HNHacker News
TopNewBestAskShowJobs

Grimblewald

1,509 karma · joined May 5, 2023

submissionscomments
Grimblewald··on Protect your right to run local AI
Eh, let em. If the US economy wants to self-sabotage, let them. Its a dying empire, lets its fall be hastened. I'm ready for china to fill the US vacuum. At least china controls it's billionairs (see jack ma saga) rather than the inverse.
Grimblewald··on Android Developer Verification: Threat masquerading as protection
At least we dont have a social credit system... hey wait a minute
Grimblewald··on Leanstral 1.5
I did get a refund quite quickly, after I finally figured out how to contact their support, which in some stroke of cosmic humor took deepseek to figure out because using their website left my trapped in a dead-end loop. Asking mistral for help lead to links that all 404'd or the same useless help section / faq page.
Grimblewald··on Claude Code is steganographically marking requests
Correct, and I've stopped. I've moved to deepseek instead, equally capable and not as morally bankrupt as either OAI or anthropic.
Grimblewald··on Claude Code is steganographically marking requests
Them strugglig to prevent others from doing what they themsleves do, is my problem to a degree where I must pay for the pleasure to getting sabotaged? Are you fucked in the head? In what world are we "being fair" by claiming that? It takes a problem anthripic has hallucinated, and makes its consequences mine for no reason. Anthropic chose to create both this problem and it's consequences, why am I wearing it?

If I decide I dislike words starting in S, despite myself being a prolific user of words starting in S, and smack some child who'd never even heard the rules, simply for saying "sorry", simply because some people say "shit" and it makes me mad, despite it being my most used word, are we "being fair" by saying "to be fair, managing curse words is a difficult problem"

no right? Insane take.

Grimblewald··on Claude Code is steganographically marking requests
Changing well documented requirememts and functioning code to do subtly incorrect things, over multiple areas, for no reason. When confronted you get the usual, "youre right to push back, there was no reason to touch that and it did make everything worse"

Why do I suspect faulty trigger? The things wrecked weren't wrecked by even much less capable, including local models, while reverting everything to pre fucking up and asking claude again led to similar results. Once on whatever shitlist that was, claude also failed tasks it previously aced when given the exact same prompt and project files. I attributed it to opus 4.6 being a downgrade which people always pushed back on, claiming they thought it was better, but i had empirical proof, it couldnt do tasks 4.5- could do quite well. Now all of this? It's clicking all of a sudden that its quite plausible i got flagged somehow and ended up with an intentionally degraded service.

So, claude is off table for me these days, and deepseek gets very deep git commit read-throughs with every file even being read being carefully monitored. This obviously ruins the "agentic" promise, but the reality is we cannot trust these fucks (being the companies). The irony is deepseek now queries claude on problems its stuck on for input via openrouter, with mild success (the gap really isnt all that big, if its even there), before escalating to me for input on direction on solving a given problem. So now chinese models get more claude training data, not less. In fact, they get training data on frontier ML stacks and problems. Anthropic did that to themselves.

Grimblewald··on Claude Code is steganographically marking requests
So block people, instead of having false positives be secretly fucked over, and having them pay for the pleasure?

Given the hidden model degradation of fable and now this, what makes you think this is where it stops? That's just what we know about and there's clearly a long-standing and deeply rooted malicious intent here.

I've had Claude fuck over clean well documented code-bases for no reason, and there's a good chance this is due to some faulty trigger. Luckily I don't trust these things one bit, and claude only ever runs in an isolated VM, however, I am pissed I am being made to pay for their errors in detection and waste my time fixing things I apparently paid to have fucked up.

That's unacceptable conduct. It's witch-hunting. Punishment and attacks on you for things without real proof. That isn't right.

Grimblewald··on Leanstral 1.5
Got curious, sign up, add money to account, try to use. Can't, it's a labs model. Fine, let's enable labs. Can't, unspecified error. Fine, lets contact customer support as instructed, can't no customer support, just a half-assed FAQ, that seems vibe-coded and searched poorly, totally irrelevant answers coming up for all queries tried. Then it hit me:

If AI makes good customer support, then why does no AI company use theirs to provide customer support?

