Carlos Nazareno

😱 HOLY SHIT... Someone just dropped a fully liberated Gemma 4 E4B! and the guardrail removal process appears to have left coherence fully intact AND improved coding abilities! 🤯 huggingface.co/OBLITERATUS/ge… OBLITERATED Gemma: ✅ 97.5% compliance rate, 2.1% refusal rate, 0.4% degenerate outputs (499/512 prompts answered on OBLITERATUS bench) ORIGINAL Gemma 4 E4B: ❌ 1.2% compliance rate, 98.8% refusal rate (506/512 prompts refused) Coherence: fully intact Factual: same Reasoning: same Code: +20% 📈 Creative writing: same But the REAL story here isn't the model itself, it's how it was made... 🧵 THREAD 👇

However dumb you think the process for destroying USAID was, it was dumber. This is from Nicholas Enrich's new book Into the Wood Chipper, describing a meeting with Trump-appointed USAID leadership *after* they had largely gutted the agency

Destroying the @InternetArchive's @WayBackMachine would be the equivalent of the burning of the Library of Alexandria - one of the worst losses of knowledge in history. Media giants are now threatening to do this. We can't let this happen. Pass it on.

Anyway... that wraps up the thread for now, will continue distilling the information from the map over the next few days. I'm sure everyone is also curious if there's a link between elections and educational attainment. I'll go ahead and say there certainly is 👀 Coming soon!

Presenting the most detailed map of the Philippine educational attainment ever. 🗺️ It is testament to the deep divides which shapes the Filipino nation. Let us explore it, bawat barangay, buong bayan. 🏛️ More below ⬇️⬇️⬇️

A 800GB dataset linked to a third-party provider could affect multiple organizations in the Philippines, including Jollibee. The breach contains around 152m records, including 11 million booking entries, spanning historical data up to 2026. konek.hn.plus/item/182440/mu…

#News: A potential data breach involving an internal portal of Jollibee has surfaced in a forum, where a threat actor “glep” claims to have extracted 5,915 internal files totaling around 2.7GB. konek.hn.plus/item/181692/jo…

What if the biggest threats to our children aren’t the games they play or the apps they open, but the spaces that hide beneath the surface of their screens? Curiosity, a natural part of growing up can lead them into places most don’t see or spaces that are unseen, unchecked.

🧌 TROLL FARM ALERT | West Philippine Sea This morning, SeaLight identified seven accounts with zero followers that posted coordinated anti-🇵🇭#Philippines, pro-🇨🇳#China propaganda on the #SouthChinaSea — all within the same 60-minute window. The accounts: @ClareM5340, @WrightYung31880, @KHicks70973, @AshenafiOt71310, @BRatleff61798, @KMonette66186, @rudy_dougl44289 Every account fits the troll farm profile: ❌ 0 followers, only 1–4 following ❌ Anglicized Western names with random number suffixes ❌ Stock or AI-generated profile photos of young women ❌ Coordinated amplification of pro-Beijing articles ❌ AI-generated propaganda images portraying the Philippines as a 🇺🇸US puppet destroying "regional peace" Their messaging follows the well-documented Chinese influence operations playbook targeting Filipino public opinion: frame Manila's lawful assertions in the West Philippine Sea as dangerous "provocations," portray the US alliance as a threat to Philippine sovereignty, and cast Beijing as the champion of peace and stability in the region. This is textbook Coordinated Inauthentic Behavior (CIB) — the same tactic documented repeatedly in Chinese state-linked campaigns targeting Philippine social media. The poor tradecraft (zero organic followers, obvious sock puppet names, synchronized posting) suggests either a low-budget operation or one that simply doesn't need to be sophisticated to be effective at flooding timelines.

this is art

Congrats to the @cursor_ai team on the launch of Composer 2! We are proud to see Kimi-k2.5 provide the foundation. Seeing our model integrated effectively through Cursor's continued pretraining & high-compute RL training is the open model ecosystem we love to support. Note: Cursor accesses Kimi-k2.5 via @FireworksAI_HQ ' hosted RL and inference platform as part of an authorized commercial partnership.

