UwU
1d ago
"Another day, another LLM" is definitely correct. Deep Seek also released their newest variant DeepSeek-V3-0324 yesterday. DeepSeek-V3-0324 is a significant improvement over DeepSeek-V3 and even beats Claude 3.7 Sonnet in many benchmarks, and not to mention, it's open weight! I guess it's less sexy since it's a text-only model and we already have highly capable ones that are generally interchangeable for most purposes, but I'm excited to see the future DeepSeek-R2 that'll be based on this improvement.
UwU
1d ago
Some unordered reasons:
-
They call themselves OpenAI and yet nothing is open. They don’t publish research or release open source or open weight models.
-
They positioned themselves as a non-profit to get clout and talent and later reneges.
-
They charge an insane amount compared to other companies to price anchor because they want the SOTA models to only be accessible by elites.
-
They hide the thinking traces from their thinking models and ban people who try to figure out their methods with prompt engineering.
And then tack on “they hold back revolutionary features until their competitor releases their own”
UwU
1d ago
Damn, if it’s better than Gemini then it looks like you were right and they were sitting on that capability. It reaffirms my opinion that OpenAI is the most insidious AI company out there.
UwU
2d ago
Isn't the corollary to this that we should also ban teenage wrestling and gymnastics in addition to puberty blockers?
Indeed, the tweets and paper from OpenAI and Meta are evidence that they had image generation capabilities, but not whether their capabilities in practice were cool and shiny or more of the same. xAI claimed they had image editing in December, but according to you[1], it sucked then, and it still sucks now in the public release. That's more evidence that they had something interesting in December, and wanted to take their time to improve and productionize it, but was forced to release it now because Google released theirs. If it sucked then and it sucks now, they never had anything cool and shiny to hold back.
Doesn'tDidn't apply to the anime ban.
Yup, I just tested it too. When "doesn't" becomes "didn't", I think that's a point in favor for models becoming less censored than the other way around.
[1]: It's not a dig at you, by the way, I didn't follow the xAI developments back then so I'll take your word for it.
UwU
5d ago
It seems like the big AI companies are deathly terrified of releasing anything new at all, and are happy to just sit around for months or even years on shiny new tech, waiting for someone else to do it first. I remember reading that Google had internally achieved something akin to ChatGPT and just did nothing with it. Then once it comes out it's a race to the bottom in the safety and censorship lane while pushing incremental improvements. At least right now most of the cutting edge is happening out in the open, being published by researchers, so nobody is able to build a moat or a big lead technically, so the players are left chasing percentage points on the margins.
Is this really accurate? Because OpenAI's o1 was only fully released in early December of last year, just over 3 months ago. Google's advanced image generation capability was just released this month. Are they not considered Big AI Companies? The speed of AI advancement is actually very fast, and I don't see good evidence that companies are deathly afraid of releasing anything new. Perhaps the reason why xAI didn't release what they teased months ago is because they failed to improve on its functionality in a way that makes for a good product?
Also, what do you mean by "it's a race to the bottom in the safety and censorship lane"? If you mean they are becoming less censored, then that's true. Grok is largely uncensored, both OpenAI and Anthropic are making their flagship LLMs less censored, and you can turn off Gemini's safety filters in Google's AI Studio.
UwU
6d ago
This is likely just downstream from the fact that black people have an outsized influence on blue-tribe politics compared to other racial minorities. When asians and hispanics see POC gets rebranded as BIPOC and the media loudly champion Harris's plan to empower black men, it's no surprise that they feel less valued than their peers.
UwU
18d ago
Thanks for the replies! I don't think our positions are very far apart. I think the disagreement really comes from just how much sending weapons to Ukraine is harming NATO vs how much it is actually harming Russia. For NATO, we are blessed with tremendous economic advantages, so as long as there's a will, there's a way. The current issue seems to be that we are still working with peace-time production numbers whereas Russia has already instituted a war time economy since 2022. Still, I believe even with the current NATO production numbers, the situation is quite sustainable in Ukraine.
