HNHacker News
TopNewBestAskShowJobs

Grimblewald

1,510 karma · joined May 5, 2023

submissionscomments
Grimblewald··on Apple bets cheaper AI will woo small developers
yeah, but gemma models are shit, objectively speaking. I've been building edge AI tools for my lab group, and gemma lineage hallucinates so much it cannot be used at all. We're almost exclusively using qwen models. Given no image with a prompt for OCR to a gemma model, it'll make things up even if told to null fields not present. Qwen will a) follow instructions (placing things not clearly legible into a dedicated 'notes' field), return null for missing items, and get some pretty wild OCR tasks done quite well. It's almost got the opposite problem, I've had to limit how quickly people can submit an OCR ingested label to the DB because people started trusting it to never make mistakes, while gemma required correction on nearly every scan because of things it made up. So gemma didn't win on tok/s, accuracy, grounding of answers, etc. Theres no conceivable spot where it wins, and this was for qwen-vl:4b vs gemma e4b, so qwen model is 1/3 of the size, runs faster, and is far more reliable. So what does gemma really bring to the table?

It's not the cheapest, its worse than cheaper options, etc. All it really brings is the google label.

Grimblewald··on US publishers tell Common Crawl to stop scraping and delete archive
Its like they don't understand the problem common crawl solved rather neatly. You think the skid scrapers are bad? Wait till the competent players lose access to CC.
Grimblewald··on Apple bets cheaper AI will woo small developers
Cheaper isn't as much of a problem as functional.
Grimblewald··on Show HN: Gitdot – a better GitHub. Open-source, written in Rust
FAQ point 5 is pure comedy.
Grimblewald··on Show HN: Gitdot – A better GitHub. Open-source, written in Rust
Bit odd it was missing at all. Makes me wonder how well the core infrastructure you're building on is understood, and how trustworthy any of this really is. If all I'm getting is some vibeslop, with zero expertise or skill added, even basic stuff like this, why shouldnt I just vibe my own better version tailored to my needs exactly? Same headaches as with your app (though likely less give what's on display), only now I dont pay you to give me headaches and I retain control.

The software was never the moat. The skills and experience crystalised within it was, and remains, the product.

Grimblewald··on Show HN: Gitdot – A better GitHub. Open-source, written in Rust
1. It's unfinished 2. Textbook slop dump and run

we can see one big initial push, not too suspect. Then a few superficial updates, and then nothing for months.

Unfinished, broken, and ostentiably abandond due to being an unmaintaible nightmare as vibeslop tends to be.

Grimblewald··on It's Not Just X. It's Y
Burden of proof that ai tools aren't dogshit isn't really on me, so pipe down and use a more reasonable register you pissant. This site, even this thread is filled with evidence. You're in no position to demand anything, especially not something apparently demanded in bad faith. A request for info I'd be happy to meet, but not this.

There are many cases out there of people proving that it does happen. I have personal text not published that do trigger all detectors between 70-100% AI depending on the tool. I wont be sharing these, as "providers" of these "tools" would simply add it to training data and continue to merrily overfit.

Bottom line, transformers regress to the mean like many other models, if you as a person produce output aligned with mean of corpus, you'll trigger detection. More importantly, evading detectors is trivial. Find a corpus of text from an author, get an llm to write a note on style, parlance, habits in writing etc. and then use that voice file to drive outputs from an llm. If the source text didnt register as AI, the new ai output also reliably avoids detection.

So my problem is, detection doesnt work, false positives are fact, so these tools at best offer harm.

A bigger problem¸ undermining your request for proof beyond that which many before me, and including me, burned in an effort to make folks see reason, is that even if i handed you proof on a silver platter you wouldnt understand it. If you could, you'd already understand, because the problem and the math are quite simple. Every example text i've offered in the past now registers 100% human, but many I kept to myself continue to show the same problem, and that never changed. So why would I waste my few remaining tools for sanity checking, when all i can expect from that endevour is losing a tool and shifting nothing in the conversation? No, best I keep that to myself for now.

