tolstoy's conclusion in his essay on Anarchism:

"There can be only one permanent revolution — a moral one: the regeneration of the inner man.

How is this revolution to take place? Nobody knows how it will take place in humanity, but every man feels it clearly in himself. And yet in our world everybody thinks of changing humanity, and nobody thinks of changing himself."

https://theanarchistlibrary.org/library/leo-tolstoy-on-anarc...

Even worse, it mistakes the possibility of partial ordering to imply an actual total ordering.

> Try opening https://kmaasrud.com/blogroll.xml in a new tab (Ctrl+click or whatever)

I can't reproduce your issue (on Firefox 115.7.0esr (64-bit) from Debian). I open it in a new tab with middle click and it loads fine immediately.

A static version (archive) of DMOZ is still available at http://www.odp.org/

Can't tell if this comment is supposed to be griping from dour conservative or praise from a communist lichess fan

Yes. Even without enabling root, you can install Termux[1] and have a full Linux cli environment with ssh.

> don't understand not more people want to access their DCIM folder via sshfs

I agree. I sync my camera folder with Syncthing[1], so as soon as I take a photo it is available on my laptop.

1: https://termux.dev/ 2 https://syncthing.net/

It's the name of a "Lo"ng "Ra"nge wifi-like technology:

https://en.wikipedia.org/wiki/LoRa