Hacker News: Jacques2Marais

New comment by Jacques2Marais in "Step 5 Preview: Advancing the Pareto Frontier"

Jacques2Marais — Sun, 20 Sep 2026 05:36:15 +0000

For anyone else looking for the pricing: https://platform.stepfun.ai/docs/en/guides/pricing/details#p...

New comment by Jacques2Marais in "Ask HN: What are you working on? (August 2026)"

Jacques2Marais — Tue, 11 Aug 2026 18:38:37 +0000

Thanks for the feedback! Will definitely take into account. This is still very much an early version, and so there is a lot of work still to be done. But the feedback is good, thanks.

New comment by Jacques2Marais in "Ask HN: What are you working on? (August 2026)"

Jacques2Marais — Mon, 10 Aug 2026 16:07:21 +0000

Thanks for your reply! I've sent you an email (from the website in your HN bio).

New comment by Jacques2Marais in "Ask HN: What are you working on? (August 2026)"

Jacques2Marais — Mon, 10 Aug 2026 05:54:59 +0000

I've been working on a vocal pitch-and-time editor called PitchDance, something like Melodyne. It works both as standalone software or a plug-in for your DAW. The idea is to make software like this more accessible by selling at a much cheaper price than the majors, and only competitors I could find on the market.

https://www.pitchdance.com/

Btw, currently I'm looking for some people to test the software, so if you're interested let me know and I'll give you a free license.

GPT-5.4 says it's GPT-5 in Codex

Jacques2Marais — Fri, 29 May 2026 11:55:03 +0000

Article URL: https://old.reddit.com/r/codex/comments/1tqza0x/gpt54_says_its_gpt5_in_codex/

Comments URL: https://news.ycombinator.com/item?id=48321981

Points: 3

# Comments: 0

New comment by Jacques2Marais in "Qwen3.7-Max: The Agent Frontier"

Jacques2Marais — Wed, 20 May 2026 19:56:38 +0000

Interesting. I wonder if current LLMs can break out of human limitations and understand the world more correctly.

Hacker News, the day after Hacker News launched

Jacques2Marais — Fri, 17 Apr 2026 09:43:27 +0000

Article URL: https://web.archive.org/web/20070222081450/http://news.ycombinator.com/news

Comments URL: https://news.ycombinator.com/item?id=47804140

Points: 3

# Comments: 0

Expert Personas Improve LLM Alignment but Damage Accuracy

Jacques2Marais — Tue, 24 Mar 2026 07:05:05 +0000

Article URL: https://arxiv.org/abs/2603.18507

Comments URL: https://news.ycombinator.com/item?id=47499419

Points: 1

# Comments: 0

New comment by Jacques2Marais in "Ask HN: What Are You Working On? (March 2026)"

Jacques2Marais — Mon, 09 Mar 2026 14:45:54 +0000

Working on a Vercel-like cloud hosting PaaS, but specifically tailored to South Africa. Everything is hosted on local servers, and pricing is ZAR instead of USD. It's called Zanode if you want to check it out :)

https://www.zanode.co.za/

New comment by Jacques2Marais in "GPT-5.4"

Jacques2Marais — Thu, 05 Mar 2026 18:22:35 +0000

I guess a big chunk of their target market won't know how to use APIs.

Doing a Video Call over a Database

Jacques2Marais — Tue, 03 Mar 2026 07:15:33 +0000

Article URL: https://www.youtube.com/watch?v=zwIc9fFcYVw

Comments URL: https://news.ycombinator.com/item?id=47229189

Points: 1

# Comments: 0

New comment by Jacques2Marais in "Show HN: Better Hub – A better GitHub experience"

Jacques2Marais — Thu, 26 Feb 2026 11:05:40 +0000

I think most of the permissions can be toggled?

