Hacker News: Tiberium

New comment by Tiberium in "Minimax M3"

Tiberium — Thu, 28 May 2026 22:55:10 +0000

Amazing benchmarks!

> Dimensions: Height 490mm, Width 206mm, Depth 445mm

> Flow Rate: Min < 3 litres per hr, Max 55 litres per min.

> Water Pressure: Min 1 bar, Max 8 bar*

> Regeneration Time: 12 mins approx

> Water 17 litres

New comment by Tiberium in "Xiaomi MiMo Token Plan is Now Globally Available"

Tiberium — Wed, 27 May 2026 02:28:35 +0000

ChatGPT (the setting is shared with Codex) and Claude (shared with Claude Code) also have sharing enabled by default, so why aren't they cheaper?

New comment by Tiberium in "We reduced a real Node.js production Docker image from 1.2GB to 78MB"

Tiberium — Sat, 23 May 2026 22:46:18 +0000

Musl is much less optimized than glibc, random example: https://andygrove.io/2020/05/why-musl-extremely-slow/

New comment by Tiberium in "JWT is a scam and your app doesn't need it"

Tiberium — Sat, 23 May 2026 20:21:32 +0000

Supporting evidence isn't what makes Pangram mark something as AI-generated, though. And I did call out that in my post specifically. But the post is still 100% generated, it has a distinct Claude-adjacent LLM style.

Some of the extremely obvious examples:

> The pitch is: the server signs it, the client carries it, every subsequent request only needs a signature verification — no database round-trip.

> You can't. That's the answer. The token is valid until it expires, full stop.

> A single opaque token, looked up in Redis with Postgres as the backing store, gives you the same security in one line of middleware. No refresh. No second token. No retry loop. Nothing.

> With opaque tokens this is just… how it works. No mismatch, no hidden tax, no "did they implement the checks correctly" question to lose sleep over.

New comment by Tiberium in "JWT is a scam and your app doesn't need it"

Tiberium — Sat, 23 May 2026 18:56:31 +0000

This article is completely AI-generated, anyone who reads LLM output often can confirm that. And Pangram confirms it as well. It's one of the more reliable LLM detectors and has an extremely low false positive rate, although in this case it marked Markdown parts too (they're irrelevant since it's a web article, not a normal post).

https://www.pangram.com/history/c68104cd-6072-4f7a-850b-e534...

New comment by Tiberium in "JWT is a scam and your app doesn't need it"

Tiberium — Sat, 23 May 2026 18:24:44 +0000

Why post this AI-generated article? Just to ragebait?

Linux LPE via GRO managed-frag UAF (io_uring SEND_ZC and veth)

Tiberium — Fri, 22 May 2026 20:37:41 +0000

Article URL: https://gist.github.com/lcfr-eth/2566a5cef312c94a5ff8d62fa417955f

Comments URL: https://news.ycombinator.com/item?id=48241304

Points: 3

# Comments: 0

DeepSeek makes the V4 Pro price discount permanent

Tiberium — Fri, 22 May 2026 15:59:26 +0000

> (3) The deepseek-v4-pro model API pricing will be officially adjusted to 1/4 of the original price after the 75% discount promotion ends on 2026/05/31 15:59 UTC.

https://x.com/deepseek_ai/status/2057854261699195173

Related ongoing thread:

DeepSeek reasonix, DeepSeek native coding agent with high caching and low cost - https://news.ycombinator.com/item?id=48256953 - May 2026 (135 comments)

Comments URL: https://news.ycombinator.com/item?id=48237663

Points: 620

# Comments: 549

DeepSeek makes the V4 Pro price discount permanent

Tiberium — Fri, 22 May 2026 13:45:07 +0000

Article URL: https://api-docs.deepseek.com/quick_start/pricing

Comments URL: https://news.ycombinator.com/item?id=48235792

Points: 1

# Comments: 1

New comment by Tiberium in "Remove-AI-Watermarks – CLI and library for removing AI watermarks from images"

Tiberium — Wed, 20 May 2026 02:11:22 +0000

It's not upscaling for NB2, 4K outputs are very different from 1K, and output tokens count is also different.

New comment by Tiberium in "Remove-AI-Watermarks – CLI and library for removing AI watermarks from images"

Tiberium — Tue, 19 May 2026 22:55:25 +0000

I mostly agree about the justification in the repo being wrong, but wanted to engage about this point:

> Image generators cannot insert watermarks into things they did not generate

It's actually very easy to take a real image, ask Gemini/ChatGPT to modify some tiny part of it (could be something as silly as lighting/shadow/etc), and often the resulting image will be detected by their watermarking tools. This way you can easily present any real image as AI-generated.

