geuis
geuis on Hacker News
Location: San Francisco
Remote: Yes
Willing to relocate: No
Technologies: TypeScript/JS, React, node.js, Redis, Postgres, pgvector, LLM agents, Cloud (AWS/GCP/Azure), ci/cd pipelines, other tech
Résumé/CV: on request
Email: [email protected]
Sew a million seeds, see what grows.
I've been thinking about this for about a year. In my head I've called it "context bombs", though that may not be super accurate.
Think back to the movie The Manchurian Candidate. Produce highly capable LLM models then give them away for free. Train them with key phrases/codes etc to perform nefarious actions later when those codes are present in some piece of ingested context.
There were a couple of papers written on the topic in 2018 (I think, I'll need to find them again) and there's been surprisingly little more work done on the topic since. At least from what I've seen.
I really love the use of local models over subscription services but personally I would never recommend using open weights models from certain sources in a commercial environment.
Posted by
geuis on Sunday October 25, 2026 @11:30PM
from the self-post dept.
Since it was released Opus 5.5 has solved almost all the problems I've had with previous Anthropic models over the last year. On medium thinking level (default in Claude code) it's been remarkably more succinct and really good at understanding my questions and even anticipating my needs. Has been a markedly huge improvement over previous models and even had me commenting to a colleague the...
I'm old enough to have been in the industry when Chrome launched. It was amazing. It was such a relief to move away from IE and all the arcane levels of details I had to remember back then to artisnaly get IE to behave like a moderately modern (for then) browser.
Chrome was fast, did the basic job of correctly rendering pages, and its js runtime was groundbreaking. Node.js was direct outcome of that revolution.
Now these days the only thing keeping me tied to chrome is finger muscle memory. The death of manifest v2 is the 2nd to last nail.
Honestly Opus 5.5 had gotten good enough I've been considering a project to have it build a new browser based on the open source Chrome and keeps the v2 in place. Just customize it specifically for my preferences and call it done. No more unexpected breaking updates. Just a browser that does its job and nothing else.
The problem is I use safari on iOS. Generally not a bad browser. Equivalent and superior to Chrome is certain aspects. I don't really need to run Chrome on my phone. Safari is good enough. However... for a local build only I use maybe it's possible.
Passing opinion, but I hope this isn't indicating a lean in direction to 90's era Apple now that Tim retired. Too many overpriced skews and non-core products that diluted the brand and led to their market shrinking.
so responses like "I've turned on the lights", but you're still standing in the dark. That's my anecdotal level of interactions with any Gemini model.
Direct link in case it's needed.
https://rachelbythebay.com/w/2026/10/08/idle/For some reason the first time I tapped the link it didn't go directly to the article, which confused me. Subsequent times it goes straight through.
Look at it from a training data perspective. If the models' behavior is being trained on a large corpus of text from human interactions, then it should be expected that certain conversational queues will be picked up in the behavior of the models.
So saying please and thanks isn't anthropomorphizing the models, it's more simply using input patterns that get better outputs more aligned to the desired result.
It also has the benefit of reinforcing polite behavior among us, the humans. Side benefit, but it is a thing to consider.
I would definitely be more satisfied if claude code would stop with the "how's the session going" popups. I find it very distracting.
Back a few years ago with gpt3.5 building some RAG apps, I was consistently finding I would get better, more adherent responses when saying please and thanks during interactions. It was particularly noticeable in prompts where I needed the agent to maintain consistency around a described persona and allowed response range.
I don't think there's anything going on inside other than predictions. That being said, being nice never hurts.
If I'm mean to some people and nice to others, that just means I'm an a-hole. I don't wanna be that. Being kind all around helps me be a better person in the world. And in some cases with the robots, it helps as well.
I appreciate a channel that calls itself out. That actually makes the content a little more interesting, because its not hiding what it is.
(channel name is AI Slop for those not clicking the link.)
The entire site looks like it was generated with Claude.
The site hijacks scrolling.
I scanned through paragraph after paragraph of overbloated and repetitive descriptions of what and how RAG works compared to how "modern agents" need to access documents.
What I never saw at any point was whatever this "alternative to RAG" was supposed to be.
Honestly the entire body of copy feels like it was llm generated.
Location: San Francisco
Remote: Yes
Willing to relocate: No
Technologies: TypeScript/JS, React, node.js, Redis, Postgres, pgvector, LLM agents, Cloud (AWS/GCP/Azure), ci/cd pipelines, other tech
Résumé/CV: on request
Email: [email protected]
I don't understand why people continue to use em-dashes. As far I know, the comment system on HN doesn't offer them. And iOS and android don't make them a default. They aren't something normally used by humans because - is a keyboard option that's easier to use.
