In my experience, deep seek models are massively overrated in terms of how good they actually are at agantic usage, coding and writing, just because they are kind of the first open source entrant and the name a lot of people know. Try GLM 5.1, coding and writing just because they are kind of the first open source entrant and the name a lot of people know. Try GLM 5.1.
They have a single switch that will remove all AI features from the interface. Why do you need more than that? This is not a rhetorical question. I genuinely don't understand it — if you can get all of those features completely out of sight, deactivated, with no trace of them left except that one switch, why is that not enough? Is it as though any kind of AI integration like contaminates the purity of the code or something?
I need more than that because I have no guarantee that its true. I need the source. Or I at least need them to provide a build that they promise doesn't have that stuff in it at all, so that if any analysis was done on a decompilation, there would be some level of certainty that they were telling the truth. Anything that leaves any of it in complicates that effort and makes the certainty that less certain.
RESPONSE EDIT (clear and intentional rate-limit evasion):
It's not paranoia; I'm not concerned if they "take" my content. I write open source, CC0 licensed software. I couldn't give a fuck about anyone doing literally anything they want to do with the code I write. Literally take it and call it your own, for all I care. If I can return the interrogation, why are you so concerned with ownership? Why was that the first place your mind jumped to? Paraphrasing: where is the need for this insane level of "if you've got nothing to hide..." submission?
Like I said: it's about trust. They want me to trust them. You, for some inexplicable reason, seem really upset that I won't trust them. Neither of the parties have given me any reason to trust them. Just insistence that I should, if I want to use their product. And while I entirely agree with that rationale, I don't understand why I get clapbacks for stating that I intend to adhere to that agreement entirely! Won't use the product because they won't give me what I need to trust them. That should be making everyone happy, right? I know I'm happy with the arrangement, at least.
Aside from all that, and far more relevant to my actual comments: another user pointed out the repository where they DO offer the transparency that I'm asking for. So your entire hissy fit is moot when you could have just pointed out that I was wrong in my understanding of what they offered. I mean, that would have gotten in the way of your sycophantic leap to the defense of the company I was so hellaciously attacking, so I understand why a good capitalist bootlicker might not think of that first, but at least now we both know!
You still haven't explained literally anything. Yeah, okay, if there's a switch, you can't be sure that every single AI related code path is fully disabled.
But if you flip the switch and there isn't any AI integration visible in the interface anymore to bother you, why does it matter whether the code is there or not, or technically active or not? Raw integration points and settings windows don't send data literally anywhere at all until you explicitly configure an API key and a URL or sign in to an AI provider or whatever. It'd have nowhere to go, and AI inference costs money. It's just local code providing a set of integrations. Where is the need for this insane level of paranoia?
So you trust a build they say doesn't have AI features, but not a switch that they say turns off the AI features? Doesn't seem like a logically consistent stance to me.
Plus, you can just packet sniff and see if they're doing anything AI related when the switch is off.
Why are you so concerned by there being AI code in the editor that you need this level of trust?
The point I have been attempting to make is that needing this level of trust and verification when you can't even explain or articulate what you're worried about at all is weird and confusing and unjustified, and I've been trying to get you to explain what you're so concerned about.
Required trust/verification should be proportional to what you're concerned about happening.
> Paraphrasing: where is the need for this insane level of ... submission?
It is not an "insane level of submission" to point out that trusting that a toggle does what it's supposed to do, when the possible consequences are basically nonexistent — as I said, without a connection to an AI provider set up, where is it going to be sending your data? No one is doing AI inference for free; and now you've even knocked the concern about code ownership out, so again, what really is the concern? — is probably reasonable.
Also, this is not remotely the equivalent of the old "if you've got nothing to hide" canard, because "if you've got nothing to hide, you've got nothing to worry about" is a justification for surveillance of your personal life; this is literally not doing surveillance. Also, even worrying about surveillance needs to be justified with an actual explanation of what negative things you expect to happen; with surveillance that's obvious; with this, it is not, which is why I'm asking you to explain it.
