Hacker News: WiSaGaN

New comment by WiSaGaN in "Why isn't the industry freaking out about DeepSeek 4.1 Flash?"

WiSaGaN — Fri, 09 Oct 2026 10:41:29 +0000

Deepseek v4.1 flash is my baseline model to use at original provider's api. The issue is, for my personal use, it's cheap enough that i don't need it to go cheaper compare to the time I spent using it. And it's already a very capable model in dealing everyday simple things. For sure, for research level questions or large scale coding projects, I would want to use frontier model. But more and more daily tasks can be done now just using pi with deepseek api directly without thinking about much.

New comment by WiSaGaN in "The Mathocalypse"

WiSaGaN — Thu, 08 Oct 2026 14:27:01 +0000

This is with the unreleased model, which is probably significantly larger (thus costlier) than astra.

New comment by WiSaGaN in "DeepSeek v4.1 Flash"

WiSaGaN — Thu, 10 Sep 2026 13:23:35 +0000

This is definitely not on par with GPT-6 astra. Not with GPT-5.6 sol either. But probably will set as a new baseline for modern API based LLM because it's so cheap.

New comment by WiSaGaN in "Our decision on Cursor following its acquisition by SpaceX"

WiSaGaN — Sat, 29 Aug 2026 08:57:48 +0000

It's one of the things you need to do if you want your company to later become the only company in the world. They also promised that they would treat you nicely afterwards after they get what they want.

New comment by WiSaGaN in "Qwen 3.8-Flash-Next releasing tomorrow (125B a6B)"

WiSaGaN — Tue, 25 Aug 2026 16:15:31 +0000

You probably meant Qwen3.8-35B-A3B. But judging from some of the words from their team, it seems unlikely unfortunately.

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

WiSaGaN — Fri, 21 Aug 2026 14:38:37 +0000

That would've been a very strange arrangement.

New comment by WiSaGaN in "OpenAI’s head of ethics leaves less than a year after joining"

WiSaGaN — Tue, 11 Aug 2026 15:02:42 +0000

It's unlikely HugginFace colluded given they specifically cited they used open models to save them from the hacking.

New comment by WiSaGaN in "How Claude marks AI-generated content"

WiSaGaN — Tue, 11 Aug 2026 11:57:49 +0000

My guess is that they will later "reveal" some "violations" but provide little evidence citing proprietary algorithm.

New comment by WiSaGaN in "AI's top startups are barely publishing their research"

WiSaGaN — Thu, 30 Jul 2026 13:02:04 +0000

Top startups in China publish a lot?

New comment by WiSaGaN in "PGSimCity - How PostgreSQL Works"

WiSaGaN — Mon, 27 Jul 2026 10:56:54 +0000

I assume this is done with the help of AI? I have done similar vibe project that explaining catastrophic forgetting with the help of AI. It's just so satisfying now that if you really want to learn something, you can always do it with the help of AI. Before, it's hard to find good resources, now it's not a problem, the issue now becomes one's own focus and agency.

New comment by WiSaGaN in "DeepSeek pause fundraise after comments on compute gap to US leaked (transcript) [pdf]"

WiSaGaN — Sun, 26 Jul 2026 01:58:29 +0000

U.S. policymakers believe that even if the gap is small—like six months to a year—whoever reaches AGI first (whatever that means) could gain such an overwhelming advantage over their perceived adversary that it would effectively kneecap them. (You can look at the kinds of things they mention—cyber, WMDs—to get a sense of what they mean.) Jensen Huang disagrees and has said AI is a marathon.

New comment by WiSaGaN in "Show HN: Trust – Coding Rust like it's 1989"

WiSaGaN — Thu, 07 May 2026 10:12:11 +0000

Maybe I should start a project rewriting pctools 5.0 in rust!

New comment by WiSaGaN in "The Claude Code Source Leak: fake tools, frustration regexes, undercover mode"

WiSaGaN — Wed, 01 Apr 2026 12:11:55 +0000

Yeah, this seems like a personal choice, which does work out given the current result.

New comment by WiSaGaN in "Day 1 of ARC-AGI-3"

WiSaGaN — Fri, 27 Mar 2026 06:01:50 +0000

Harness is fine. I think people here are arguing what provided here to take the test is not harness.

New comment by WiSaGaN in "Detecting and Preventing Distillation Attacks"

WiSaGaN — Mon, 23 Feb 2026 22:50:27 +0000

This violates the ToS, but I don't think it's distillation. Distillation requires knowing the logits, which current API does not provide. This is just synthetic data generation. Anthropic definitely knows the difference.

New comment by WiSaGaN in "Google restricting Google AI Pro/Ultra subscribers for using OpenClaw"

WiSaGaN — Mon, 23 Feb 2026 12:41:44 +0000

Google has gigantic power over its users. Consider that for some reason, Google banned your gmail account, which you are using for large number of logins for different essential services.

New comment by WiSaGaN in "Consistency diffusion language models: Up to 14x faster, no quality loss"

WiSaGaN — Fri, 20 Feb 2026 09:08:05 +0000

I think diffusion makes much more sense than auto-regressive (AR) specifically in code generation comparing to chatbot.

New comment by WiSaGaN in "Improving 15 LLMs at Coding in One Afternoon. Only the Harness Changed"

WiSaGaN — Fri, 13 Feb 2026 06:59:32 +0000

Great article. To me, this highlights a key question in the era of rapidly advancing machine intelligence: if we know machine intelligence is progressing, what is more valuable to build for? As humans, we still find many tools useful even when doing knowledge work. For instance, a calculator. Sure, a smart person can perform calculations in their head, but it’s much easier to teach everyone how to use a calculator, which is 100% reliable in its intended domain.

In this era, we should build these kinds of tools for problems we know are straightforward ones you can’t get smarter than, even as intelligence continues to advance. Using tools like "bash" or command-line interfaces originally designed for humans is a good initial approach, since we can essentially reuse much of what was built for human use. Later, we can optimize specifically for machines, either accounting for their different cognitive structures (e.g., the ability to memorize extremely long contexts compared to humans) or adapting to the stream-based input/output patterns of current autoregressive token generators.

Eventually, I believe machine intelligence will build their own tools based on these foundations, likely a similar kind of milestone to when humans first began using tools.

New comment by WiSaGaN in "Gemini 3 Deep Think"

WiSaGaN — Fri, 13 Feb 2026 05:06:20 +0000

Yes, agentic-wise, Claude Opus is best. Complex coding is GPT-5.x. But for smartness, I always felt Gemini 3 Pro is best.

New comment by WiSaGaN in "Prism"

WiSaGaN — Wed, 28 Jan 2026 13:13:22 +0000

OpenAI has a former NSA director on its board. [1] This connection makes the dilution of the term "PRISM" in search results a potential benefit to NSA interests.

[1]: https://openai.com/index/openai-appoints-retired-us-army-gen...