Gemini 3.1 Pro is surprisingly good at classifying banking transactions

Jacques2Marais — Wed, 25 Feb 2026 06:31:55 +0000

Article URL: https://butternut.click/blog/gemini-3-1-pro-banking-transactions

Comments URL: https://news.ycombinator.com/item?id=47148078

Points: 2

# Comments: 0

LLMs are expert movie script writers

Jacques2Marais — Fri, 20 Feb 2026 11:27:20 +0000

Article URL: https://butternut.click/blog/ais-are-expert-script-writers

Comments URL: https://news.ycombinator.com/item?id=47086633

Points: 3

# Comments: 0

New comment by Jacques2Marais in "Qwen3.5: Towards Native Multimodal Agents"

Jacques2Marais — Mon, 16 Feb 2026 12:50:47 +0000

Yes, I also see that (also using dark mode on Chrome without Dark Reader extension). I sometimes use the Dark Reader Chrome extension, which usually breaks sites' colours, but this time it actually fixes the site.

Qwen3.5

Jacques2Marais — Mon, 16 Feb 2026 12:01:18 +0000

Article URL: https://huggingface.co/Qwen/Qwen3.5-397B-A17B

Comments URL: https://news.ycombinator.com/item?id=47033981

Points: 5

# Comments: 0

New comment by Jacques2Marais in "I want to wash my car. The car wash is 50 meters away. Should I walk or drive?"

Jacques2Marais — Mon, 16 Feb 2026 11:07:38 +0000

An LLM's take on this thread (GPT 5.1):

""" - Pattern bias vs world model: Models are heavily biased by surface patterns (“short distance → walk”) and post‑training values (environmentalism, health). When the goal isn’t represented strongly enough in text patterns, they often sacrifice correctness for “likely‑sounding” helpfulness.

- Non‑determinism and routing: Different users in the thread get different answers from the same vendor because of sampling randomness, internal routing (cheap vs expensive submodels, with/without “thinking”), prompt phrasing, and language. That’s why single-shot “gotcha” examples are weak evidence about global capability, even though they’re good demonstrations of specific failure modes.

- Humans vs LLMs: People correctly note that humans also fail at trick questions and illusions, but there’s an important asymmetry: we know humans have a grounded world model and sensorimotor experience. With LLMs, we only have behavior. Consistent failures on very simple constraints (like needing the car at the car wash) are a real warning sign if you’re imagining them as autonomous agents.

- Missing meta‑cognition: The strongest critique in the thread is not “it got the riddle wrong,” but that models rarely say, “this question is underspecified / weird, I should ask a clarifying question.” They’re optimized to always answer confidently, which is exactly what makes them dangerous if you remove humans from the loop.

- Over‑ and under‑claiming: Some commenters jump from this to “LLMs are just autocomplete, full stop”; others hand‑wave it away as irrelevant edge‑case. Both are overstated. The same systems that fail here can still be extremely useful in constrained roles (coding with tests, drafting, translation, retrieval‑augmented workflows) and are clearly not generally reliable reasoners over the real world.

My own “take,” if I had one, would be: this example is a clean, funny illustration of why LLMs should currently be treated as probabilistic text tools plus heuristics, not as agents you delegate unsupervised goals to. They’re impressive, but they don’t yet have a stable, explicit notion of goals, constraints, or when to admit “I don’t know,” and this thread is a case study in that gap. """

New comment by Jacques2Marais in "Lost Soviet Moon Lander May Have Been Found"

Jacques2Marais — Mon, 16 Feb 2026 09:13:35 +0000

'“One of them is wrong,” an expert said.'

New comment by Jacques2Marais in "I want to wash my car. The car wash is 50 meters away. Should I walk or drive?"

Jacques2Marais — Mon, 16 Feb 2026 08:17:36 +0000

You would be surprised, however, at how much detail humans also need to understand each other. We often want AI to just "understand" us in ways many people may not initially have understood us without extra communication.

Show HN: Host OpenClaw with native template and multi-agent support

Jacques2Marais — Wed, 11 Feb 2026 15:47:35 +0000

Hi HN,

I built ClawClaw because currently OpenClaw hosting still requires a lot of manual setup for certain useful features. The plan with ClawClaw is to make OpenClaw ready out-of-the-box for all kinds of different use cases with minimal manual configuration.

Enjoy and let me know what you think :)

Comments URL: https://news.ycombinator.com/item?id=46976408

Points: 1

# Comments: 0