> Like I said: it's about trust. They want me to trust them. You, for some inexplicable reason, seem really upset that I won't trust them. Neither of the parties have given me any reason to trust them. Just insistence that I should, if I want to use their product. And while I entirely agree with that rationale, I don't understand why I get clapbacks for stating that I intend to adhere to that agreement entirely! Won't use the product because they won't give me what I need to trust them. That should be making everyone happy, right? I know I'm happy with the arrangement, at least.
I'm annoyed, because your standards for trust are insane when you can't even articulate what you're trying to guard against. You're getting clapbacks not because you won't use their product if you don't trust them, but because your standards for trust are extremely high and you seem to be completely unable to explain why to any degree, instead just getting irrationally angry at me for just asking you why.
> So your entire hissy fit is moot when you could have just pointed out that I was wrong in my understanding of what they offered
You're very clearly the one having a hissy fit lol.
> I mean, that would have gotten in the way of your sycophantic leap to the defense of the company I was so hellaciously attacking, so I understand why a good capitalist bootlicker might not think of that first, but at least now we both know!
Ah, and now we get to the personal attacks. Of course.
They literally have a quote from Tao in the article saying it was a novel approach humans hadn't tried, and that the problem hadn't been solved even after a lot of professional attention.
You're not arguing in good faith here, but just to make this apparent to everyone else: the disclaimers talk about the general case of Erdos problems as a whole.
The article explicitly acknowledges them, but then says that the disclaimers don't apply in this specific case:
> ...experts have warned that these problems are an imperfect benchmark of artificial intelligence’s mathematical prowess. They range dramatically in both significance and difficulty, and many AI solutions have turned out to be less original than they appeared.
The new solution—which Price got in response to a single prompt to GPT-5.4 Pro and posted on www.erdosproblems.com, a website devoted to the Erdős problems, just over a week ago—is different. The problem it solves has eluded some prominent minds, bestowing it some esteem. And more importantly, the AI seems to have used a totally new method for problems of this kind. It’s too soon to say with certainty, but this LLM-conceived connection may be useful for broader applications—something hard to find among recently touted AI triumphs in math.
So I don't see why I have to trust only one of only the other.
Furthermore, their assessment is backed up by direct quotes from Tao himself:
> “This one is a bit different because people did look at it, and the humans that looked at it just collectively made a slight wrong turn at move one,” says Terence Tao, a mathematician at the University of California, Los Angeles, who has become a prominent scorekeeper for AI’s push into his field. “What’s beginning to emerge is that the problem was maybe easier than expected, and it was like there was some kind of mental block.”... “We have discovered a new way to think about large numbers and their anatomy,” Tao says. “It’s a nice achievement. I think the jury is still out on the long-term significance.”
Would you care to explain in what way his claims have been misleading? Because I have read all of his articles and attacked his math and his sources, and so on, and I haven't found them misleading at all. The biggest way I've seen him accuse of being misleading makes the exact mistake he responds to from Joshi in [this piece](https://blog.andymasley.com/p/replies-to-criticisms-of-my-ch...).
Is he wrong about AI water usage, though? Using the fact that he defended AI water usage, as a reason to ignore everything he says, only works if he was wrong to defend AI water usage.
I read that article in depth and checked all the math and the extensive sources and found it very much accurate and it convinced me that the water usage issue is not serious.
He’s kinda wrong. The facts he states are correct. Others are clearly exaggerating or making mistakes. But he jumps from that to the incorrect conclusion that “data centers don’t use water”.
There are water accounting games on how much water is used for the electricity - Masley is right about that. But he ignores the water actually used to cool data centers, which is about 1/10th that. (About 25 billion gallons per year vs the 200B misstated as consumed during electric generation.) 25B gallons is actually still quite small in the big picture, but it’s growing very fast. And in the local regions where these data centers are built that can be a huge strain on the water supply.
