Hacker News: msdz

New comment by msdz in "GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2"

msdz — Sat, 20 Jun 2026 19:57:57 +0000

> it hallucinates like crazy but looks like its by design to boost benchmarks.

Wasn’t there a discussion around some new-ish benchmark _punishing_ hallucination answers (over not replying at all) recently? Maybe in the not-so-distant future, this “spam replies until one’s correct” strategy won’t be able to game a benchmark much at all anymore.

New comment by msdz in "Want your images back? That'll be $5"

msdz — Wed, 17 Jun 2026 16:33:42 +0000

You forgot about the best part, in terms of the “GDPR threat” effectiveness:

Fines can be up to €20 million or 4% of global revenues…, _whichever is greater._

New comment by msdz in "SpaceX to buy Cursor for $60B"

msdz — Tue, 16 Jun 2026 23:37:03 +0000

You don't think that a $60b valuation is having something worthwhile?

(Only half-joking…)

New comment by msdz in "SpaceX to buy Cursor for $60B"

msdz — Tue, 16 Jun 2026 23:23:31 +0000

Somewhat doubtful. (The first part of your statement, anyway. Of course the difference between abstract "value" and hard, spendable "money" is a thing.)

Like Mr. Hanson said in my sibling comment, some rulers are (or were!) bound to have amassed incredible amounts of resources. For historical/non-present-day examples, consider looking into figures like Jakob Fugger or Mansa Musa.

New comment by msdz in "SpaceX to buy Cursor for $60B"

msdz — Tue, 16 Jun 2026 21:31:08 +0000

Who's to say it won't?

New comment by msdz in "Apple is about to make Hide My Email useless"

msdz — Tue, 16 Jun 2026 20:59:14 +0000

Which has more market pull: Some web site or Apple?

New comment by msdz in "Kimi K2.7-Code: open-source coding model with better token efficiency"

msdz — Fri, 12 Jun 2026 15:29:52 +0000

> If you ask Claude in Chinese to introduce itself, it will claim it's Kimi :)

That's a funny anecdote, buut I'm not able to reproduce. Where/how/when did you get this, or hear about it? It might've been patched by now, at least that's the feel I get from my limited testing.

Using bare aichat [1] with no system prompt and no temperature nor top_p (and I'm truncating the response after the first line that contains the name the model gave, because the point has been made clear by then), and with the same prompt (approx. "Introduce yourself!") every time:

Claude Sonnet 4.5:

> 请做个自我介绍！

你好！我是Claude，一个由Anthropic公司开发的AI助手。 […]

Claude Haiku 4.5:

> 请做个自我介绍！

# 你好！

我是 *Claude*，一个由 Anthropic 公司开发的 AI 助手。

Claude Opus 4.5:

> 请做个自我介绍！

# 你好！

我是 *Claude*，由 Anthropic 公司开发的 AI 助手。

Claude Opus 4.6:

> 请做个自我介绍！

# 你好！我是 Claude

Claude Opus 4.7:

> 请做个自我介绍！

你好！我是 Claude，由 Anthropic 公司开发的人工智能助手。很高兴认识你！

Claude Opus 4.8:

> 请做个自我介绍！

你好！我是 Claude，由 Anthropic 公司开发的人工智能助手。

Claude Fable 5:

> 请做个自我介绍！

# 自我介绍

你好！很高兴认识你！

我是 *Claude*，由 Anthropic 开发的 AI 助手。 [2]

I don't see a Kimi mention, unfortunately. :-)

[1] https://github.com/sigoden/aichat

[2] This model really is noticeably more verbose even with supposed-to-be-brief responses huh, lol

New comment by msdz in "Build a Basic AI Agent from Scratch: Long Task Planning"

msdz — Fri, 12 Jun 2026 14:56:40 +0000

That is in fact a better explanation due to bringing up different reasons (zero cost to host as you mentioned, vs. network/visibility out-of-the-box in the linked comment).

New comment by msdz in "Kimi K2.7-Code: open-source coding model with better token efficiency"

msdz — Fri, 12 Jun 2026 14:52:51 +0000

> China is a communist country with elements of capitalistic markets baked in.

While I get the point you're making (it should be pretty obvious to anyone who's held a newspaper), I think it's important regardless to point out that Chinese companies AFAIK aren't worker-owned or -controlled, so you can't exactly call it communism, either. And they obviously do not have a "free market capitalism", as you just discussed.

It's simply a highly authoritarian state then, I guess?

New comment by msdz in "Build a Basic AI Agent from Scratch: Long Task Planning"

msdz — Thu, 11 Jun 2026 15:49:52 +0000

What a strange comment.

The original post is also available at the poster’s own blog [1], so the question is a very valid one. Clearly, “posting articles for free” is a hurdle already cleared by the author.

