Good morning, it’s Tuesday. Nokia’s CEO is shaking up the data center world in pursuit of greener AI—a move that fits right into Finland’s goal of being carbon-neutral by 2035.
Meanwhile, DeepSeek is stepping up as a challenger to OpenAI, and developers are donning managerial hats thanks to AI tools. Plus, Trump’s record-breaking fundraising puts tech and crypto in the spotlight.
Oh, and don’t miss 👾The Magic of Prolonged Thinking: Test-Time Compute Part 2, now live. Read on!
🗞️ YOUR DAILY ROLLUP
Top Stories of the Day
🤖 DeepSeek's R1 Rivals OpenAI with Massive Model
DeepSeek's R1 model challenges OpenAI's o1, excelling in benchmarks like AIME and MATH-500. With 671 billion parameters, R1 is open-source, cost-effective, and tailored for math and science tasks. Distilled versions support smaller devices, but responses comply with Chinese regulations. As U.S.-China tensions over AI tech exports rise, DeepSeek joins Alibaba and Moonshot AI in the competitive race for global AI leadership.
💉 Isambard-AI Supercharges UK Vaccine Development with AI
The £225m Isambard-AI supercomputer in Bristol is revolutionizing drug and vaccine development with cutting-edge AI. Operational this summer, it will rank among the world’s top 10 fastest supercomputers. Already advancing treatments for Alzheimer’s, cancer, and heart disease, Isambard-AI simulates molecular drug interactions to fast-track testing. Despite its energy demands, it plans to repurpose waste heat for local homes, bolstering its groundbreaking potential.
🤐 Epoch AI Faces Backlash Over OpenAI Funding Secrecy
Epoch AI, creators of the FrontierMath benchmark, faced criticism for delaying disclosure of OpenAI funding. Showcased with OpenAI’s o3 model, FrontierMath's credibility was questioned due to perceived secrecy. Critics argue this undermines trust and objectivity, despite assurances that OpenAI won't use the data for training. Epoch AI acknowledged missteps, sparking debate about transparency and impartiality in AI benchmarking amidst rising scrutiny.
💰 Tech and Crypto Leaders Propel Record Trump Fundraising
President-elect Trump’s inaugural committee has raised over $200 million, smashing records with contributions from major corporations and tech and crypto leaders. Their prominent roles in the celebrations spotlight concerns over big money’s influence on politics, especially in high-stakes industries like AI and crypto. Critics caution that such trends risk blurring lines between public interest and private power in policymaking.
🌱 GREEN DATA
Nokia CEO Calls for Smarter Data Centers to Address AI’s Sustainability Challenges
The Recap: Nokia CEO Pekka Lundmark highlights the growing energy demands of AI-powered data centers and emphasizes the need for smarter, more sustainable infrastructure designs. By leveraging innovations like efficient networking, localized data processing, and smaller, purpose-driven AI models, Lundmark envisions a more sustainable future for AI-driven technologies.
AI has the potential to reduce global emissions in industries like manufacturing and logistics, but data center energy consumption in the U.S. could triple by 2028.
Nokia’s Finland headquarters data center uses waste heat to warm 14,000 homes, showcasing sustainable energy reuse.
Innovations in optical networking have cut power consumption by 60% per bit, though global network traffic is expected to grow significantly by 2033.
Distributed and edge data centers are expected to grow in response to data sovereignty laws, low-latency demands, and industrial AI applications.
Nokia Bell Labs is focusing on small language models (SLMs) to save energy compared to large language models (LLMs), reducing energy use by up to 50% with minimal accuracy trade-offs.
AI pruning techniques inspired by human brain development streamline neural networks and further lower energy consumption.
NVIDIA and other partners suggest building data centers in locations with access to renewable energy, ambient cooling, and cheaper real estate.
Forward Future Takeaways:
As AI adoption accelerates, the industry faces an urgent need to balance performance with sustainability. Distributed and localized data centers, energy-efficient networking, and the development of leaner AI models will play critical roles in addressing these challenges. By focusing on smarter infrastructure and sustainable innovation, companies like Nokia are paving the way for an AI-driven world that minimizes environmental impact without compromising growth. → Read the full article here.
