Qwazic

[Qwazic @qwazzic]

@CodeByNZ Just ask GPT and Claude about the DEI agenda and you'll see the magic of Western censorship. The fact that they refuse to answer politically risky questions says absolutely nothing about a model's actual capabilities.

[Qwazic @qwazzic]

@jun_song So they tried to double the base price with a new "mythos tier" and failed. Now we'll get Opus 5 at the old price, and Fable will just quietly die

[Qwazic @qwazzic]

@AiHubMix Well, at least Kimi K3 did its job an order of magnitude better than the competition.

[Qwazic @qwazzic]

@jun_song What did you expect from the cult? They don't actually use the models or neural networks; they just root for them and worship their idols. It’s exactly like PlayStation vs. Xbox fanboys—all they care about is their chosen brand winning.

[Qwazic @qwazzic]

@zeroday_whales @AndreasSteno Well, when you're paying $50 per 1M tokens instead of just coasting on a subsidized subscription, it's not so funny anymore.

[Qwazic @qwazzic]

@AndreasSteno So don't panic.😂

[Qwazic @qwazzic]

@MacroMicroNMike @irving08392944 @AutomationMikke @haider1 That’s literally how LLMs work: they train on raw data, not on their own identity. Plus, distillation wouldn't make a model think it's Claude—Claude doesn't prefix every response with "I am Claude." You really don't get how this works.

[Qwazic @qwazzic]

@MacroMicroNMike @irving08392944 @AutomationMikke @haider1 Gemini once called itself GPT, and Claude called itself Qwen in Chinese. Turns out, if you don’t explicitly state the model's name in the system prompt, it just hallucinates one, auto-filling the most common or "fitting" name for that language.

[Qwazic @qwazzic]

@pomis172 @kimmonismus Rich imagination. As soon as the Chinese started squeezing them, Opus in their imagination suddenly slimmed down from the fictional 3-5T to a 1.5T. In reality, Musk called Grok 4.5, with its 1.5T, super cost-effective, and he is charging 4 times less for it than what Opus costs.

[Qwazic @qwazzic]

@guilleflorvs Not even close. They didn’t even invent the Transformer. This is easily the most embarrassing tweet I’ve seen in years.

[Qwazic @qwazzic]

@Valuable A cunning plan to get an inheritance. I see no other logic behind this. If this isn't trolling, you need to be locked up in prison.

[Qwazic @qwazzic]

@yacinelearning @oneill_c Musk says Grok 4.5 is a budget model with 1.5T params at $6. Comparing that to Sonnet at $15, Sonnet must be at least 2T–2.5T parameters. Either that, or Claude is severely overpriced.

[Qwazic @qwazzic]

@The_Guy_Space The Chinese approach is just simpler: no need for perfect precision, lower risks, and way easier to scale.

[Qwazic @qwazzic]

@RouleurCapital @thehypedotnews @rohanpaul_ai @OpenAI GLM 5.2 is half the size of Grok 4.5 in terms of weights, utilizing a Mixture of Experts (MoE) system for inference. It’s a matter of optimization, not government support. The model is inherently more economically viable, not artificially subsidized.

[Qwazic @qwazzic]

@Lynnette_Novak So I don't feel like my days were wasted

[Qwazic @qwazzic]

@RonanHunt88 @StockMKTNewz Have you ever heard of Android? Or did your brain's context window collapse to programming only?

[Qwazic @qwazzic]

@iphone15pormax @StockMKTNewz дурак

[Qwazic @qwazzic]

@davis7 What's even more surprising is that nowadays there's a bunch of fools evaluating AI based on coding benchmarks, as if that's the only application for AI. The AI market is way bigger than just a handful of programmers.

[Qwazic @qwazzic]

@bateman_pa12732 Нахуй этот овощ нужен. Если что я про женщину.

[Qwazic @qwazzic]

@haider1 В тех задачах, которые я решаю, искусственный интеллект за последний год фактически деградировал. GPT-4.5 намного лучше справлялся с текстом, чем современные модели. Компании просто забыли, что искусственный интеллект - это не только кодирование.