nik736

5 ms·

1.
by nik736 4d ago
Anthropic trained their LLMs with copyrighted stuff but distillation is bad. Anthropic has closed models, China releases as open weight. DeepSeek even allows distilling their models, but China = bad. Understood.
2.
by nik736 8d ago
Mistral Small 4 is also a MoE model, with way more params, so I would expect it to perform better. Our benchmark involves around 10% communication and writing in German and English and Gemma 4 26B A4B beats it although creative writing is t
3.
by nik736 9d ago
Mistral has solid OCR, STT and TTS models and I would love to support them by switching with all of our business workloads to Mistral... but their LLM models are sadly not competitive at all. In our business benchmarks their Mistral Medium
4.
by nik736 24d ago
For me Opus 5 was the nail in the coffin. Fable without the restrictions was a great model, but became unusable with the security guardrails. Opus 5 became so bad and slow it's unbearable. And since Fable falls back to Opus all the tim
5.
by nik736 2mo ago
Gemma 4 31B is underrated. It surprises me a lot.
6.
by nik736 3mo ago
I have put together an internal benchmark on 1000s of business documents with weird tables, structure, etc. that I run on every relevant model release. Opus 4.8 performs very very well. But it is obviously overkill for the task (and expensi
7.
by nik736 3mo ago
Opus is very good at OCR. Way better than the small 1-4B VLMs. If Opus failed, most likely those smaller models will fail as well.
8.
by nik736 4mo ago
Why would you need the 3rd run if you pick the "one in the middle"?
9.
by nik736 6mo ago
> (I originally was going to say a computer that plays chess, but computers play chess with no intuition or instinct--they just search a gigantic solution space very quickly.) Isn't that how LLM models are trained right now? Trying
10.
by nik736 6mo ago
In Germany we have several accounting software solutions like that for 5-10+ years that integrate with bank accounts, paypal, etc. - automatically suggests booking accounts and exports it via a REST API to the software accountants use. Acco
11.
by nik736 7mo ago
No lightmode?
12.
by nik736 8mo ago
Well, we have to "register" every new IP or new mail server with them as well. It's annoying and a weird system, but they respond quickly and it's just one todo we have to think about.
13.
by nik736 9mo ago
> Meaning that the technology was there and ready to make an experience that was truly excellent In general I would agree, but Siri is honestly still so bad.
14.
by nik736 10mo ago
What I am missing with Gnome is the global menu I have with macOS. It's just my preferred way of working. This is also what I liked about Unity. Gnome seems to follow the same direction as Windows. Additionally miller columns in Finder
15.
by nik736 10mo ago
Yes, but only because of data privacy concerns.
16.
by nik736 10mo ago
GitLab is very very heavy with a lot of bloat and sadly still a bad UI/UX. I prefer Gitea for its simplicity. Gitea Actions are similar to Github Actions and they work great.
17.
by nik736 10mo ago
Which models will this be able to run at an acceptable token/s rate?
18.
by nik736 10mo ago
The problem with ONCE is that software is never finished. This is why most ONCE software that is still available today is charging a one off licensing fee + update fee (e.g. charge yearly for major updates or 10% of the one off fee per year
19.
by nik736 11mo ago
It's only on-die ECC not real ECC
20.
by nik736 11mo ago
It's an interesting article, thanks for that. What people forget about the OVH or Hetzner comparison is that for those entry servers they are known for, think the Advance line with OVH or AX with Hetzner. Those boxes come with some dra
21.
by nik736 11mo ago
The most annoying thing for me currently is that when connecting to local smb shares with Finder and adding favorites (directories on shares), after a reboot they are still there under favorites, but it won't connect to them when click
22.
by nik736 11mo ago
They changed their license to AGPL, removed features (Web UI, etc.) and now they don't provide docker images/binaries. It's their project but; what's next?
23.
by nik736 11mo ago
Is there a fork already?
24.
by nik736 11mo ago
Twilio seems to be affected as well
25.
by nik736 11mo ago
They limit them to 7500 IOPS, as stated in their docs. It also doesn't scale with size, the limit is there for every volume of any size.
26.
by nik736 11mo ago
Thanks for asking! The infrastructure is actually the less interesting part for us, since our platform is written to be completely portable. The USP is the platform itself, that means the managed aspect. You can run your workload on our pla
27.
by nik736 11mo ago
There is also https://www.hetzner.com/ (IaaS), https://www.leaseweb.com/en/ (IaaS) and https://www.nodion.com/en/ (PaaS). Disclaimer: I am the founder of Nodion.
28.
by nik736 11mo ago
If you have enough memory to load a model, but not enough bandwidth to handle it, you will get a very low token/s output.
29.
by nik736 11mo ago
This is only the base model, no upgrades yet for the Pro/Max version. The memory bandwidth is 153GB/s which is not enough to run viable open source LLM models properly.
30.
by nik736 1y ago
https://arstechnica.com/security/2025/06/meta-and-yandex-are...

More ›