HackerTrans
トップ新着トレンドコメント過去質問紹介求人

kappuchino

no profile record

コメント

kappuchino
·28 日前·議論
Needs [2025], as it was written last year.
kappuchino
·2 か月前·議論
I have two of the M3s due testing of models at work and with exo I can run decent quantization with 1 millon tokens for memory and derailment tests.

Slow? Yes, but ... private. Unconditionally.

And recently with https://github.com/antirez/ds4 one can use just one system to a very, very decent speed and ttft for chat inference. Again, private.
kappuchino
·2 か月前·議論
And the list lacks wire for example. I wonder why.
kappuchino
·2 か月前·議論
LLM generated comment.

It is said don't feed the trolls. Looking at eieiyo's comment history, I can really tell - kiddo? Bot? Confused Oldtimer?

Whatever.
kappuchino
·2 か月前·議論
Thanks for the reply. Um. In this particular text, I tend do disagree.
kappuchino
·2 か月前·議論
I read the text myself and checked it with pangram, it's 95% plus human written (people hate hundred percent so the discount). So I'm asking myself if it is ragebait or Idiocracy. I know that's harsh to say, but throwing in written by AI is harsh as well. (Analysis: Pangram believes that this document is fully human-written https://www.pangram.com/history/c59610e0-aba7-469c-b762-21f7...)
kappuchino
·3 か月前·議論
It's lovable when people use a wrench to hammer a nail in. Watch your thumb.

Meaning: You managed your ways around the system prompt and usage intention - Congrats! Now it doesn't work any more - Bummer!

Have you tried opus 4.7 in comparison to 4.6 with a general purpose / writing system prompt in the app? Thats where this would make more sense.
kappuchino
·3 か月前·議論
My line is: No advice from llms. This seems to be one. Sigh.
kappuchino
·5 か月前·議論
Angry Upvote. (This isn't reddit, I know)
kappuchino
·6 か月前·議論
I love services that have not a single person as a contact/responsible for the site. /sarcasm

It's a red flag if you hide behind a contact form with no reachability beyond that whatsoever.

And as other said: 99.5% accuracy means you should have millions of working fingerprints, since mine and others are faulty as hell.
kappuchino
·7 か月前·議論
I purchased physical eSIM Multicard. It worked quite easily, so now I can handle esims in a virtual/physical way.

And I know what you might think: Isn't that defeating the purpose? No, it is not. I can recive a digital simcard activation, render it on a physical card and use it in any phone. Thats perfect.

And yes, once the phones without a sim-card slot come ... we'll see.
kappuchino
·7 か月前·議論
He corrected that in the comment section of the youtube video. Six is actually the maximum amount. He just didn't want to buy another one.

He also published the Benchmarks in Detail and with two/four Macs in Comparison: https://github.com/geerlingguy/beowulf-ai-cluster/issues/17
kappuchino
·7 か月前·議論
Until https://github.com/vllora/vllora/tree/v0.1.6 it was Apache licensed. Then Elastic Search 2. Nah.

IMHO the "don't remove anything with a licensekey ever" part in the license is the kind of potential poison that I would never recommend this to my or any other company. More than a few fellow engineers consider nagware an insult and see the potential to twist your arm late in the game making former free functions part of a new "optimized pay package", which you need because you can't fix the bug in the goddamn license part that is a security risk. LOL. (Not saying that you ever do. See below)

And there is no moat, debugging AI flows is a few prompts and a claude code max, google gemini pro or codex whatever for a couple of days while doing the usual things will happen.

Note: Its not about this software specific. I learned that the cuts and bruises of incidents before you come along are the ones that shape behaviour of your partners/colleagues/peers. You may have the purest intentions and best approaches, but someone longe before you ruined it. Its not you, its you chosing the same path.
kappuchino
·11 か月前·議論
That only works for weeks or so, since they won't be updated, according to the PR.

It's time to build your own from core / foundational images - something I recently learned and now seek to master.
kappuchino
·12 か月前·議論
well, if you read about the exchange beween the author and owners ... add "schwurbeln" (german) to the list of whats weird about the domain.
kappuchino
·昨年·議論
"Complete Guide to Meta Prompting while recommending our Product" would be a more honest title.

I personally reject advice that is muddled with directly offering their own services: Its conflict of interest in my face.
kappuchino
·昨年·議論
! Dataset is not available to me or at my location - the only offer I get is Spotify 2023 and such. Gone?
kappuchino
·昨年·議論
Well, he commented on his post the prompts: Two LLM prompts to Gemini 2.5 were used to help with the content.

> The following is a blog post. Please identify additional points of leverage and sanction in each context mentioned in the blog post.

and

> The following is a blog post. I want to make the content more engaging. I am reluctant to illustrate the points with stories, so I'm looking for other ways to make the content accessible and engaging.

Since it's mostly a list of lists and a starter text (for engagement) ... well played.
kappuchino
·昨年·議論
Says by a green account. Oh the irony!
kappuchino
·昨年·議論
Why are the praising accounts green? Am I missing something?