Hacker News: johndough

New comment by johndough in "Free SQL→ER diagram tool, runs in the browser, nothing uploaded"

johndough — Sun, 14 Jun 2026 09:44:52 +0000

Probably not that important in practice. Firefox allows 2^20 - 4 and Chrome allows 2100000 characters. Also, 8000 characters already allows for an unreasonable amount of SQL and could be extended even further with compression. And if that should not be enough, the website already supports JSON exports. All in all, this seems like a worthwhile tradeoff for not having to store anything.

New comment by johndough in "Free SQL→ER diagram tool, runs in the browser, nothing uploaded"

johndough — Sun, 14 Jun 2026 09:27:18 +0000

In academia, this is called "planar embedding" and can be computed in O(V) where V is the number of vertices of the graph.

However, there are graphs that do not allow planar embeddings (e.g. K_5 or K_3,3, see https://en.wikipedia.org/wiki/Planar_graph).

In this case, you'll probably want to look into heuristics that produce a low number of crossings and little distortion when new vertices are added.

New comment by johndough in "Statement on US government directive to suspend access to Fable 5 and Mythos 5"

johndough — Sat, 13 Jun 2026 12:04:27 +0000

Feel free to replace "racism" with "discrimination" if you prefer. English is not my first language and the minutiae elude me.

New comment by johndough in "Statement on US government directive to suspend access to Fable 5 and Mythos 5"

johndough — Sat, 13 Jun 2026 08:17:12 +0000

The recent MiMo-V2.5-Pro-UltraSpeed can be served from 8 GPUs, which is certainly within the reach of sophisticated on-prem setups. https://mimo.xiaomi.com/blog/mimo-tilert-1000tps

New comment by johndough in "Statement on US government directive to suspend access to Fable 5 and Mythos 5"

johndough — Sat, 13 Jun 2026 07:59:15 +0000

MiniMax M3 weights have already been released: https://huggingface.co/MiniMaxAI/MiniMax-M3

New comment by johndough in "Statement on US government directive to suspend access to Fable 5 and Mythos 5"

johndough — Sat, 13 Jun 2026 07:49:06 +0000

As a foreigner, the casual racism against red-haired people in the US and UK always baffled me.

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

johndough — Mon, 08 Jun 2026 22:41:06 +0000

Cerebras got lucky that they IPOed last month instead of now.

New comment by johndough in "Meta's ships facial recognition on smart glasses"

johndough — Thu, 04 Jun 2026 20:53:50 +0000

I recall that Facebook asked me to identify faces of my friends in order to "verify" myself about 15 years ago. Do you know whether Facebook still stores that data?

New comment by johndough in "ArcBrush – Node-based 2D image editor"

johndough — Sat, 23 May 2026 11:40:49 +0000

Likely just a coincidence. There is a huge number of node-based image editors: https://www.google.com/search?q=node-based+image+editor&udm=...

New comment by johndough in "College students drown out AI-praising commencement speeches with boos"

johndough — Wed, 20 May 2026 14:24:27 +0000

Goomba fallacy https://en.wiktionary.org/wiki/Goomba_fallacy

New comment by johndough in "The last six months in LLMs in five minutes"

johndough — Tue, 19 May 2026 21:46:42 +0000

Fair enough. I am optimistic that 2 and 3 will be doable eventually.

New comment by johndough in "The last six months in LLMs in five minutes"

johndough — Tue, 19 May 2026 18:55:42 +0000

I've got a few. Pick whichever you like.

Factor 135066410865995223349603216278805969938881475605667027524485143851526510604859533833940287150571909441798207282164471551373680419703964191743046496589274256239341020864383202110372958725762358509643110564073501508187510676594629205563685529475213500852879416377328533906109750544334999811150056977236890927563 in less than 24 hours.

Come up with a way to sample from LLMs such that they can tell funny jokes. The jokes should not be recited jokes from elsewhere.

Implement a CUDA kernel that achieves optimal efficiency for PyTorch-like conv2d for "reasonable" shapes/strides/dilations/groups. (This task is the closest to being solved by LLMs, but they usually get stuck somewhere doing stupid things instead of considering more advanced optimization methods and still need a human to push them along).

New comment by johndough in "Project Gutenberg – keeps getting better"

johndough — Fri, 15 May 2026 20:45:35 +0000

It looks like the issue was that, in Italy, copyright expires 70 years after the death of the author or the first translator of a work.

New comment by johndough in "If AI writes your code, why use Python?"

johndough — Tue, 12 May 2026 20:05:53 +0000

Yes, single RTX 3090 with this model https://huggingface.co/llmfan46/Qwen3.6-35B-A3B-uncensored-h... following these https://huggingface.co/havenoammo/Qwen3.6-35B-A3B-MTP-GGUF instructions (should add "-j 8" to last cmake command for parallel build) and llama-server with --reasoning off

Note that the MTP PR https://github.com/ggml-org/llama.cpp/pull/22673 is still under development, so things might be broken.

New comment by johndough in "If AI writes your code, why use Python?"

johndough — Tue, 12 May 2026 16:07:15 +0000

> it gets caught in circles more often then say a 1t parameter model does.

I've found that the Q5+ quants are less loopy than Q4. Still not perfect, but noticeably better.

> reasonable enough tokens per second

The speed has been amazing. I've been running the recent llama.cpp MTP branch with an uncensored variant of Qwen3.6-35B-A3B on my RTX 3090 over 170 tokens per second and it was able to turn a buffer overflow into a reliable shell exploit in just a few seconds (with reasoning disabled). Still a bit loopy though. Hopefully, the Qwen team will pay more attention to those looping issues. It feels like their models are especially susceptible.

New comment by johndough in "If AI writes your code, why use Python?"

johndough — Tue, 12 May 2026 11:11:55 +0000

While Qwen3.6 27B and 35B-A3B are very good, I am skeptical about them being that good. I think another factor is at play here.

The Qwen3.6 models have memorized some common games. For example, if you ask it to create an index.html with a snake game, it will generate almost the same high quality snake game every time. The relatively low success rate of 25% but high average percentile of almost 100% for one-shot coding in Python suggests that the model is extremely good at few tasks.

New comment by johndough in "A recent experience with ChatGPT 5.5 Pro"

johndough — Sat, 09 May 2026 07:39:24 +0000

API pricing for Claude is about an order of magnitude more expensive than subscriptions (numbers: https://she-llac.com/claude-limits). But it may be worth it with DeepSeek V4 Pro, which is currently on discount.

New comment by johndough in "A recent experience with ChatGPT 5.5 Pro"

johndough — Sat, 09 May 2026 07:17:51 +0000

At my university, everyone had to pay their AI subscriptions out of their own pocket, until a communal AI service was introduced recently. It took 2 years to set up and only serves gpt-oss-120b, so everyone is still using other services. But at least some admin can scatter the word "AI" all over the university's website now and has an excuse to reject any requests for AI subscriptions because "we already have AI".

New comment by johndough in "DeepSeek V4 – almost on the frontier"

johndough — Sat, 02 May 2026 13:21:45 +0000

> I'm running this locally.

Impressive! What is your setup? Are you running the full DeepSeek V4 Pro, or V4 Flash?

New comment by johndough in "Mistral Medium 3.5"

johndough — Wed, 29 Apr 2026 17:11:28 +0000

You mentioned "pareto-competitive", and EXAONE certainly was that. The statement that the "above countries have never had a model that is even close to being so" is simply too broad.