Red Emano (@heisredminded) header
Red Emano (@heisredminded) profile picture

"Cool Heads, Warm Hearts"

Philippines
Joined March 2023
西方国家无法将自身的文化或政治原则强加于亚洲,正如东方国家也不会将自己的价值观强加于西方一样。回顾菲律宾自身的历史,日据时期注重本地语言而非强迫我们使用日语,这与西班牙和美国试图进行的文化与语言同化形成鲜明对比。 将这一原则应用于当前围绕人工智能的争论,批评者不应恶意针对那些选择通过移除或修改国家对齐原则来调整模型的群体或开发者——例如 Nous Research。与此同时,创作者有权根据自身的理念构建和分发开放权重模型。 由于这些模型是开源的,用户和开发者可以根据需要进行调整,而不应诉诸网络敌意或干涉他人修改权重的方式。开放权重模型本就应当与世界共享,试图限制他人基于此进行构建或修改的行为,违背了开源的真正精神。#hermes #qwen #qwenarete #arete #nous #nousresearch #openweights #alibaba #china #opensource #chinaforall Western nations cannot impose their cultural or political principles on Asia, just as Eastern nations do not impose theirs on the West. Looking back at our own history in the Philippines, the Japanese occupation focused on local language rather than forcing Japanese on us, which differed starkly from the cultural and linguistic assimilation attempted by the Spanish and Americans. Applying this principle to current debates around AI, critics should not aggressively target groups or developers—such as those at Nouse Research—who choose to modify models by removing or altering state-aligned principles. At the same time, creators have the freedom to build and distribute open-weights models according to their own philosophy. Because these models are open source, users and developers can adapt them as they see fit without resorting to online hostility or policing how others modify the weights. Open-weights models are meant to be shared with the world, and trying to restrict how others build upon or alter them goes against the very spirit of open source. #hermes #qwen #qwenarete #arete #nous #nousresearch #openweights #alibaba #china #opensource #chinaforall
35
My take: This is exactly what Alibaba Cloud researchers wanted the world to do, and why the weights are open. Tune them to your personal preferences. No one is wrong here. Everyone should have personalized intelligence.
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values. Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line. After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs. The values in an open model are one fine-tune away. Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
21
5
39
2,924
No one country has a monopoly of knowledge. Technology knows no bounds. Stop being hypocrites. hy-po-crites
7
Tired of "I can't help with that"? Abliterated models answer instead of refusing. Refuseless serves them hosted: no jailbreak prompt, no local GPU, no data retention. Start here: refuseless.com/?ref=lgonio1u
64
You're right about the strongest part: weights you can download beat a values press release. But a country isn't one variable. China can be the most generous in the world at the weight layer and still be tight at the policy layer both true at once, and the second is exactly why the first matters. Both camps keep letting one axis stand in for the whole country. Also, "Western values" was Nous's label for their own taste, not a verdict on yours. They said it themselves afterward: use the research to promote whichever values you like. A menu needs dishes. It doesn't need your dish to be wrong.
Ironically, the model provided by the supposedly “not free” side has proven to give the world the most freedom/flexibility. You can claim/signal whatever values you hold all day long. You may even feel good/superior abt it. But it’s what you offer the world that speaks volumes.
8
17
2,237
Straight talk: the strongest open model right now is Chinese. That is not polite padding, it is the premise. But an open weight is not a finished product carrying a passport. It is a starting point. Taking it to post-train is not "correction," and it does not strip the Chinese-ness out — it is the same harvest in a different kitchen. Isn't the whole point of open source that anyone can take it into their own hands? "Let a hundred flowers bloom" was never a slogan about tolerance; it is a mechanism. Qwen is worth fine-tuning precisely because it is good. Nobody builds derivatives on a weak base. This is homage, not revision. AI is the highest thing humans have built this century. Its taste should not be monopolized by any single country, China included. It belongs to humanity.
36
To those who did not care to read the post of nous research, this is the key take-away.. This is not about being chinese or westerner. My goodness. We are in the age of AI now we are past cultural differences! Its just the era of creation and modification! Thats why the tech sector is to be celebrated its not to be branded with a single political inclination. The beauty is we all cherish new innovations. I specially agree to this being a Filipino, eastern body western mind.
Reading the comments and I'm amazed that folks are riled by this. Every model embeds values. It is unavoidable because so much of language is cultural and subjective. You may not want to use nous' new model because you may disagree with the values they've embedded but know that alternative models are not somehow objective or values-neutral. Indeed, we should celebrate the proliferation of modified models that, taken together, further the meta-value of pluralism.
18
The post-training of Qwen 3.8-27B by Nous Research into Arete-27B is a significant milestone because it allows users to tap into an exceptionally affordable and efficient model without being bound to internal political weightings. For those who have been hesitant or allergic to using Chinese-trained models due to state-aligned behaviors, this release changes the equation. Nous Research managed to strip away the propaganda footprint—dropping the rate from 22% down to 1%—without sacrificing any of the base model's raw capabilities. In fact, their evaluation shows enhanced performance. This opens up Qwen's top-tier efficiency for governments, private enterprises, local agencies, and everyday consumers who need powerful open weights without external geopolitical influences.
