Denzel Geng
denzelgeng.bsky.social
Denzel Geng
@denzelgeng.bsky.social
17 followers 20 following 21 posts
Indie hacker, building several AI projects
Posts Media Videos Starter Packs
6️⃣ Qwen 2.5 Max from Alibaba! Alibaba's Qwen 2.5 Max inference model is set to debut. Another big win for the open-source community!
5️⃣ Google Veo 2 API drops! 🎬 Google's Veo 2 video model API is opening, promising to outperform all other video generation models. Video creation is about to level up.
4️⃣ Grok-3 API & Voice Mode Unleashed! 🎙️ Expect the Grok-3 API to open up, PLUS Grok's voice mode is arriving. Voice interaction is getting an upgrade.
3️⃣ Deepseek's Open Source Spree! 🔓 5 new code libraries from Deepseek, all focused on infra-related inference components. Open source innovation continues!
2️⃣ Claude 4 is on the horizon! 🗣️ Claude's official site hints at the imminent launch of their new inference model. Get ready for smarter conversations.
We got such a big week coming up in AI! 🚀 Buckle up for these HUGE announcements:

1️⃣ GPT-4.5 is coming! 🔥 Sam Altman announced next week's GPT-4.5 release. Microsoft is already prepping servers for both GPT-4.5 & GPT-5!
xAI is here to shake things up! What aspect are you most excited about? Let's discuss! 👇
5/5 DeepSearch integration? Answers just got DEEP. 🔎 Grok-3 + DeepSearch = info delivering machine! Real-time, comprehensive answers at your fingertips. AI-powered resourcefulness!
4/5 Pre-training ain't dead! Grok-3 is here to prove it. Massive compute + pre-training = HUGE wins! Rewriting the rules and unlocking crazy potential. Scalability? ∞ and beyond! 📈
3/5 200,000 GPUs. Let that sink in. 🚀 xAI's Colossus supercomputer is powering Grok-3 with TWO HUNDRED THOUSAND GPUs! Colossal compute for colossal AI innovation. Scale: UNPRECEDENTED.
2/5 Brain upgrade alert! Grok-3 got TWO modes: "Think Mode" for daily wins & "Big Brain Mode" for impossible stuff. Efficiency + power = beast mode for any challenge. Adaptability level: NEXT.
1/5 SOTA benchmarks are here! Math, coding, reasoning... GPT-4o & Gemini-2 Pro who? New king in town! 👑
xAI has released the new SOTA model Grok-3, so what new? 🤔
4/4 Decoupled Visual Encoding 🧠: Innovation: separate visual encoders for understanding & generation. Task-specific optimization boosts performance in both, overcoming limitations of single encoders.
3/4 Scalability via Data & Model 🚀: "Pro" means scalability. Performance scales with data & model size. Future-proof design leverages data & compute advancements for even stronger multimodal AI.
2/4 SOTA 7B Performance 🔥: Janus-Pro 7B achieves state-of-the-art (SOTA) on benchmarks (image captioning, VQA, text-to-image!). Rivals larger models, highlighting efficient architecture.
1/4 Unified Cross-Modal Synergy 💪: Janus-Pro: shared backbone for text & vision enables deep cross-modal fusion from the start. Nuanced, holistic understanding & generation like humans!
DeepSeek has done it AGAIN 🤯

They've dropped Janus-Pro and it's a paradigm shift in multimodal AI. Janus-Pro achieves true unification, integrating visual and textual processing within a single neural network architecture. 🤯

Here are facts you NEED to know about Janus-Pro: 👇
Trump got second place on a “be like trump” impression contest
Aloha
aloha, tiktok users! just a heads up, here on bluesky, we keep it a little chiller! so if you're having a mental breakdown, don't make a big fuss, just quietly signal the mods by beginning and ending an otherwise unrelated post with the same word. any word will do, even something silly like aloha
Dude certainly looks like he just got a hangover