It is unreal how good GLM 5.3 Flash is. There's an ~4 bpw EXL3 quant that fits on 2x DGX Sparks. I'm not sure it's worth it now ($7k/Spark) or even when I bought them ($4k/Spark) but there's something about knowing that the whole internet could go down and I could still have this model running at 50-60 tokens/s with 1M context, indefinitely. It's just so good. TensorFold hits 100+ tokens/s in aggregate with 2-3 M total context pool. It's like having Claude at home.
I'm using Flash and the full 5.3 model side-by-side, and for coding I'm hard-pressed to find a difference. Flash is slightly worse at steering in longer contexts, but writes code that's just as good as anything I get out of full 5.3. You could totally build a SaaS or video game with nothing but 5.3 Flash.
The full 5.3 is only really needed for super deep bugfixes/profiling passes, or ridiculous reverse-engineering in my experience.
Yeah, MiMo Pro is the first Chinese model I can really clearly identify as an upgrade over GLM 5.3 Flash. K3, Qwen 3.8 Pro, GLM 5.3, DS4.1 Flash all feel around equivalent to me. Oddly lots of people have derided MiMo but I have found it able to tackle slightly tougher problems. For instance, I tried GLM 5.3 Flash, DS4.1 Flash, and MiMo Pro on making a new map layer for my Garmin watch, and MiMo was the only one that found the actual best method (.img file for OSM).
Garmin.img thing is the interesting, but not really on my Experience.
Been running the same stuff on GLM 5.3 Flash, Deepseek v4p, Qwen 3.x model and Mimo v2.x pro model. Mimo faster yeah, but its hallucinate more and losses the thread easier. And when the context window is not limited, when usage of >200k token, is loss context. even he didnt know what im sayin.
Also, sometimes he Had a Loop over and over, for the same tool call (on my case is Grep and Read. >15x looping) until is burning a bunch of context for nothing. GLM and Deepseek doesn't do that (but the deepseek Hallucinations is huge asf too).
In my experience, like most models, it requires a ton of supervision. It is extermely verbose. Flash loves to create some wildly unecessary abstractions/spaghetti code (like most llms). It wants to implement things in the least effecient manner and needs to be reminded constantly not to do things a certain way.
Am I the only one experiencing accidental chinese tokens, wrong grammar, and other sloppy mistakes in almost every output when using GLM 5.3? I feel like using it through any kind of subscription service (z.ai, Opencode Go) just serves me aggressive quants.
On some days it's usable, but on most days it feels more like having to trust a pretty smart intern that may or may not get a small stroke every couple of minutes.
Never had that happen at all and using since launch as well, but I have a very custom harness running.
My local machine went down however and I used their web API to repair it and talking to that felt much stupider, I think harness really matters Alot.
I do get the really annoying LLM "that's A, not B" and "load-bearing" from time to time, those two tells stay no matter what I do even though the rest of the output is pretty amazing,
I was running DS731 until Qwen 3.8 Flash Next (so what, like a couple weeks? The pace has been torrid). I switched to Qwen for image input, then GLM 5.3 Flash. At the time, Qwen felt like a minor step down from DS731, GLM 5.3 Flash feels like a major step up from Qwen, so I'd say slightly better than DS731. I realize that's not exactly what the benchmarks say, but DS731 definitely flailed around on some of the more challenging things that GLM mostly gets for me.
There's a few comments (boldly) criticizing the ruling for being technically incorrect. If no child was abused, how can it be a child abuse crime?
I would counter with this: If it is technically impossible to tell the difference between AI generated images, and real ones, how could anyone ever be prosecuted for having presumptive real ones in their possession ever again? Everyone tried would simply say "AI" and that'd be that, unless there was a victim to testify, or it was a case on generation of it etc.
So AI images of this nature do have victims, just slightly less directly.
Another adjacent question: How would you feel if it was an AI generated image of a specific child. What is the precise threshold of victimhood that is necessary?
I disagree. If it's impossible to tell the difference, people WILL start distributing collections of real and artificial child pornography combined while labeling it as 100% artificial. Most public websites will make both against the terms of service, but it makes it much harder for victims to get support from law enforcement to take down genuine CSAM.
Also, now that it's legal to own and consume content featuring realistic depictions of sexualized children, do you think that will lead to fewer or more children getting abused?
I wonder how people (like yourself) would would feel about "un-watermarked AI-generated CSAM is illegal, but watermarked is not". Or is is your first argument irrelevant and the second one is what actually matters to you?