[1] https://www.ruxu.dev/articles/ai/build-an-ai-agent-planning/

New comment by msdz in "Fooling Go's X.509 Certificate Verification"

msdz — Tue, 09 Jun 2026 07:01:18 +0000

Them darn youngins[1]!

[1] Ken Thompson (over 80), Rob Pike (around 70), Robert Griesemer (over 60)

New comment by msdz in "MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second"

msdz — Mon, 08 Jun 2026 22:39:27 +0000

AFAIK Taalas, the company behind this demo, still only have their initially "hardwarized" model available to test in ChatJimmy, which IIRC is a rather stupid Llama 3ish 8b.

Don't get me wrong though, that demo is still incredibly impressive & makes me very much excited for the hardware-based model era (potentially) ahead.

Once you've experienced those speeds, you really start to think about the whole class of things that becomes possible; massively parallel decode paths, extensive reasoning loops, etc…

New comment by msdz in "The quiet renovation at Bitwarden"

msdz — Tue, 19 May 2026 14:39:20 +0000

What would happen if you lost access to phone and laptop? Is there another "backup" device, or a mechanism to register a new device to your Tailscale network that doesn't require vaultwarden?

New comment by msdz in "Rewrite Bun in Rust has been merged"

msdz — Fri, 15 May 2026 08:13:58 +0000

So the takeaway here is that they scaled to just over $5bn instead of $6.6bn in revenue in just a few years…? Still sounds like plenty demand exists?

New comment by msdz in "Rewrite Bun in Rust has been merged"

msdz — Fri, 15 May 2026 08:08:59 +0000

Push comes to shove, you could probably still ask an LLM to generate transpiler code, if you're so inclined, and then have it fix the remaining "edge cases" afterward, right…?

New comment by msdz in "Everything that went wrong with Claude"

msdz — Mon, 27 Apr 2026 08:16:57 +0000

While I also prefer companies that don't care what the client I'm using is, (part of) the issue is that alternative software was often inefficient with caching, at least up until recently (not sure whether it might have been patched).

OpenClaw heartbeats (essentially idling) could cost single-digit dollar amounts of LLM inference per day, _before any actual user activity_. Another example was IIRC the Pi agent harness sending a new timestamp at the start of every message turn (which sends along the entire chat + tool call history up until that point as context), which of course also invalidates the cache's hash, causing effectively unnecessary re-compute.

I'm not defending Anthropic per se, but just try to picture yourself in their position as a company desperately strapped for any amount of free compute trying to scale up every service as aggressively as they want/have to… And the caching topics are just one potential issue that could occur with "third-party" software. Not that I like it, but of course they'd be quick to ban such behavior in favor of first-party, "guaranteed behaving" customers.

New comment by msdz in "GPT-5.5"

msdz — Thu, 23 Apr 2026 19:23:52 +0000

Such an increase tracks the company's valuation trend, which they constantly, somehow have to justify (let alone break even on costs).

New comment by msdz in "A Roblox cheat and one AI tool brought down Vercel's platform"

msdz — Thu, 23 Apr 2026 13:24:09 +0000

"Writing is nature's way of letting you know how sloppy your thinking is." – Dick Guindon

If your text hasn't undergone that process, it's still sloppy thinking.

New comment by msdz in "GitHub's Fake Star Economy"

msdz — Mon, 20 Apr 2026 09:34:17 +0000

> I look at the starts when choosing dependencies, it's a first filter for sure.

Unfortunately I still look at them, too, out of habit: The project or repo's star count _was_ a first filter in the past, and we must keep in mind it no longer is.

> Good reminder that everything gets gamed given the incentives.

Also known as Goodhart's law [1]: "When a measure becomes a target, it ceases to be a good measure".

Essentially, VCs screwed this one up for the rest of us, I think?

[1] https://en.wikipedia.org/wiki/Goodhart%27s_law

New comment by msdz in "TinyLoRA – Learning to Reason in 13 Parameters"

msdz — Wed, 01 Apr 2026 16:24:06 +0000

This got me thinking, and it might actually even be a comparable amount. Let's estimate 12 years of schooling run at minimum $100,000 per student, at least in the US [1], and then add onto that number whatever else you may do after that, i.e. a bunch more money if paid (college) or "unpaid" (self-taught skills and improvements) education, and then the likely biggest portion for white-collar workers, yet hard-to-quantify, in experience and "value" professional work will equip one with.

Now divide the average SOTA LLM's training cost (or a guess, since these numbers aren't always published as far as I'm aware) by the number of users, or if you wanted to be more strict, the number of people it's proven to be useful for (what else would training be for), and it might not be so far off anymore?

Of course, whether it makes sense to divide and spread out the LLMs' costs across users in order to calculate an "average utility" is debatable.

[1] https://www.publicschoolreview.com/average-spending-student-...