New comment by Tiberium in "Remove-AI-Watermarks – CLI and library for removing AI watermarks from images"

Tiberium — Tue, 19 May 2026 22:52:50 +0000

This is a bit misleading as for Gemini it only properly removes the visible watermark. To remove SynthID it has to regenerate the image at low noise with SDXL, which will likely destroy a lot of small details, plus won't work for higher res properly (NB2 and GPT Image 2 support up to 4K image outputs)

New comment by Tiberium in "OpenAI Adopts Google's SynthID Watermark for AI Images with Verification Tool"

Tiberium — Tue, 19 May 2026 20:14:07 +0000

I still don't think there's a single GitHub repo that actually removes real SynthID watermarks from Nano Banana 2/NBPro outputs. Most of them are just some research projects that haven't achieved this. The only methods so far I've seen are weird tricks with transparency/overlaying the original image if you're using edits, and also using a diffusion model to regenerate the NB-generated image at low noise levels, but this also modifies the original.

New comment by Tiberium in "Gemini 3.5 Flash"

Tiberium — Tue, 19 May 2026 18:46:44 +0000

Please delete/edit your AI-written and factually wrong post.

New comment by Tiberium in "Show HN: Id-agent – Token efficient UUID alternative for AI agents"

Tiberium — Tue, 19 May 2026 11:52:20 +0000

Is this just a reinvented humanhash?

New comment by Tiberium in "Haiku OS runs on M1 Macs now"

Tiberium — Mon, 18 May 2026 18:52:18 +0000

What do you mean by that? It seems like there's Ruby in the packages:

https://depot.haiku-os.org/#!/pkg/ruby/haikuports/haikuports...

New comment by Tiberium in "1024000^2 Blocks, 2B2T Minecraft Server World Download Project, and Discoveries"

Tiberium — Mon, 18 May 2026 14:56:09 +0000

2B2T really has an extremely interesting and long history, and some of the most interesting exploits/security vulnerabilities found just for it, for example https://2b2t.miraheze.org/wiki/Nocom where the attackers had a blunt server DoS and through it influenced the server owner to report it to PaperMC, and the fix for the DoS gave them ability to track other players just as they had hoped.

> In July 2018, 0x22 and Babbaj created a coordinate exploit, using the groundwork laid out in the lag exploit. The two theorized that, if the server didn't return a response for unloaded chunks, but returned a response for loaded chunks, the rough location of players in 2b2t could be approximated. However, prplz's patch returned a response regardless of whether a chunk was loaded or unloaded, requiring a second patch to Paper that would only return a response if the chunk was loaded.

> Knowing that the issue would be resolved if Hausemaster reported it to Paper, likely through the method they laid out, 0x22 and Babbaj began intentionally, repeatedly, and blatantly sending CPacketPlayerDigging packets, causing the Paper watchdog process to output a stack trace, which included the line added by prplz.

New comment by Tiberium in "OpenClaw Creator Spent $1.3M on OpenAI Tokens in 30 Days"

Tiberium — Sat, 16 May 2026 17:12:02 +0000

I did mention that you could use ~60 $200 Codex accounts to emulate his workflow without /fast, or 2.5x that if you used /fast. Not $1.3M

New comment by Tiberium in "OpenClaw Creator Spent $1.3M on OpenAI Tokens in 30 Days"

Tiberium — Sat, 16 May 2026 12:37:03 +0000

This is quite a misleading title because this is the raw API cost, but he (obviously) has unlimited usage as an OpenAI employee. Moreover, if you use e.g. the $200 Codex sub, you get about ~$5k-$6k monthly API usage if you spend every week of your usage, if not more, which shows that the raw API cost is not how much it (likely) costs to OpenAI, unless they're subsidizing all this.

He did clarify that it was with fast mode. Without fast mode it'd "only" be $300k in raw API cost, or ~60 $200 Codex subscriptions.

QEMUtiny - QEMU escape vulnerability if cxl is used

Tiberium — Fri, 15 May 2026 14:38:21 +0000

Article URL: https://github.com/v12-security/pocs/tree/main/qemu

Comments URL: https://news.ycombinator.com/item?id=48149233

Points: 8

# Comments: 0