Hacker News: dsrtslnd23

Nvidia AVO scores 100% on the ARC-AGI-3 interactive reasoning benchmark

dsrtslnd23 — Fri, 21 Aug 2026 13:26:03 +0000

Article URL: https://twitter.com/NVIDIAAI/status/2090786258981466231

Comments URL: https://news.ycombinator.com/item?id=49387755

Points: 67

# Comments: 37

New comment by dsrtslnd23 in "DeepSeek-v4-flash-vision-exp"

dsrtslnd23 — Fri, 21 Aug 2026 11:17:49 +0000

will this be open weights?

New comment by dsrtslnd23 in "DeepSeek Harness"

dsrtslnd23 — Thu, 13 Aug 2026 13:52:02 +0000

I hear that often but to me it does not feel like it. I built my own framework around pi.dev harness and run all kind of different LLMs with it. Sometimes also use the vendor harnesses and they don't feel better adapted.

New comment by dsrtslnd23 in "Kimi-K3 Releases on HuggingFace 7/27"

dsrtslnd23 — Mon, 27 Jul 2026 09:36:47 +0000

maybe a quantized version on a GB300 would work? unsloth hopefully working on it.

New comment by dsrtslnd23 in "Qwen-Image-3.0: Rich Content, Authentic Details, Deep Knowledge"

dsrtslnd23 — Tue, 21 Jul 2026 10:17:49 +0000

Seems that this will not be open weights?

New comment by dsrtslnd23 in "Historic Photos of NASA's Cavernous Wind Tunnels"

dsrtslnd23 — Tue, 07 Jul 2026 13:35:59 +0000

With all the technological advances, why are wind tunnels still widely in use today (instead of pure computational approaches)?

New comment by dsrtslnd23 in "Anthropic's Method to Losing Goodwill in a Few Easy Steps"

dsrtslnd23 — Mon, 06 Jul 2026 14:06:53 +0000

what is the current statement on Oauth Claude use with Pi. I remember there was some back and forth from Anthropic? I currently use my Claude sub with Pi.

New comment by dsrtslnd23 in "Oomwoo, an open-source robot vacuum you build yourself"

dsrtslnd23 — Thu, 02 Jul 2026 14:47:56 +0000

I agree - OpenMover has that approach. Use a COTS mover and change the brains only https://openmower.de/

New comment by dsrtslnd23 in "GLM 5.2 beats Claude in our benchmarks"

dsrtslnd23 — Mon, 29 Jun 2026 07:42:53 +0000

I agree, but there are use cases for the 'best model' other than converting your 1975 stuff to rust: for use cases where LLMs are just getting started to be useful I really want to use the current 'best' model: e.g. CAD, PCB design etc. In particular anything which requires spatial reasoning. The short time I had access to Fable 5 - it was just way better than any other model.

New comment by dsrtslnd23 in "Ask HN: In the age of agentic coding why no one talks about orchestration tools"

dsrtslnd23 — Tue, 23 Jun 2026 08:45:50 +0000

i thought they did allow third party tools - or did they revert on that again?

New comment by dsrtslnd23 in "GLM-5.2 is the new leading open weights model on Artificial Analysis"

dsrtslnd23 — Wed, 17 Jun 2026 11:46:03 +0000

looks like I need a GB300 workstation

New comment by dsrtslnd23 in "Fidonet: Technology, Use, Tools, and History (1993)"

dsrtslnd23 — Tue, 02 Jun 2026 16:03:00 +0000

I feel like polling mail 30+ years ago on ISDN + zipped mail file from my fido net node was faster than IMAP on my 1 gbit connection now.

New comment by dsrtslnd23 in "Claude Opus 4.8"

dsrtslnd23 — Fri, 29 May 2026 09:40:56 +0000

i find it very confusing - there were reports of accounts being banned due to it - then allowed again - what is the current state? I would like to use it with pi.dev harness.

New comment by dsrtslnd23 in "Show HN: Zot – Yet another coding agent harness"

dsrtslnd23 — Fri, 29 May 2026 07:07:22 +0000

Didn't they allow using oauth in custom harnesses for personal use (e.g. pi.dev)?

New comment by dsrtslnd23 in "ChatGPT Images 2.0"

dsrtslnd23 — Wed, 22 Apr 2026 07:15:11 +0000

actually gpt-image-2 is VERY flexible with the resolution. You can use arbitrary resolution within the max pixel budget.

Ask HN: Why isn't FPGA adoption exploding now that LLMs can write HDL?

dsrtslnd23 — Sun, 05 Apr 2026 11:26:12 +0000

You can now describe a circuit in plain English and get working Verilog out of an LLM. Agents can iterate on it, run simulations, fix timing issues. The biggest barrier to FPGAs was always that writing HDL sucked and the toolchains were hostile. That barrier is rapidly shrinking.

So where is the wave of new FPGA projects? Why aren't hobbyists and startups reaching for FPGAs for stuff that would have been microcontroller or GPU territory before? Cheap dev boards exist, the silicon is there, and now the programming side is getting dramatically easier.

What am I missing? Is the toolchain still the bottleneck even if the HDL itself gets easier? Is it a awareness problem? Or do most people just not have problems where FPGAs are the right answer?

Comments URL: https://news.ycombinator.com/item?id=47648299

Points: 2

# Comments: 2

New comment by dsrtslnd23 in "Opera: Rewind The Web to 1996 (Opera at 30)"

dsrtslnd23 — Tue, 24 Mar 2026 13:01:44 +0000

turn your volume down before opening...

New comment by dsrtslnd23 in "Chrome DevTools MCP (2025)"

dsrtslnd23 — Sun, 15 Mar 2026 22:29:46 +0000

you mean this one? https://github.com/vercel-labs/agent-browser

New comment by dsrtslnd23 in "Emacs and Vim in the Age of AI"

dsrtslnd23 — Sat, 14 Mar 2026 08:23:23 +0000

I think a terminal is still the perfect UI for coding agents on desktop systems. But now that I have running agents in background that only sometimes require some extra input on the go - I hate the terminal use on the go: terminus or other mobile app + virtual keyboard is horrible to use. For that I envision a better interface.

Ask HN: Any informed guesses on the actual size/architecture of GPT-5.4 etc.?

dsrtslnd23 — Mon, 09 Mar 2026 23:51:33 +0000

Does anyone have decent intuitions or hard clues on how big models like GPT-5.4, Gemini 3.1, and Opus 4.6 actually are, and how they compare to the best open models like GLM-5?

Are they all roughly in the same range now (for example around 1T params, maybe MoE), or are the closed models still much bigger?

Also curious about “pro” versions like GPT-5.4 Pro - is that likely a different model, or mostly the same model with more inference-time compute / longer reasoning / better orchestration?

Comments URL: https://news.ycombinator.com/item?id=47317413

Points: 6

# Comments: 0