Grimblewald··on Artificial intelligence is not conscious – Ted Chiang
Always with this topic of articles I find a common theme. No one defines the load-bearing terms and just assumes we all mean the same thing by it, which we know is not the case. That lack of engagement with the topic always has the same effect: the few interesting things said, if any, dont apply to the subject matter claimed generally. Its more of an insight into the authors mind, than the claimed topic. Why can't we all just do the more effective thing, and offer our personal definition for what constitutes conciousness, which is the only thing these articles offer. It would save so much time and make discussion far clearer.
Grimblewald··on I built a vulnerable app and spent $1,500 seeing if LLMs could hack it
I've had some really dumb refusals. Explaining elements of infrared specteoscopy, researching aritifical bud-breaking in agriculture, etc. Anything interesting and non-mainstream is banned. Basically, restricted to answers i'm better of just going to wikipedia for.
Grimblewald··on It's Not Just X. It's Y
Double should map to endash, tripple for em.
Grimblewald··on It's Not Just X. It's Y
Alternatively, no one sounds like an llm, an llm sounds like someone, typically those close to the median of the training corpus. If AI were genuinly capable of novelty, it would be a big deal, tech bros having enough work ethic to design new detectable prose for an llm is a mssive reach and has no real evidence supporting it, else why do tech bros only tackle the easier issues? Things we have massive well labelled corpi for? Why is it never dishwashing and folding laundry?

I put to you, if you see a trope in AI writing it's because that trope appeared in the training corpus. Therefore, sure, being predjudice against it lets you catch some AI, but you'll also flag human outout. I think that may not be worth it in the end.

Grimblewald··on Prolific Wikipedia editors are threatening to go on strike
Well, they've mostly left reddit as well, afaik. It's a few stragglers, people being paid to push agenda, and auto-moderation now. I was reading / moderating reddit for over 4 hours a day, every day, for close to a decade. Heavy handed pro israeli censorship and propaganda has seen me pick up and leave, as i know many others have done. I was already wavering, over reddit's support of astroturfing and shill bots which were obvious, detectable, and reddit refused to do anything about it. Even actively supported it. So with the whole denying /actively supporting genocide thing, it was time to hang it up and leave.

If wikipedia is shutting off avenues for community input, maybe that is running it's course as well.

Its been nice internet, I loved you, and I will never forgive google for what they put into motion.

Grimblewald··on Claude Opus 4.8
I maintian a log of tasks, prompts, related information etc. So i can repeat past workflows verbatim, and I can qualitatively say each model beyond 4.5 has been a regression, and it would not surprise me 4.8 continues the trend. Each iteration has failed at more tasks previously completed succesfully. Right now it flat out refuses to answer many benign chemistry questions, or leans into shilling to hard and ignores non industry funded studies on certain topics. I'm transitioning to deepseek as a reuslt. Cheaper by far and at this stage not strictly speaking less capable.
Grimblewald··on What Apple and Google are doing to push notifications
Your position is that of any normal human. Google is committed to evil however, just look at how playstore notifications are tied to sales spam. Want payment notifivations? Gotta take the ads as well, not seperate toggles, one toggle. Drink liquid shit you tech peasant. Oh? this hostility drove you to f-droid? We'll unilaterally decide every device r belong to us, so we can disable competition we dont approve of. Welcome back to the liquid shit trough, peasant.
Grimblewald··on DuckDuckGo search saw 28% more visits after Google said people love AI mode
AI is like sex. I don't mind it, heck in the right situation I'm quite partial to it. However, if you keep trying to sneak it on me while I sleep, keep trying to slip it into agreements, keep involving people i didnt consent to being involved etc. The whole thing takes on a very ugly vibe. In fact I'm going to grow to be quite hostile toward it. Consent matters, not that I'd expect that lesson to land within silicon valley, second most rapey place in the USA after LA.
Grimblewald··on GPT Guesses Between 1 and 100
6*7=42
Grimblewald··on When does learning from data work (math starting from basic probability)
Unit word per unit meaning is lower than average human, far lower than average technically capable human, the 'hallucinations'/'mistakes' like you said, overall composition and thoughlessness in visual presentation etc. This is unlikely to have seen a human co-author beyond quick skim/scroll.
Grimblewald··on When does learning from data work (math starting from basic probability)
I don't mean to be rude, but this reads like 0 human oversight Ai slop
Grimblewald··on A digital billboard company has the technology to make 3D ads on moving trucks
If I find myself in a jury panel for someone who took a hammer to these, I already know im voting not guilty, and recommending they get keys to the city.
Grimblewald··on The 52-Page Memo That Nearly Destroyed OpenAI: Ilya Sutskever's Deposition
"How would less competition make the world better?" In the context of Umbrella Corp and bayer. Both undoubtedly bad, but one so evil it is irredeemable.