> Also, now that it's legal to own and consume content featuring realistic depictions of sexualized children, do you think that will lead to fewer or more children getting abused?
I'm not sure. Is the intuition that it will lead to more? because I think there are data-based arguments to the contrary.
I'm not convinced you can stop someone from taking real CSAM, and recreating it with AI to get a copy with a watermark.
Also, assuming your correct and someday we have some watermarked attested "victimless" child pornography. I'd love to hear your best "data-based" argument that shows that predators will be satiated by pictures they have no doubt about being fake. It's laughable.
Should we allow victims of child pornography to endorse the distribution of images of themselves as a child once they are an adult too since at that point, each additional viewer doesn't generate an additional victim?
No. What it means is different. Let's use fire as an example. There is good reason for the fire brigade to come with full force every time a significant amount of smoke emerges from a high rise building. Meaning banning people from simulating a fire with the use of fog grenades and smoke machines without having clear agreement with the fire brigade is totally reasonable. It is in all our interest that the fire brigade comes when smoke is seen.
Every convincingly real CSAM encountered online has to be treated as potentially being the real thing. Even material with clear AI markers needs to be treated as such since that could be an attempt to obfuscate real material. Allowing AI generated CSAM while banning real material seems to be impossible to enforce and would only further a normalization of encountering CSAM and pedophilia.
If you catch a pedophile and they have actual CSAM on their machines, you also do not have to litigate which ones are real and which ones are fake. If fake ones are legal the burden of proof is now shifted as the pedophile can easily argue that pic that contains documented abuse of your daughter is in fact just AI generated.
Society is always about weighing rights against each other. And while "protect the children" has become the hypocritical rallying cry of politicians that want to introduce a surveillance state, in this case I would happily argue for keeping a strong stance e en as someone who has repeatedly argued against surveillance. Pedophilia is shunned in most societies for a reason and in my opinion it should remain so. If people have that unfortunate sexual urge, society does not profit from safely enabling it while producing many potential risks for actual victims and a lot of work for those policing it.
Now it is known that many pedophiles never touch a child. Normalizing pedophilia as a society would likely change that. Allowing CSAM if it is AI generated (from real training material btw.!) may change that. And this is not a thing we need.
Instead society should normalize it for non-acting pedophiles to seek therapeutic treatment.
You're just making more criminals this way. It's the same as saying air rifles that are indistinguishable from fully automatic weapons should be treated as fully automatic weapons.
I hate to extend a bad analogy, but If you rob a convenience store with an airsoft rifle that is visually indistinguishable from a real gun, you will be charged with armed robbery.
People are already using AI to generate porn of real minors, so you don't have to get hypothetical to see how the technology is used to abuse children. If it's easy for any predator to do this, then it's not hard to imagine that they will blackmail minors by threatening to release AI nudes to their friends and family unless they send them real CSAM.
You've combined two crimes into one. Robbing a bank with anything is armed robbery. My analogy is apt.
If I get pulled over with an airsoft identical to an AK47 and charged with trafficking illegal firearms, I think you will agree that I shouldn't be. It's not an AK47, it's an airsoft.
Similarly, calling AI generated imagery CSAM might be a pure category error. If there is no C being SA, it's not CSAM by definition.
Now if that imagery came about because real CSAM material was in the training data, or because someone used AI to generate sexual imagery of a real person, that's a different story. You now have a derivative work based on a real crime with real victims. But we'd also have some very real and important questions to be asking the model trainers.
In a vacuum though? Sorry no, it doesn't make any sense.
Generating sexualized pictures of minors and normalizing pedophillia is soliciting more abuse. You are willfully ignore that at this point to defend your shitty analogy. Your analogy sucks because child porn, and especially it's distribution is inherently bad for society. Trafficking an airsoft gun does nothing. Your semantic arguments about the acronym are extremely cliche especially since you know exactly what everyone means. Before some people started calling it CSAM, it was called child pornography AND IT WAS STILL ILLEGAL
Hold on, is there any evidence of this? This is a very delicate topic and gut reactions are often wrong. Playing violent video games doesn't cause violent behavior, for example.
We should be preventing harm to real children, as well as treating/managing people's psychological disorder of being attracted to children (instead of simply criminalizing it).
The war on drugs comes to mind - it criminalized a substance use disorder, gave power to organized crime, made it harder for people to seek help, allowed the market to be flooded with the most harmful versions of those substances which ultimately killed more people than it saved.