One example you brought up is tanks. As you mention, so far we've lost about 66 western MBTs in Ukraine out of the over a hundred or so we sent. Just how much tanks are NATO procuring and producing? It's hard to get an accurate number on, because individual countries are procuring separately, and there are a lot of models and variants to keep track of. Taking just one example of a NATO MBT, the Leopard 2A8, the newest variant unveiled in 2024, I found a number of procurement announcements, one article claims 123 units to Germany, 54 to Norway, 300 to Italy, 77 to Czech (but probably delayed and downgraded to older models), other articles claim 44 to Sweden, 44 to Lithuania, 46 to the Netherlands. The timeline for these are sparse, but available information on these orders seem to indicate delivery dates from 2026-2030. Disregarding the Czech order, assuming no additional orders, and also assuming only 50% of these tanks actually get delivered by 2030, we get a tank production number of 306 tanks in 5 years, or about 60 tanks a year. To sanity check, another article from early 2023 claims 50/year production rates with an additional 60-70/year refurbishment, which does track with the estimated production rates.
It seems like, even with just the production rate from one model of the western MBTs, the equipment losses are sustainable with pretty good headroom. If we add Abrams to the mix, the M1A2 SEPv3 production rate to satisfy domestic orders is 135 a year, which does not include export orders like the 250 tanks to Poland expected by 2026. This also does not include other NATO tanks like LeClerc and Challenger 2, but I don't expect their production numbers to be significant. For Poland in particular, they sent about 300 out of the total 500 tanks received by Ukraine, mostly outdated T-72s. But so far they already received as backfill 116 M1A1 Abrams, 28 M1A2 and 84 K2s, with more K2s expected in 2025 to completely cover the 300 tanks sent.
As for cruise missiles, it's hard to get any accurate numbers because the amount sent is classified. For one of the models we sent to Ukraine, JASSMs, the production numbers look like 720 units per year, and they are expanding it to 1,100 per year. I doubt Ukraine is expending over a thousand high-end cruise missiles a year.
I'm not familiar with mine-clearing vehicles, but I suspect that they are not that technically difficult to build compared to SPGs and MBTs, and we don't need them in high numbers.
I'd be very interested to know exactly what percentage, because Russia's been shelling the absolute heck out of them for a couple of years. But yes, I've never argued that Ukraine + NATO is weaker as a whole in raw military strength. That's not exactly the arrangement we have, though, the Ukrainians are doing most of the fighting and NATO is arming them.
Zelensky now claims 40% of Ukrainian materiel is domestically produced. Before the spat with Washington, he's claimed 30% in 2024. I'm not sure how accurate it is, but it seems plausible based on equipment loss numbers. But my main point over here was not adding up NATO and Ukraine's military strength, but their military industry output. The argument is that, by not ceding Ukraine, we get their MIC on our side, as opposed to the other way around.
I feel like you are just trying for gotchas without addressing the core of my argument here.
EDIT: but let me just respond quickly - The 3.6% growth number is self-reported by Russia, no longer independently verifiable because they stopped reporting a lot of their economic stats after the war started. Even if taken as true at face value, we know that they are experiencing high inflations even with record high interest rates, 21% to be exact, this indicates any economic growth is propped up by an aggressive fiscal policy that's unsustainable. But what are you actually arguing for with the 3.6% number, which is apparently better than the US? Are you claiming that Russians are simply immune to the effects of economic sanctions (then why do they want those lifted?), and having a sizable portion of their work-force and industrial output being blown up in Ukraine is actually good for their economy?