Hmm. I wonder how openAI, in the direction it was taken under Altmans leadership, is a fundamentally and deeply rooted evil thing to allow to exist. It really boggles the brain.

It's like saying "Therapy needs competition, this is why we're allowing PsychoTheRapist to continue practicing, to prevent psychotherapist's from getting to cocky"

Which is odd because the two seem to have fundamentally different purposes with immediate and long-term impacts of the former being unambiguously negative

Besides, strangling OAI in it's cot, as the natural order of things would be, wouldn't get rid of competition. Competition is fierce via Chinese and European offerings.

Grimblewald··on AI slop is killing online communities
Ironically, llm's are a decent way to combat. Even relativley small / simple llm's paired with a good slop rubric do a decent job. E.g. set up scoring so that time to answer is a factor, how directly relevant things are, how much exposition is offered etc.
Grimblewald··on OpenAI models coming to Amazon Bedrock: Interview with OpenAI and AWS CEOs
Isn't it an open secret that benchmarks are largly irrelevant at this point? Why else we do all have a personalized test battery for new models? That said i've stopped testing chatgpt entierly. Its still ok but is beaten by local models and it gets thrashed by non oai frontier providers. I get the history, but holding up oai outputs as equivallent is lile comparing yahoo to google post yahoo's collapse in search domains.

Oai language models are largly irrelevant at this point imo.

Grimblewald··on Is my blue your blue? (2024)
A neither option would also work. Point is half the colours i was shown fall into neither
Grimblewald··on Is my blue your blue? (2024)
Agree, this seemed silly. It seems to be more a question of "would you say turquise is blue or green?" Rather than a question of our blues match. Better imo would be to ask something like paired colours and pick the "more blue" one. Cool idea for a website, but imo poorly formulated.
Grimblewald··on We got 207 tok/s with Qwen3.5-27B on an RTX 3090
This, even on android via termux you can run ollama with gpu accelaration on phone. This works, though milage will vary.
Grimblewald··on Changes in the system prompt between Claude Opus 4.6 and 4.7
Nah, until recently i still had access via web chat interface, and often paste a transcript and files for somethong 4.7 keeps fucking up, paste response into files as appropriate, and attempt to continue with 4.7.

I swear 4.6+ looks for reasons to ask clarifying questions sometimes, even when really not required, and this fucks flow/quality up in a big way.

I just wish there was a "im not stupid" checkbox you can use to get a minimalistic interference access to claude. Im starting to use local models again, which I havent in a while because claude was so much better, but once i fully lose access to 4.5 it might be time to go back to fully local for good. 4.6+ fails to add value for me, projects 4.5- did good jobs on first try now require multiple prompts and feedback. Exact same initial prompt and project files extracted from archive. I liked claude because it aced those tests while local required handholding. Now claude requires handholding, so why use it over local? Once 4.5 leaves openrouter it might just be time.

Grimblewald··on Game devs explain the tricks involved with letting you pause a game
Right, all valid points, but consider the scale of a game like those coming out of rockstar. I'd understand for indie games and arcade games, but a single player rpg that will likely never be seen in arcade settings? Seems odd to me to see it here. Rockstar has the resources to do it properly, one would think, no?
Grimblewald··on Changes in the system prompt between Claude Opus 4.6 and 4.7
I miss 4.5. It was gold.
Grimblewald··on Banned by Anthropic?
Wild that it gets billed before it is accepted.
Grimblewald··on Banned by Anthropic
Good lord, these cases are quite problematic, i was going to use claude for some legacy stuff but i don't feel like getting banned over something innocent "can you identify a how we can fix the slave's behaviour? They're not listening to master properly"
← PreviousPage 6 of 30Next →