mhaberl

4 ms·

1.
by mhaberl 17d ago
I might do that myself in the end checked the bug reports and PRs first though, there are a few related, nothing merged yet I ran opencode task again last night since it's slow and this morning i got a "SSE read timed out" lo
2.
by mhaberl 17d ago
I dont get it, so for slow inference its: hermes > pi > opencode ?? or is hermes better in general? habent tried it hermes/pi yet
3.
by mhaberl 18d ago
I would agree, but I want to add that I have real issues with combination of opencode plus slow inference (4-5tok/s). I get weird interruptions. I can only guess its related to some kind of timeouts in the harness or something. Its not
4.
by mhaberl 18d ago
2 things here, one is related to benchmaxxing, another to being good enough 1. Mistral isn't benchmaxxing. That doesn't mean they're better, but it does mean the benchmark gap not a good reflection of the actual gap 2. I thin
5.
by mhaberl 21d ago
Well for some company already running IBM Z, using Bobcoins might make sense (we will see what happens) for a FE dev or some nodejs dev this sounds stupid, who cares but for someone who deals with their stack it might mean something to get
6.
by mhaberl 21d ago
This is so ridiculous that I love it I seriously think I’m going to use it just for the lolz Bobcoins! Are they blue? hahaha
7.
by mhaberl 26d ago
DDR5?? dude, DDR6 was planned for 2027 in 07/2025 [1] No one is going to build factories in 2026 to produce DDR5. [1] https://www.techpowerup.com/339178/ddr6-memory-arrives-in-20...
8.
Single-User Inference Card(markohaberl.substack.com)
3 points by mhaberl 1mo ago|0 comments
9.
by mhaberl 1mo ago
> Back in the day I learned the hard way that sending a long reply to an email just ensures the other side won't read it. You can't assume that you dealt with their request effectively that way Then I guess you didn't deal
10.
by mhaberl 1mo ago
Thank you! It seems Qwen is kicking ass, and Fable made me laugh when I saw it all alone on the far right of the graph :))
11.
by mhaberl 1mo ago
lot of tech folks think their writing is boring and ask AI to make it more interesting or whatever llm gets them some of that sweet_llm_padding and readers smell the slop and attack the writing :) people, go write "boring stuff" -
12.
by mhaberl 1mo ago
Did you evaluate any that could be self-hosted (or at least ow models), if so which one is the best you seen?
13.
by mhaberl 1mo ago
> We are however doing our best to fix this: Anthropic is ramping up its efforts very quickly in biology and medicine, and we hope to have incredible results in the coming years and some early glimmers in the coming months. When we’ve ac
14.
by mhaberl 1mo ago
You could run Qwen 3.8 locally at almost 200 tok/s
15.
by mhaberl 1mo ago
I wish we could have better hardware and I think the tech is there for a few years already. I've gone in (too many) details last night with the calcs: https://news.ycombinator.com/item?id=49324600
16.
by mhaberl 1mo ago
what??? not true! for inference the compute is the last thing we need more of. memory bandwidth is the numebr one blocker, after that the inefficiencies that where introduced with MoE models (and all new large models are made that way) Here
17.
Single-User Inference Card(markohaberl.substack.com)
7 points by mhaberl 1mo ago|1 comments
18.
by mhaberl 2mo ago
> massive improvement over IntelliJ for my daily workflow. Why? I used Zed for about 6 months but I lost motivation. I have the full jetbrains subscription and use it for some debugging etc. I got vim for quick edits and cli tooling for
19.
by mhaberl 11mo ago
why? looks like DW-5000C - not so good, but "yuck" is a bit too much, no?
20.
by mhaberl 11mo ago
..and the Wren compiler :) > Context switching is a skill that gets easier the more you practice it, just like any other. Totally agree with this!! I learned this when I started off as a junior dev. We had some shitty machines and the pr
21.
by mhaberl 11mo ago
and he wrote Game Programming Patterns [0] [0] https://gameprogrammingpatterns.com/
22.
by mhaberl 1y ago
>The new way: The entire premise of AI coding tools is to automate the thinking, not just the typing. That’s the promise, but not the reality :) Try this: pick a random startup idea from the internet, something that would normally take 3
23.
by mhaberl 1y ago
This perfectly aligns with my experience. Every large project I have worked on showed a clear correlation between the ease of setup and running and the number of problems on the project, like bugs and missed deadlines.
24.
by mhaberl 2y ago
There is one with the AI and gut bacteria combination: https://www.linkedin.com/company/anibiome/ They recently raised series A. > Ani Biome uses machine learning models to provide personalized solutions for e
25.
by mhaberl 2y ago
> It's wild to me that Teams is so fucking horrible that many businesses who effectively get it for free as part of other dealings still choose to pay for Slack. I don't agree. For many years now I use Linux. I pay for Slack be
26.
by mhaberl 2y ago
This is so true. I remember one of my interviews that went well in the end. In the middle of the interview, they mentioned using Java and building web apps. At the time, I was a C++ developer with no web development experience. I said: &quo
27.
by mhaberl 2y ago
> going to be useless in 2 years, why would I buy it for my kid Don't worry about it. I will be broken or lost far before those 2 years :)
28.
by mhaberl 2y ago
Serious question: I am looking at the measurements of speed on the landing page between Zed and the other editors, and they don't seem drastic. Do you actually feel the difference when typing?
29.
by mhaberl 2y ago
Perhaps it is just wishful thinking, but at the speed its getting better I feel like next year it will suprass Unity overall. Godot is awesome. Thank you John Riccitiello!
30.
by mhaberl 2y ago
> Even my last team leader sent me a message out of the blue saying “I think I’ll run a bar. I want to be a bartender and listen to other people’s stories, not figure out why protobuf doesn’t deserialize data that worked JUST FINE for th

More ›