Hacker News: Kostic

New comment by Kostic in "Xiaomi-Robotics-1"

Kostic — Mon, 20 Jul 2026 09:09:35 +0000

Sure but that holds true for most new tech. Expensive and slow but it can only get better or stay the same. I'm thinking it's only going to get better.

New comment by Kostic in "Kimi K3: Open Frontier Intelligence"

Kostic — Thu, 16 Jul 2026 22:48:13 +0000

Your video is showing Kimi 2.6, not 3? Once the weights are released, there might be providers that serve it without the censorship filter.

New comment by Kostic in "Oxide computer 3D rack guided tour"

Kostic — Thu, 25 Jun 2026 19:14:35 +0000

A very beautiful website and a machine. Oxide folks should be proud, you can see the love that went into it.

New comment by Kostic in "Oracle shed about 20k roles globally in the last year"

Kostic — Tue, 23 Jun 2026 14:39:17 +0000

Time to copy&improve the niche solution.

New comment by Kostic in "Steam Machine launches today"

Kostic — Tue, 23 Jun 2026 06:09:08 +0000

70% of the Steam users have a PC that's weaker than the Steam Machine[0]. These things will sell out, that's not the issue. The issue is that Valve probably won't have enough power to secure hardware deals to fulfill all the orders, thus limiting the growth of their hardware side.

[0] https://www.techpowerup.com/342970/valve-claims-steam-machin...

New comment by Kostic in "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"

Kostic — Tue, 16 Jun 2026 08:26:21 +0000

Auth with Github Copilot and then point it to localhost[0]. Hopefully the auth to Copilot requirement will be dropped for local models at some point. Would love to use a fully open stack (VSCodium and everything) in the future. My config:

``` [ { "name": "http://127.0.0.1:8888/v1", "vendor": "customendpoint", "apiKey": "llama.cpp", "models": [ { "id": "gemma4-31b", "name": "Gemma 4 31B", "url": "http://127.0.0.1:8888/v1/chat/completions", "toolCalling": true, "vision": true, "maxInputTokens": 65536, "maxOutputTokens": 8192 }, { "id": "qwen3.6-27b", "name": "Qwen 3.6 27B", "url": "http://127.0.0.1:8888/v1/chat/completions", "toolCalling": true, "vision": true, "maxInputTokens": 180224, "maxOutputTokens": 8192 } ] } ] ```

[0] https://code.visualstudio.com/blogs/2025/10/22/bring-your-ow...

New comment by Kostic in "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"

Kostic — Mon, 15 Jun 2026 17:49:48 +0000

For personal needs I connected VSCode with llama.cpp running Qwen 3.6 27B or Gemma 4 31B and it's good enough to cancel my cloud subscription.

Qwen running on my 1st GPU at q4@176k context from 70 to 50 tok/s with MTP, pretty good for coding.

Gemma on the other hand is using both GPUs, running q8@64k context, doing document sentiment analysis, summarization, proofreading and translating, at consistent 25 tok/s. Somewhat slow but usable for batched workflows. Might get some more once llama.cpp starts supporting MTP with tensor split mode.

Still using frontier LLMs at dayjob since I'm not paying it and those are obviously better. Hopefully we'll have a Sonnet 4.6/Opus 4.5 level 30B model in a year or so.

EDIT: Prompt processing starts from 800 t/s and drops to 400 t/s. In most cases my starting prompts are around 16k-24k of tokens and require from 60 to 90 seconds to be processed. Not great but acceptable.

New comment by Kostic in "Ask HN: What was your "oh shit" moment with GenAI?"

Kostic — Fri, 05 Jun 2026 22:02:12 +0000

"I'm sorry" is not gaslighting but an admission of fault it learned from our texts. And if an LLM managed to delete your database, it's time to slow down the vibe train and put up some guard rails.

LLMs are awesome but not without supervision.

New comment by Kostic in "Gemini Robotics-ER 1.6"

Kostic — Wed, 15 Apr 2026 15:19:29 +0000

Taalas showed that you could make LLMs faster by turning them into ASICs and get 10k+ token generation. It's a matter of time now.

New comment by Kostic in "The MacBook Neo"

Kostic — Wed, 11 Mar 2026 17:59:37 +0000

For now. These will be pretty cool Linux machines if Asahi starts supporting them at some point.

New comment by Kostic in "Qwen3-Coder-Next"

Kostic — Tue, 03 Feb 2026 20:53:19 +0000

I would not go below q8 if comparing to sonnet.

New comment by Kostic in "A few random notes from Claude coding quite a bit last few weeks"

Kostic — Wed, 28 Jan 2026 12:06:43 +0000

I assume you're living in a city. You're already renting out a lot of things to others (security, electricity, water, food, shelter, transportation), what is different with white collar work?

New comment by Kostic in "GitHub: Git operation failures"

Kostic — Tue, 18 Nov 2025 21:09:22 +0000

At least with GitHub it's hard to hide when you get "no healthy upstream" on a git push.

New comment by Kostic in "ChatControl update: blocking minority held but Denmark is moving forward anyway"

Kostic — Sun, 14 Sep 2025 20:34:31 +0000

I don't understand how you can ship a car without proper papers out of the country so easily. Maybe focus on that first?

New comment by Kostic in "Mistral Releases Deep Research, Voice, Projects in Le Chat"

Kostic — Thu, 17 Jul 2025 18:49:59 +0000

Agentic editing is really nice. If on VSCode, Cline works well with Ollama.

New comment by Kostic in "Cloudflare Introduces Default Blocking of A.I. Data Scrapers"

Kostic — Wed, 02 Jul 2025 18:06:57 +0000

This would be true if not for open-weights (and even some open source) LLMs that exist today. Not everything should be done for profit.

New comment by Kostic in "Ask HN: What do you spend your money on?"

Kostic — Mon, 02 Jun 2025 13:47:57 +0000

I'm spending fair amount of money on helping teachers and professors in Serbia who had their pay reduced to to absurdity because they are supporting their students in demands for justice in the case of Novi Sad canopy collapse[0].

[0] https://en.m.wikipedia.org/wiki/Novi_Sad_railway_station_can...

New comment by Kostic in "Claude 4"

Kostic — Fri, 23 May 2025 08:00:46 +0000

It's relevant from an engineering perspective. They have a way to develop a new model in months now.

New comment by Kostic in "Ask HN: Do you still use search engines?"

Kostic — Wed, 09 Apr 2025 07:03:21 +0000

I've stopped using search engines actively 18 months ago. My first stop is an LLM. Once I understand what I actually need, I do a web search to go to the product/tool website. I do this not because LLMs are that good but because web search result quality went way down in the same period.

One interesting trend that I like is that I started using local LLMs way more in the last couple of months. They are good enough that I was able to cancel my personal ChatGPT subscription. Still using ChatGPT on the work machines since the company is paying it.

New comment by Kostic in "Military grade sonic weapon is used against protesters in Serbia"

Kostic — Mon, 17 Mar 2025 10:09:36 +0000

This happened during 15 minutes of silence for 15 persons that died in the collapse. Everyone was silent and mourning.