The usage of them immediately makes your commentary suspect. Either you aren't using the standard web interface to make a comment, you're using and odd 3rd party client, or its LLM generated.
This isn't peer reviewed, just a press release. 10 other "papers" will be published in the next 2 weeks counterfinding anything in this post.
Science moves forward yet again with a few backward steps.
Fast download for me. But immediately denied. Doesn't run on Mac intel hardware.
Location: San Francisco
Remote: Yes
Willing to relocate: No
Technologies: TypeScript/JS, React, node.js, Redis, Postgres, pgvector, LLM agents, Cloud (AWS/GCP/Azure), ci/cd pipelines, other tech
Résumé/CV: on request
Email: [email protected]
Cyberpunk 2077:
> Gameplay Type: First Person Shooter Role Playing Game
Summary: A long anticipated release with eight years of development time, the game’s launch has been plagued with bugs and poor performance as well as decidedly mixed reviews. Regardless, the title offers gamers the largest open world cyberpunk universe thus far, so despite the rather uninspired art design and a plot that brings nothing new to the genre, this game is the best available to allow players to immerse themselves in a cyberpunk setting.
This is a very dated and seriously opinionated review.
Wonderful piece of art. Excellent callback to what the web used to look like when I was a teenager.
I don't understand the problem space. Meaning I don't understand what this solves.
My use case, even on phone:
View HN
Right click link to open in new tab
OR: tap/hold a link then open I a new window.
What problem is this solving?
Good question! I think I'll use that as a counter question next time I interview. Ask a technical member of the team I'm interviewing with how they handled that situation.
I finally made progress today on a deeply intractable problem that Sonnet was just incapable of solving despite hours of experimentation and multiple attempts. I swapped over to Opus in a new handoff session with the expectation of using at least one 5 hour session. It wasn't quick and took several hours but eventually Opus dialed in on several cascading base failures resulting from how the base vision model was being converted to coreml. It eventually was able to provide a combination of conversion and Swift changes that semi fixed the issue but still didn't solve the core conversion issues.
Then since Anthropic was so kind to provide $100 for Fable credits, I did another handoff to let Fable attack the root issues again. Several more hours and I'm back to seeing the same issues again and Fable is wandering around trying different things. I'm down about $40 of free money and still don't have a solution.
This is where having the human engineer in the loop benefits from a deep understanding of the problem domain. In this case, I don't yet.
I know the high level architecture I'm building, but the deep specifics of how vision models work and how conversion across platforms should be done isn't something I know yet.
So I'm left learning as I go and relying on constant feedback with the models to provide what guidance I can while learning exactly what is being done.
Someone who already knows these architectures would likely be able to get to the solution much faster.
Glad you posted your experience. I'll definitely be keeping my eye out for shenanigans on my own domains.
I've been a long, long term customer of Namecheap as well.
Domain privacy protection is a feature that Namecheap provides for all valid domains. Its included by default at no extra cost.
The poster didn't indicate if they had this feature enabled or not for the domain. It would have prevented the college club person from even seeing their email address to initiate a password reset.
This clearly isn't an answer for NC's customer support personnel and company policies.
But I've been a happy customer for many years and I discourage others from immediately reading other comments and rushing to jump to other registrars without doing your due dilligence.
Remember that in any situation, the people most likely to leave negative comments and reviews are the people that have had genuine bad experiences or feel like they've been slighted, even if unwarranted.
No. Even with the amount of co2 current being dumped into the atmosphere there's not enough total solar energy capture to Venus things up. A few thousand years, absolutely.
To add: desertification and more or less removing the support systems for modern population levels can easily happen with what's going on. But the oceans won't boil first. Everything else will die first.
Lot of excellent information. Thanks. I'm glad someone out there doesn't have arachnophobia. I'm also glad it's not me.
Location: San Francisco
Remote: Yes
Willing to relocate: No
Technologies: TypeScript/JS, React, node.js, Redis, Postgres, pgvector, LLM agents, Cloud (AWS/GCP/Azure), ci/cd pipelines, other tech
Résumé/CV: on request
Email: [email protected]
Doesn't work for my use-case. GroundingDINO is a text to bounding box model. SAM2 supports coordinate based masks (user taps or clicks somewhere in an image), which is what my research app needs.