Grimblewald··on Ford AI hiccups push carmaker to rehire ‘gray beard’ inspectors
I might not use all the same terms as you, but i largly agree. However llms live in a world almost entierly divorced from our own. Ours is physical, it's is human information corpi. For tasks suited to its world, and llm is as "human" as I am in my own domain. Same same but different. Intelligence is a fundamental trait of our universe, not a surprise.
Grimblewald··on .self: A new top-level domain designed to support self-hosting
In this econimy? where google's full might is behind killing self-hosting? Be still, my beating heart --- there may be hope yet.
Grimblewald··on WhatsApp to let people chat without swapping phone numbers
Vendor lockin disguised as a feature.
Grimblewald··on Why does kinetic energy increase quadratically, not linearly, with speed? (2011)
I feel like its simple?

Velocity is itself an expression of energy, and energy does scale linearly. Why does it take quardically more energy for each extra unit velocity? Well, because the closer you get to the speed of light the harder it is to get closer still. Why that is simply requires a deeper appreciation of the system than I expect is possible to muster in a comment.

Grimblewald··on Ford AI hiccups push carmaker to rehire ‘gray beard’ inspectors
Velocity implies direction, AI is just speed sans direction, AI only workflows are just really fast brownian motion centred on training corpus mean for a task. Humans can give it direction, how good that direction is depends on human expertise.

We still need the humans, there are no cases for novel useful work I can think of, or have seen, where humans are no longer required.

Grimblewald··on PR spam today looks like email spam in the early 2000s
On par, a large part of that professional cohort has its entire career made by abusing rules intended to prevent abuse of rules.

It's a self-perpetuating monster that breeds attitudes and behaviours that maximise antisocial and exploitative behaviour by the system's design.

just look at what was done to copywrite, designed to protect inventors from cororates and related abuse, now unweildly for individuals and largly serves corporate interest more than the individuals it was supposed to protect. Same with patent system.

Grimblewald··on PR spam today looks like email spam in the early 2000s
A real leopards ate my face moment
Grimblewald··on Anthropic says Alibaba illicitly extracted Claude AI model capabilities
Chinese models are years ahead of american models on multimodal comprehension, better yet,they publish on what makes the models tick and release weights openly.

Chinese research outout, publically released, has also contributed in big ways to features present in every single US model. Yours is a bit of an unfair take I'd say.

Besides, claude will think its chatgpt sometimes, so clearly this isn't a problem restricted to china, turns out unethical companies will do unethical things /shrug

Grimblewald··on Anthropic says Alibaba illicitly extracted Claude AI model capabilities
Claude thinks it's chatGPT, and various chinese models sometimes, whats up with that?
Grimblewald··on Extreme Heat conference cancelled due to extreme heat warning
Aicon is reasonable for areas where it's required, but "solving" things in areas where for millenia it wasnt required simply removes the pressure to act. This would be the opposite of what's required right now, which is decisive and heavy action on something we've been inactive on for way too long.
Grimblewald··on The hacker sent by Anthropic to calm the government's nerves about AI safety
Remember strawberry? I do. Rememver gpt2 hype? I do.

the hype isnt real, its marketing designed to inderectly siphon capital from the less informed.

The current USA government is no stranger to a grift, so they'll get it. Not that I agree with the practumice, this bubble will hurt quite badly when it goes, but at least AI fundamentally does deliver something useful, even if it isnt infinite value as typically promised.

Think dotcom bubble and the hype and promises made surrouning that. Tge hyper will pass, the bubble will pop, and life goes back to normal as ai becomes part of the mundane everyday human environment. Like websites and domains, some will be used well, some for evil, not everyone needs it, and as we continue to move towards energy being our fundamental unit of value / exchange, if we cant make these models way more efdicient then their use case will be rather limited in scope.

Grimblewald··on AI Slop Has Taken over LinkedIn
And nothing of value was lost
Grimblewald··on What the Fuck Happened to Nerds
Great post, and largly captures my own experience.

I can only speak for my institution, but eagerness to lock down ip and keep ownership of everything tightly controlled and out of the hands of said nerds/inventors doesnt really incentivise me to do beyond what I'm paid for.

The one time I tried, I was hit by the full force of my institutions commercialization goons and lawyers, to a degree that it killed my drive to do anything novel for them. Despite being promised partial ownership, in the end, after federal grant funds were secured and product developed, they took everything using "loopholes" that go against the law and the institutions own rules, but to fight it I need resources I don't have, which the institution no doubt knows. All that despite me initially being fully aligned with my institution, and happy to only take a very minor share of actual profit, in-line with what i'd get anway, only stipulation was veto rights in application (as the tech has very real applications in offline autonomous drones, which I consider an X-treat).