I think you're being overly emotional and haven't addressed any of my points. You just doubled down and said it's child porn again when I've stated to you a couple times that if the porn contains no child, it can't be child porn by definition.
And to be clear I'm not a consumer of such things nor am I attempting to normalize anything. I just want to find out the truth about stuff.
And yes, weapons that are indistinguishable from automatic weapons should be expected to be treated as such. I likely live in a world region with far less people dead from gun violence than you. And I gladly make that choice to keep my fellow humans alive, every single day. Only a brainwashed sociopath would not.
I know this is a different discourse in the US with different positions you all became propagandized to accept as "normal", but the numbers speak for themselves.
If you cannot tell the difference, you cannot tell the difference. Unfortunate as it may be, the standard of proof for criminal conviction must be high. There's a reason that various legal systems across the world have converged on such principles as proof beyond reasonable doubt, in dubio pro reo, intime conviction etc. basically stating that the legal burden of proof is on the prosecution to present compelling evidence of crime and that the benefit of doubt must go to the accused. It is considered so fundamental and so important that it has made it into the International Covenant on Civil and Political Rights which a majority of countries in the world (including Denmark) have ratified. There must no doubt have been painful lessons in history which led humanity to eventually adopt such ideas. We must be careful to forget those and inadvertently take steps going back towards that direction.
> I would counter with this: If it is technically impossible to tell the difference between AI generated images, and real ones, how could anyone ever be prosecuted for having presumptive real ones in their possession ever again?
If AI CSAM is legal, the market for real CSAM is effectively destroyed. It's heartbreaking to me that--incredibly--we're making our best weapon in the fight against child abuse illegal. It's the methadone of CSAM. What the fuck are we doing?
You can jail the actual pedophile in that scenario, you cannot jail an agent. The real question is whether the person who instructs the computer to do the thing that leads to hacking should be liable, or whether the company that made the tool that when used unsafely lead to hacking should be liable. Gun manufacturer's typically are shielded from liability. They also have a lot more disclaimers on their products though.
>You can jail the actual pedophile in that scenario, you cannot jail an agent.
How is this relevant? Should we just jail the closest person that caused a bad, just to prove some sort of point?
>The real question is whether the person who instructs the computer to do the thing that leads to hacking should be liable, or whether the company that made the tool that when used unsafely lead to hacking should be liable.
This totally ignores the civil vs criminal distinction. Even in the case of guns, no one seriously thinks that gun company CEOs should be sent to prison for murder.
Technically beta keratin is helical. That said you are correct, there is no requirement to be helical for fiber diffraction. It just turns out it's very difficult to have non helical uniaxial symmetry in nature.
When I was studying spindle shaped viruses by electron microscopy we'd get beautiful fiber diffraction patterns in the fourier transform of the micrographs. I always wanted to try and solve the helical parameters but never got around to it.
Specifically the antiparallel base matching was the important bit, agreed, but minor nitpick. The first alpha helix wasn't solved until 1951, so one year earlier. Their ubiquity was very novel.
I think the reqs were 12 years old and $171 or $271
Coming from $300/yr colorado insurance its a steal
I'm full time in a camper right now so this entire conversation is extra fun because I don't have a physical address anywhere which is a whole other ball of worms just insuring and registering my truck.
Not comparable. The markets can still charge wherever they want, but it needs to be priced per item and not per customer. In your house counterexample, you could set whatever list price you want.
Not sure why I'm being down voted. She was initially diagnosed with complex regional pain syndrome aka "the suicide disease" for how much pain it causes, and heavily medicated. Her official genetic test didn't found anything, but trough years of iteration of a mixture of codex and Claude code over her raw genetic data we built a working theory of what might be wrong and trough cupious daily notes, labs, and over several iterations of feeding the notes to different ais (whatever was the best Ai at the time, with the exception of fable due to the refusals), we manage to get her to a much better situation where she can now tolerate touch up to the toe line and can also tolerate phisioteraphy. Turns out her genes point towards a very weird energy and electrolyte deficit that must be approached from a very unique angle. The first real impact of an llm on the condition was via o1-preview
I tried to connect Claude to my Google workspace calendar and email and tasks. Calendar and email no problems. But no public Tasks mcp server seems to exist. I'd have to host my own.
It also has 196B ngram params that sit on disk. So it's really pretty comparable to GLM 5.3 in size.
reply