German DeepSeek Runs 3x Quicker

Welcome to another edition of AI PlanetX.

DeepSeek LLM hits 200% speed; Sutskever becomes CEO of Safe Superintelligence; Xbox exec backs AI Copilots for laid-off workers.

• Hottest AI News

• Top AI & SaaS Tools

• Top AI & Tech News

• AI Trend Watch: How Gemini CLI Makes AI Coding Easy

• AI Art Spotlight

• Prompt of the Day: Prompt to Build AI Chatbot

• AI Video Tutorial

• F-R-E-E AI Course of the Day: Advanced LLM Agents by UC Berkeley

Hottest AI News

DeepSeek

German Lab Unveils Supercharged DeepSeek LLM with 200% Speed Boost

TNG Technology Consulting has released DeepSeek-TNG R1T2 Chimera, an accelerated version of DeepSeek’s R1-0528 model. It maintains most of the original’s reasoning ability while operating at more than double the speed, just a month after the original launched.

Details:

• R1T2 Chimera reaches 90–92% of the original model’s scores while using under 40% of the tokens. It runs over twice as efficiently as R1-0528 and 20% better than January’s R1

• TNG used an "Assembly-of-Experts" (AoE) method—unlike standard MoE—to merge weight tensors from several models without retraining. The “Tri-Mind” setup combines R1-0528’s reasoning, R1’s structure, and V3-0324’s clarity

• Benchmarks on AIME-24, AIME-25, and GPQA-Diamond confirm its strength. Hugging Face’s Vaibhav Srivastav praised the MIT-licensed release hosted on the platform

TNG’s Assembly-of-Experts merges pre-trained weights instead of activating experts per input like MoE. This cuts inference costs while keeping strong reasoning performance.

Marketing ideas for marketers who hate boring

The best marketing ideas come from marketers who live it.

That’s what this newsletter delivers.

The Marketing Millennials is a look inside what’s working right now for other marketers. No theory. No fluff. Just real insights and ideas you can actually use—from marketers who’ve been there, done that, and are sharing the playbook.

Every newsletter is written by Daniel Murray, a marketer obsessed with what goes into great marketing. Expect fresh takes, hot topics, and the kind of stuff you’ll want to steal for your next campaign.

Because marketing shouldn’t feel like guesswork. And you shouldn’t have to dig for the good stuff.

Sign Up Free

SSI Inc.

Ilya Sutskever Steps Up as CEO of Safe Superintelligence Following Co-Founder's Departure

Ex OpenAI co-founder Ilya Sutskever has become CEO of Safe Superintelligence (SSI) after Daniel Gross stepped down on June 29. He also named co-founder Daniel Levy as president of the $32B startup focused on safe superintelligent AI.

Details:

• Meta CEO Zuckerberg reportedly held talks to hire Gross and Nat Friedman, even attempting to acquire Safe Superintelligence. Sutskever said they were “flattered” but remain focused

• Sutskever founded SSI after leaving OpenAI, where he played a role in ousting CEO Sam Altman. SSI focuses solely on safe superintelligence, unlike Meta’s broad AI approach

• As CEO, Sutskever will face challenges like fundraising and hiring, while Gross may take a product-focused role at Meta with other top AI researchers recruited for Zuckerberg’s AI projects

Sutskever is confident about SSI’s future, saying, “We have the compute, the team, and know what to do.” Yet, Gross’s departure raises questions about possible strategic challenges or differences at the AI safety startup.

Top AI & SaaS Tools

• Presenti (Life-time Deal): AI-powered presentation maker that turns text, Word, PDF, Markdown, and mind maps into polished, visually appealing presentations

• Glass: An open-sourced version of Cluely released by another founder in just 4 days, delivers the same real-time meeting assistant features with better output and design [F-R-E-E]

• MiniMax-M1: It’s currently #1 in math on the LMArena leaderboard, thanks to its record-long context window advancing long-context reasoning and tool use [F-R-E-E]

• Pally: AI relationship management unifies all your contacts, helping you track, search, and strengthen key relationships [F-R-E-E]

• Browse Anything: AI agent that can browse the web, automate tasks, extract data, generate reports - all from a simple prompt [F-R-E-E]

Learn AI in 5 minutes a day

What’s the secret to staying ahead of the curve in the world of AI? Information. Luckily, you can join 1,000,000+ early adopters reading The Rundown AI — the free newsletter that makes you smarter on AI with just a 5-minute read per day.

Sign up to start learning.

Top AI & Tech News

• An Xbox executive, Matt Turnbull, suggested using AI chatbots like ChatGPT Copilot to help laid-off workers manage the emotional and cognitive challenges of job loss

• Soham Parekh is an Indian software engineer who went viral for secretly working multiple jobs at Silicon Valley startups at the same time

• xAI's Memphis data center was permitted for 15 methane turbines, but satellite and thermal imaging show 24 running, raising local pollution concerns

• Cluely, an AI real-time meeting notes and analysis tool, has doubled its annual recurring revenue to $7 million after launching an enterprise product

AI Trend Watch

How Google Gemini CLI Is Changing the Game for Developers 🤯

AI Art Spotlight

Model: Midjourney V7

Prompt:

very closeup shot of Tom Holland riding a dust motorbike bike in a tropical forst, red tank of the motorbike visible, his face is contorted with panic, mouth wide open, he has tight grip on the handle of the motorbike he is riding, behind him a huge giant Spider chasing him while tom holland is trying to flee, spider eyes, legs and furs are very clear. It's a sunny day, there is a clear view of tongue, teeth and open mouth of Tom Holland, while he is looking at the camera --ar 9:16 --v 7

Prompt of the Day

Prompt to Build a Simple AI Chatbot with RAG for Beginners

Learn how to build a simple RAG chatbot in a Jupyter notebook with a friendly AI tutor, explaining each step in plain English—from embeddings to using Pinecone's Python SDK and OpenAI models.

Top AI Video Tutorial

These AI Skills Will Give You an Immediate Advantage at Work

Complimentary AI Course of the Day

Advanced Large Language Model Agents by UC Berkeley [Spring 2025]

This course covers advanced topics in LLM agents, including reasoning, AI for mathematics, code generation, and program verification. It introduces advanced inference and post-training for building agents that search and plan, then explores how LLMs can prove math theorems and generate or reason about code.