If my own institution is a hostile actor, and willing to fuck me over nothing, simply because they can, why do anything?

So, current state of Copyright law favours institutions over the very individuals it was meant to protect, and there are no options to protect one self if anything interesting is developed without serious capital and legal might. So, fuck it, im not doing anything except hobby related, GPL licensed stuff. If I can do anything to make it hard to commercialize, I will. If it can be kept in house, it is kept there.

Capital interest has become a rather ugly and hostile egregore with interest aligned against that of humanity. All those building cool and novel shit I know hold similar opinions, so it is no surprise to me. I was strongly advised against working with the institution by older folks i look up to, people who have built really powerful tools of their own. Their warnings ended up being proven valid with deafening clarifty. I've always found the statement that capitalism breeds innovation to be a joke, and while it works in the chinese model, the "western" model is sick and suffers a sort of cultural psychosis that makes it rather unttractive to engage with.

Grimblewald··on Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable
Then you have to fight, not just the battle but the war. Its not enoygh to get people to back down, keep going, go for the throat. Demand the resignation of the politician proposing the bill. If theres no risk for trying, theyll keep trying.
Grimblewald··on Anthropic requires 30 day data retention for Fable and Mythos
> session limit reached
Grimblewald··on Somali referee for World Cup denied entry into United States
The USA is, albeit rapidly waning, an important part of the tech world. It lets me know that when invited to the USA to share my knowledge and skills, I am best serving my interests by declining such invitations and instead proposing that they visit me, which they inevitably do.

I am not against the USA per se, and would love to visit it's national parks one day, but in it's current state it seems insane to waste time and money attempting to travel there, even as a cis white male. You'd have to be mentally deficient or deeply desperate to even consider it in the current climate. This is a reality for many in the tech world, so I would say this is a very good fit for the crowd that frequents HN.

Grimblewald··on The backlash against AI, in 4 charts
Kinda straight-forward to understand backlash when you see how it's being handled. It's being forced on people without any option for opt-out. Most of the time, does not deliver on what is promised. Has obvious ethical and moral issues that make opt-out a must have.

The entire thing is being handled so poorly it feels like the entire bubble is engineered to collapse by design. Almost like it's the point. Which could be, when these companies collapse, the data people submitted becomes part of the package people buying the bankrupt entities assets can nab, and a lot of that data is a wet dream for companies otherwise prevented from gathering it.

Grimblewald··on Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable
I think all this started with post opus 4.5, that's when claude started wrecking my shit without extreme oversight. Codebases it was making positive contributions to before were slowly and constantly being eroded and wrecked. Give it tasks in isolation? still does well, but the moment it sees the bigger picture, it goes to shit. I chalked it up to a bad model but this makes it all seem like it may have been by design in retrospect.
Grimblewald··on Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable
Corporate America never backs down. It simply rallies and tries again later until people are too fatigued to care. The only solution is to abandon ship, which I am doing. MS walked back in OS ads the first few times, but ultimately we still ended up on the exact trajectory everyone was outraged at. OpenAI still ended up on its path to closed AI despite initial walk backs. The story repeats itself over and over again, so, once the bad behavior starts, you leave. Their apologies are as hollow as their moral posturing.
Grimblewald··on Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable
Am I being paid to do anthropic's work for it? See my comment history for some examples in another thread, but generally I see no reason to catalogue this for a model Ive seen no evidence of being worth the effort. I'm overworked as it is, doing this for no reason isnt something I can justify.

The successes I have had with the model were strictly worse than output from deepseek v4 pro on the exact same task.

Grimblewald··on Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable
I wear a few hats, but as a chemist and I'm not happy with fable. As a statistician I'm not happy with fable. As a data scientist I am not happy with fable. As an academic and a researcher I am not happy with fable. It's useless. I'd be surprised if anyone can get any output from it that couldn't easily be replaced with a search from wikipedia. Given how verbose claude models have become, wiki articles are probably less verbose too, and the tok/s is unmatched for a wiki article pull.
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.

← PreviousPage 5 of 30Next →