So I would say Masley is biased in his reporting, and should be read critically. But on infrasound I think he’s right.
I think my takeaway from the water usage deep dive was about the scale of the numbers and a better intuition about water usage, but also that you really need to consider each data center uniquely. He'll say in broad strokes that data centers are fine, and then mention the few exceptions (in the infrasound article, that's the xAI DC). That's fine for the moment when he wrote the article, but if I'm evaluating a proposed data center in my local area, I don't know what bucket it falls in. Is it the exception or the norm? Still, because I read that deep dive, I feel better equipped to make that evaluation.
I don't think saying "as a general rule, data centers don't use remotely enough water to be any kind of significant threat, when you see through the accounting games, media hype, and look at things in a proper context" is made wrong or misleading by admitting that there are a few exceptions.
> That's fine for the moment when he wrote the article, but if I'm evaluating a proposed data center in my local area, I don't know what bucket it falls in. Is it the exception or the norm?
It does help you set your priors though, and not fall for the "all AI guzzles water, and DCs are dehydrating every town they're in and are always bad, without doing the research" rhetoric that's driving DC bans around the country right now.
I think we agree? I can't tell if you meant this as additional support for what I wrote or as a rebuttal.
Regardless, yeah, I don't think that's wrong or misleading. I only meant to say that because there are exceptions now, there might also be more exceptions in the future. Which, to me, means it's important to evaluate each new local DC individually.
And your point about setting my priors is exactly what I'm saying, too.
> Others are clearly exaggerating or making mistakes.
I'd be interested to hear a specific example, so I can get a sense for what you mean.
> But he jumps from that to the incorrect conclusion that “data centers don’t use water”.
He doesn't ignore the amount they are using, though — he goes to great lengths to contextualize how much water that actually is, compared to other industries (at a national scale) and other industries and recreational things (like golf courses and water parks and so on) at the local scale, specifically to point out that "25B gallons is actually still quite small in the big picture," as you yourself say — and yes, the water usage is growing "fast," but I don't know that anyone's actually quantified that growth rate, and it's still small in comparison to plenty of other industries that also grow year over year, and I neither he nor I think it'll continue to grow forever (AI bubble and all that).
> And in the local regions where these data centers are built that can be a huge strain on the water supply.
He explicitly deals with this, and as far as I can tell he's also right here, that there isn't a meaningful strain on most local water supplies either, as a fraction of total water production or in comparison to other industries those places also choose to host that are water intensive, despite being in arid climates, like the aforementioned golf courses, water parks, and other more industrial things. He goes through all the specific news headlines that claim that the water thing is a serious issue, and show that either they're talking about something different (like data center construction temporarily dirtying well water in nearby houses) or just pointing out that data centers "use water" and are also in arid areas, as if that's self-evidently bad, when other water using industries are already there and it isn't a big issue.
If you could point me to sources that he missed that disprove this, I'd be open to it for sure — I'm open to being wrong, and not committed to absolutely defending the honor of a guy I've never met on the internet against all odds. But I'm not personally aware of any contradictory evidence. I've been linked to a few reports from various foundations before, but they always are referencing numbers from other reports that link to other reports that, if you follow the whole process to the end, bottoms out in random news articles with unsubstantiated numbers that don't line up with what any actual math or other reports say.
> So I would say Masley is biased in his reporting, and should be read critically.
I read everything very critically, especially when it seems "too good to be true," like a lot of the stuff he says, but I might've missed something?
LMArena isn't very useful as a benchmark, however I can vouch for the fact that GLM 5.1 is astonishingly good. Several people I know who have a $100/mo Claude Code subscription are considering cancelling it and going all in on GLM, because it's finally gotten (for them) comparable to Opus 4.5/6. I don't use Opus myself, but I can definitely say that the jump from the (imvho) previous best open weight model Kimi K2.5 to this is otherworldly — and K2.5 was already a huge jump itself!
reply