Hacker News: DonsDiscountGas

New comment by DonsDiscountGas in "Reverse engineering Gemini's SynthID detection"

DonsDiscountGas — Thu, 09 Apr 2026 22:05:25 +0000

If you want to make an AI generated image but don't want other people to know that it's AI, the most obvious solution is to not use Gemini. Synth ID is watermarking. It's only ever going to be useful to good actors, who want an AI generated image and aren't trying to hide the fact that it's AI generated.

New comment by DonsDiscountGas in "Newly created Polymarket accounts win big on well-timed Iran ceasefire bets"

DonsDiscountGas — Thu, 09 Apr 2026 12:46:33 +0000

The value of that to the public is pretty minimal though, seeing as that private information becomes public 1-4 hours before it would've been publicized in the more traditional way. And the signal itself is extremely noisy.

New comment by DonsDiscountGas in "I've sold out"

DonsDiscountGas — Wed, 08 Apr 2026 12:12:43 +0000

Or fork it?

New comment by DonsDiscountGas in "I've sold out"

DonsDiscountGas — Wed, 08 Apr 2026 12:12:07 +0000

Pi isn't

New comment by DonsDiscountGas in "Project Glasswing: Securing critical software for the AI era"

DonsDiscountGas — Wed, 08 Apr 2026 12:06:54 +0000

Everybody remembers the fable of the boy who cried wolf and how he died at the end. Left out of the story is the multiple other villagers who died of starvation because their flock of sheep was eaten. So because they didn't want to feel like suckers. Tuning out completely because of the existence of false positives is not a good choice.

New comment by DonsDiscountGas in "In Japan, the robot isn't coming for your job; it's filling the one nobody wants"

DonsDiscountGas — Mon, 06 Apr 2026 18:41:10 +0000

Mechanical dishwashers have existed for awhile. I guess they only do half the job and but we've had them for decades

New comment by DonsDiscountGas in "I stopped hitting Claude's usage limits – things I changed"

DonsDiscountGas — Mon, 06 Apr 2026 12:17:07 +0000

Also use /compact. Slightly easier than copy pasting into new chat. But I forget not everybody uses Claude code

New comment by DonsDiscountGas in "AI that copied musical artist files copyright claim against artist [updated]"

DonsDiscountGas — Sun, 05 Apr 2026 08:46:25 +0000

Doubtful. People seem to be moving in the opposite direction since they hate AI so much.

New comment by DonsDiscountGas in "CC leak: skills are better than I thought"

DonsDiscountGas — Thu, 02 Apr 2026 21:04:15 +0000

Grandiose language but it's not wrong. The short article was worth reading IMHO

New comment by DonsDiscountGas in "Trump fires Pam Bondi as attorney general"

DonsDiscountGas — Thu, 02 Apr 2026 20:37:04 +0000

I suspect many do, but you can grift a lot before that happens

New comment by DonsDiscountGas in "Full network of clitoral nerves mapped out for first time"

DonsDiscountGas — Mon, 30 Mar 2026 16:42:11 +0000

> her colleagues used high-energy X-rays to create 3D scans of two female pelvises that had been donated through a body donor organ programme.

Would be really interesting to get a few dozen of these and map out the variability. Gotta start somewhere though

New comment by DonsDiscountGas in "ArXiv declares independence from Cornell"

DonsDiscountGas — Fri, 20 Mar 2026 15:10:19 +0000

Considering the value and prominence of arxiv to the world, this seems low to me. Although more importantly the rest of the staff needs to be well paid too, and if that's the ceiling its a bit concerning. It's crazy to me that people thought this was too high.

New comment by DonsDiscountGas in "I'm OK being left behind, thanks"

DonsDiscountGas — Fri, 20 Mar 2026 15:05:28 +0000

I started working in AI/ML about ten years ago. Reasonably early. Today, professionally and financially I'm doing about as well as a typical programmer. I find the field interesting so I have no regrets but I tend to agree with OP.

New comment by DonsDiscountGas in "Show HN: What was the world listening to? Music charts, 20 countries (1940–2025)"

DonsDiscountGas — Fri, 13 Mar 2026 14:28:27 +0000

I would love more granular data, like state or zip code. It would help settle a decade old (and zero stakes) dispute I have with a friend. I'm sure that's your top priority, so if you could get on that, that would be great

New comment by DonsDiscountGas in "ATMs didn't kill bank teller jobs, but the iPhone did"

DonsDiscountGas — Thu, 12 Mar 2026 18:59:14 +0000

I'm the same way but we're both posting on hacker news. Many people prefer phones

New comment by DonsDiscountGas in "ATMs didn't kill bank teller jobs, but the iPhone did"

DonsDiscountGas — Thu, 12 Mar 2026 18:57:17 +0000

I've had the same thought. The only major difference that I can think of is the built-in camera making check deposits easier. It may also be that people were just generally using computers more and using the internet more over this same time period, although a lot is that because of smartphones

New comment by DonsDiscountGas in "The Gervais Principle, or the Office According to “The Office” (2009)"

DonsDiscountGas — Tue, 10 Mar 2026 17:33:27 +0000

Hot take: being clueless is better than these essays make it out to be. The examples are all really socially annoying people (Michael, Dwight) but I've known some pretty nice and pleasant middle managers who had generally great lives. They probably could've gotten all of that with less work but perfectly hitting the Pareto frontier is quite difficult.

New comment by DonsDiscountGas in "Debian decides not to decide on AI-generated contributions"

DonsDiscountGas — Tue, 10 Mar 2026 17:17:50 +0000

It's absolutely not a straw man, because OP and people like OP will be affected by any policy which limits or bans LLMs. Whether or not the policy writer intended it. So he deserves a voice.

New comment by DonsDiscountGas in "Is legal the same as legitimate: AI reimplementation and the erosion of copyleft"

DonsDiscountGas — Tue, 10 Mar 2026 00:35:15 +0000

Trademarks don't, patents do. Different kind of creativity but still.

New comment by DonsDiscountGas in "Phi-4-reasoning-vision and the lessons of training a multimodal reasoning model"

DonsDiscountGas — Mon, 09 Mar 2026 16:34:55 +0000

I'm very happy to read about this progress but I don't find it particularly surprising. The big labs optimize for accuracy/high scores on benchmarks first; I automatically expect that (with some research effort) a model with 100x few parameters can achieve the same scores.