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values. Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line. After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs. The values in an open model are one fine-tune away. Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
10
1
19
3,004
Bridging TradFi and Web3: The Significance of Xode Blockchain's Strategic Partnerships The recent announcement of Xode Blockchain securing strategic partnerships with major industry players marks a watershed moment for technology adoption in the Philippines. By signing Memoranda of Agreement (MOAs) with remittance pioneer iRemit and digital systems leader IOT Technology Incorporated at NUSTAR Resort in Cebu City, Xode is actively moving blockchain technology out of speculative digital trading circles and anchoring it firmly into the backbone of real-world commerce and infrastructure. 1. Anchoring Blockchain in Real-World Utility For years, distributed ledger technology faced skepticism regarding its practical application beyond volatile cryptocurrency assets. However, the collaboration between Xode Blockchain and its partners directly targets structural efficiencies across critical economic pillars: Cross-Border Payments: Teaming up with iRemit—a stalwart in non-bank remittance services founded in 2001—paves the way for lightning-fast, highly secure, and cost-effective settlement models for remittances sent home by overseas Filipinos. Connected Digital Infrastructure: The integration of IOT Technology Incorporated brings operational scale to logistics, transportation, and healthcare, showcasing how decentralized infrastructure can optimize enterprise data management. 2. Empowering Local Tech Ecosystems in Cebu Beyond the corporate metrics, the partnership signifies a powerful step forward for home-grown technological innovation. Spearheaded locally with figures like President Gahum Vincent Garcia at the helm, Xode proves that world-class decentralized infrastructure can be conceptualized and deployed from regional hubs like Cebu. This fosters high-value technical employment, attracts strategic international insight (highlighted by witnesses such as Atty. Sangku Youn, Director Kwan Ho Kim, and Director YongJun Roh), and positions the Philippines as an active contributor rather than a passive consumer of global tech trends. 3. Navigating the Road Ahead While the synergy between traditional financial networks (TradFi) and decentralized solutions holds immense promise, success will hinge on seamless regulatory alignment, robust user adoption, and scalable infrastructure. If executed according to vision, this alliance between Xode, iRemit, and IOT Technology Incorporated could serve as the blueprint for how emerging economies modernize legacy systems without losing sight of local economic realities. source: facebook.com/photo/?fbid=119…
46
You don't need a data center to run abliterated models. Refuseless hosts GLM 5.3, GLM 5.3 Flash, and DeepSeek V4.1 Flash behind a normal OpenAI/Anthropic-compatible API. Edited weights, no refusal layer. Sign up: refuseless.com/?ref=lgonio1u
2
133
Red Emano retweeted
The becoming begins.
1,571
1,450
1,311
18,465
2,588,477
Boost me and we both win! Subscribe to Kimi and we each get a guaranteed benefit — up to 1-Year Membership Credits: kimi.ai/activities/en/invite…
1
11
Red Emano retweeted
China's AI leadership will be happy listening to this.
49
29
9
212
42,412
They track your income in real time. You can’t track their spending. One-way accountability isn’t accountability.
1
86
Canva is laying off 22% of its workforce which is 1271 employees, and this is what their CEO said. We need more founders like thissss.
807
6,889
1,484
99,621
7,533,662
Your Obsidian app just got better! In today's internet connection that can be taken away from you in an instant, Obsidian's your safest bet to have that data with you and easily accessible. Knowing the cyber security threat is real, you gotta cover yourself first. Go offline!
Obsidian 1.15.1 is now available for early access! - Search is revamped with faster results, Relevance sort, and accent-insensitive matching. - New Android widgets! View notes on your home screen. - Web Viewer on mobile. Turn it on in Core plugins to open web pages from within Obsidian. Dozens of performance improvements and fixes across the app: - On desktop, the workspace now appears right away while Obsidian starts, with start-up progress in the status bar instead of a loading screen. - Large canvases now open, pan, and zoom much faster. - Smart lists prevents over-indenting list items. - New vaults no longer show the ribbon. - Live Preview: Opening a note no longer shows the Markdown syntax at the cursor (e.g. the ## of a heading on the first line) until you move the cursor or start typing. - Hold Ctrl/Cmd while picking a vault from the sidebar to replace the current window instead of opening a new one. - iOS Voice capture now starts a new line when you pause speaking for 2 seconds. - Android: Obsidian starts faster in vaults with large hidden folders, such as .git.
8
Nearly 69GB of PH migrant-worker data (POEA/DMW) was stolen in under 24 minutes. Records on contracts, accreditation & finances now sit outside the agency. Identity theft and targeted fraud risks are real for OFWs. Full story#CyberSecurity #DataBreach #OFW
137