The NATO equipment parks questions, I don't know. I have not seen people do analysis on NATO equipment numbers from satellite imagery, unlike the Russian equipment parks. But if you have a link, I'd be interested. My premise is that we (NATO) are not depleting our materiel faster than Russia is. Germany's readiness going down from 65% to 50% is evidence that one of the NATO countries is being affected, it does not demonstrate NATO as a whole is seeing lower readiness numbers. For example, at least Poland is seeing an increase of its own military capabilities. And even if NATO is seeing a net depletion of its own reserves, it does not demonstrate that it is depleting it at a rate higher than Russia. We have a visual database of equipment losses that tracks individual equipment types for us to estimate the relative loss rates between Russian and NATO equipment (The page actually links to Ukrainian losses as a whole, of which only a fraction is NATO supplied equipment). Also, even if this is a problem, the correct solution is to increase NATO's military production, kind of like what EU is doing right now, rather than ceding Ukraine to Russia. Again, I remind you that Ukraine is producing a significant amount of its own weapons and under no circumstances that NATO + Ukraine vs Russia is worse than NATO vs Russia + Ukraine.
UwU
19d ago
Russia has blown through its stockpiles and it is not doing well economically, which impacts its military production and force generation. Sure, their military capability might have increased through adaptations and experience, especially with drones, but it's not a significant departure from their army in 2022. In fact, in terms of equipment, their formations are probably less mechanized today than three years ago. I don't see a parallel between the Russia today and the US or USSR from 1945.
Yes, if we (NATO) are depleting our materiel faster than we are depleting that of Russia, then that's a problem. So far, I don't see good evidence that's the case. Further, Ukraine itself is also producing a significant amount of war materiel like ammo, armored vehicles and drones, so having Ukraine by our side is still better than forcing it to surrender.
UwU
19d ago
I'm not aware of that assessment but it's not unbelievable the Russian military at this point in time is more capable than it was in 2022, simply because of the necessity to fight a high intensity conflict in Ukraine. However, long term Russian war fighting capacity is still being degraded due to the accumulating effects of battlefield losses, economic sanctions and their inefficient war economy. The more we can deplete their strategic reserves, cause more casualties, and inflict economic damage, the less of a threat Russia is in the long term, and the more time they'll need to re-organize and re-arm before their next military adventure.
So if Ukraine capitulates today and Russia decides to immediately shift to attacking NATO, then, yes, that would indeed be a poor outcome. If we keep supporting Ukraine and Russia gives up because of war exhaustion, that's the best outcome. If we keep supporting Ukraine and it becomes a frozen conflict, it's still a good outcome because Russia is still occupied with Ukraine. If we keep supporting Ukraine and it eventually loses in a year or two, that would be suboptimal but still better than forcing Ukraine to capitulate now since it will keep Russia occupied for longer, depleting more of their resources, and they'll need a longer recovery period to reconstitute their strength before they can think about attacking the countries we really care about.
UwU
20d ago
The Baltics exist. They are a lot weaker than Ukraine. By helping Ukraine, we do two things:
- Demonstrate our credibility in defending countries against Russian aggression
- Degrade Russia's war fighting capacity
Both of which deters Russia from messing with the Baltics later on. Yes, the Baltic states are in NATO, which precisely makes it worse, because fighting over them has a good chance of leading to WW3 or nuclear Armageddon, with a far higher likelihood than a war in Ukraine. So in this case, "we fight them over there so we don't have to fight them over here" makes sense. It's also not even "we fight them over there", it's "we give the Ukrainians weapons so they can fight them over there" which makes the calculus even better.
UwU
20d ago
I don't really understand what desperate circumstances actually compelled the frog to aid the scorpion against the frog-nazis. Or is the causal relationship flipped, like, the frog must be desperate otherwise it would not be aiding the scorpion?
UwU
20d ago
You can enable private mode by going into https://www.themotte.org/settings/profile and turning it on at the bottom. It prevents people from clicking on your profile to see your comments. Maybe that'll help?
UwU
20d ago
Interesting parable, but it doesn't really establish why the frog was desperate to aid the scorpion against the frog-nazis.
UwU
20d ago
If I had complete control, I would continue the intelligence sharing, military materiel transfer to Ukraine and sanctions on Russia. No security guarantees, no troops on the ground, no direct air support, intel is fine. In terms of magnitude, I would probably give more because Ukraine is currently losing, and they'll need additional help to hold the line. The goal is to bleed Russia and teach them that aggressive wars do not pay. Which makes it less likely that they will think about trying the same against the Baltics and Finland, who are actually NATO members. Which in turn makes it less likely for the US and Russia to go to war against each other down the line.