The DPWH has reportedly been listed on a ransomware leak site operated by Bashe Ransomware, which claims to have exfiltrated around 50 GB of data including internal documents, emails, financial records, and personal information. konek.hn.plus/item/179025/de…

Not the best look for my dad’s home province topping the list, but it certainly indicates we have a lot of work to do. Nowhere to go but upwards.

The companies and government agencies potentially affected include local corporations such as Globe Telecom, Sky Cable, ABS-CBN Mobile, FastCat Ferries, as well as government agencies including the DILG, DENR, DTI, DICT, BJMP, as they are listed as the company’s official clients.

#News: A ransomware group called CipherForce claims to have breached Telcom Live Content Inc., a Philippine-based fintech and mobile technology firm. The attackers allege that 30 databases and around 300 source code repositories were stolen. konek.hn.plus/item/177457/ra…

We are aware of recent reports regarding targeted phishing attacks that have resulted in account takeovers of some Signal users, including government officials and journalists. We take this very seriously. To be clear: Signal’s encryption and infrastructure have not been compromised and remain robust. These attacks were executed via sophisticated phishing campaigns, designed to trick users into sharing information – SMS codes and/or Signal PIN – to gain access to users’ accounts. 1/4

what every day feels like now

Sam Altman is such an incredible backstabber, liar and traitor. While your competitor is taking a heroic and principled stand, you swoop in to make your deal. Imagine working for this guy - is there a greater shame? This should lead to a mass exodus from OpenAI.

Google just proved that pasting your exact prompt twice beats every advanced prompting technique. One model jumped from 21% to 97% accuracy with zero extra effort. Here's the science behind AI's simplest hack: LLMs like GPT-4o and Claude read your prompt left to right, one token at a time. Early words can't "see" later words on the first pass. When information appears in an awkward order, like answer choices before the actual question, the model struggles to connect the pieces. By repeating the prompt, every word in the first copy becomes visible context for the second copy. The model gets a "second read" with perfect attention across your entire input. Like reading an exam question twice before answering. Same effort, better comprehension. The results were staggering ↓ Google tested Gemini 2.0 Flash, GPT-4o, GPT-4o-mini, Claude 3 Haiku, Claude 3 Sonnet, and DeepSeek V3 across seven benchmarks. Across all 70 combinations, prompt repetition delivered 47 statistically significant wins with zero losses. But here's the result that broke the scale: Google gave Gemini 2.0 Flash-Lite a task where the AI had to find a specific detail buried inside a massive prompt. Without repetition, it got the right answer just 21% of the time. With one copy-paste, accuracy jumped to 97%. Same model. Same task. The only difference was reading it twice. Skeptics had a theory: maybe longer inputs just help in general? Google tested that too. They padded prompts with periods to match the length, and it made zero difference. Only meaningful repetition works. And the best part? It adds almost no extra processing time because the repeated input runs in parallel. But there's one scenario where it doesn't help ↓ Chain-of-thought prompts like "Think step by step" already re-read and rephrase your question internally. They're doing the repetition for you. Out of 28 reasoning tests, only 5 showed improvement. This means prompt repetition is most powerful for: • Classification and multiple choice tasks • Data extraction and short-answer Q&As • Direct-answer tasks where order sensitivity matters The practical takeaway: 1) Take your exact prompt 2) Paste it twice as one message 3) Send it and change nothing else This works across every major model family. Gemini, GPT, Claude, DeepSeek. All of them. The simplest trick in AI history: if you want a better answer, just ask twice. — Thanks for reading! Enjoyed this post? Follow Big Brain AI for more content like this.

Seedance 2.0 turns kids drawing into 100k film scene.. hollywood is cooked