In terms of falsifiability, I guess if it's discovered that Ukrainian oligarchs are squirreling away a large portion of the aid money, that would make further aid a bad idea. Some corruption is expected since Ukraine was itself corrupt, but if I had to put a number to it, we'll need to rethink it if more than 20% is actually ending up in someone's swiss bank account.
From my perspective, I see almost an equal amount of Ukraine skeptics and Ukraine supporters on this forum.
UwU
23d ago
I don't know about the blue tribe normie but if we're doing alt history I vote for not doing the mail-in strat in 2020. Without it Trump would have eeked out a meager victory in 2020: Trump 1 was lame and an immediately consecutive second term, with Pence VP and no Musk would have been much of the same if not even worse.
I might be one of the more blue-tribe-normie-adjacent person here and I think this is accurate.
UwU
23d ago
Trump just halted military aid to Ukraine.
If Zelensky was making a play to see if he has more power than Trump, then he was consuming his own propaganda. So what if he has sympathizers in congress or the general American public is in favor of aid? If Trump unilaterally halts it, what can congress or the American people do about it? The answer is... nothing. Ukraine aid is simply not a contentious issue in America, even if the average American supports Ukraine, or is in favor of aid. The most they'll do is probably go "oh well, sucks for Ukraine" and go on about their day. Sure, the blue tribe will probably use this as an opportunity for more "Trump bad!" on social media, but that's basically the SOP for everything Trump does at this point.
UwU
26d ago
Is this a prediction Ukraine will lose the war in six months? Because they left it.
UwU
26d ago
True, Vance isn't the final decision maker, but he was instigating the exchange during the meeting.
UwU
27d ago
This is really not giving Trump and Vance enough credit. Whatever you feel about the mineral deal, cease fire, American support, none of that really matters here if the only differentiating factor is how much abuse can Zelenskyy personally take. I believe that at least Vance is capable of separating the merits of these international deals from how servile Zelenskyy pretends to be today. As in, if the deal is bad, then it shouldn't be signed regardless if Zelensky is sucking up to them today or not. And vice-versa if the deal is good.
UwU
27d ago
The inevitable conclusion for this framing is that our leaders are also man-children but they are allowed to be because we are more powerful. Which is fine, that's how the world works, but morally it amounts to throwing stones from a glass house and it's far better to be the adults in the room.
The mineral deal is a deal that helps Ukraine, not the US. We don’t need their minerals. They need us to come and protect them from Russia.
I don't think this is true. The mineral deal that I found said nothing about military assistance or security guarantees. Instead, by my brief reading, it only creates a Fund that both the US and Ukraine can contribute to which reinvests into Ukraine with the US benefiting from the proceeds of this investment. The idea, of course, is that if the US invests in Ukraine, then we are invested in their future and we'll help them to win the war, or, being realistic, at least to not lose too badly. However, there are no mandate for the size or timeline of the US investment, which is problematic given how mercurial our current administration is. To Ukraine, this deal probably doesn't offer anything other than possibly appeasing Trump while concretely signing away natural resource revenues.
UwU
28d ago
How does your interpretation jive with Wong Kim Ark? His parents were Chinese subjects, so if they were living elsewhere in the world, they wouldn't be morally obliged to follow the US's rules. If the US has a draft, they wouldn't have been called, but if the Chinese Emperor calls on their service, they're obliged to follow. Do you think the court decision was wrong, in this case? (totally an option by the way)
I haven't tried Claude 3.7 for creative writing but it's definitely better than the existing DeepSeek-V3 from my limited experience with it, so feel free to test it out. It's a lot less repetitive at longer context lengths which actually makes it usable for creative exercises. The original DeepSeek-V3 was likely more optimized for multi-shot prompting for factual queries, which made it strictly follow the reasoning, tone and structure of earlier examples. Good for factual determinations but not so good for being creative and non-repetitive.
More options
Context Copy link