👾 FORWARD FUTURE ORIGINAL
Train-Time Compute, Test-Time Compute, and Test-Time Training
“Our large-scale reinforcement learning algorithm teaches the model how to think productively using its chain of thought in a highly data-efficient training process. We have found that the performance of o1 consistently improves with more reinforcement learning (train-time compute) and with more time spent thinking (test-time compute).”
OpenAI
Train-Time Compute (TTC) refers to the total computing power required during the training phase of a model, i.e. the period in which the model recognizes patterns in a large amount of data and learns from them. A particularly important part of this is the pre-training, in which the model is usually trained on a wide range of data sets to develop a general understanding. Methods such as reinforcement learning are often used before or after this pre-training to improve the model in a targeted way. In addition, the concept of “Chain of Thought” (CoT) plays a role, in which the model learns to answer complex questions in several mental steps and thus expand its ability to draw conclusions.
After training comes the deployment phase, in which a distinction is made between test-time compute (TTC) and test-time training (TTT). In test-time compute, the model uses only the skills it has already learned during application. It processes new input and provides output (e.g. answers in the case of language models) without changing its internal parameters – CoT can be used here to achieve more precise results through multi-level thought patterns. In contrast, with test-time training, the model does not remain static: it continues to learn during use by completing additional training steps. A language model, for example, could adapt to a new dialect and thus improve its responses step by step. However, this process is more computationally intensive because the model actively updates its parameters. In other words, while test-time compute is like a musician playing their practiced notes, test-time training is like a musician learning a new instrument in the middle of a concert in order to be able to react even more flexibly. → Continue reading here.
👨💻 AI CODING
The Second Wave of AI Coding: A Race Towards Smarter Software Development
The Recap: A new generation of AI coding tools is transforming software development by moving beyond simple autocomplete to creating, debugging, and testing code autonomously. These innovations are not just reshaping programming but are being seen as potential stepping stones toward artificial general intelligence (AGI).
Over a quarter of all new code at Google is generated by AI, demonstrating how AI tools are speeding up development processes.
Startups like Poolside, Cosine, and Zencoder are building advanced tools that aim to mimic how human developers reason about code, rather than just autocompleting it.
First-generation tools focused on producing syntax-correct code, but the new generation targets functional correctness, ensuring programs do what they are intended to do.
Companies are creating custom datasets that capture the logic of how code is built, rather than just training on finished code from repositories.
Techniques like reinforcement learning from code execution (RLCE) allow models to self-improve by simulating and testing their own code, much like DeepMind’s AlphaZero learned to master games.
AI coding tools are shifting the role of developers toward problem-solving and oversight, allowing for rapid prototyping and debugging at unprecedented speeds.
Fewer programmers will be needed to maintain large codebases as AI tools become more capable, potentially reducing tech workforce sizes while increasing efficiency.
Forward Future Takeaways:
The next wave of AI coding tools is pushing software engineering into a new era where human input focuses more on managing and guiding AI-driven processes rather than direct coding. With companies like Poolside and Cosine positioning code generation as a stepping stone to AGI, these tools may have profound implications beyond software, paving the way for machine-driven reasoning across industries. While efficiency will rise, the economic and workforce implications of reducing the need for traditional coding teams will demand careful consideration in the years to come. → Read the full article here.
🛰️ NEWS
Looking Forward
📈 Spain Allocates $155M for AI Growth: Spain pledges €150M to accelerate AI adoption among companies, boosting innovation and tech integration. PM Pedro Sanchez announced the initiative.
🏆 Best AI Tools for Coding in 2025: ChatGPT Plus and Perplexity Pro lead as the top AI coding assistants, excelling in real-world programming tests. Meanwhile, Grok impresses as a free contender to watch.
🤔 Less AI Knowledge = More Openness: Research reveals people with lower AI literacy are more receptive to it, viewing it as "magical." Policymakers face balancing education and maintaining wonder.
💻 AI Supercomputing for Laptops: DIMON, a new AI tool, solves complex equations in seconds, transforming laptops into high-performance tools. It empowers breakthroughs in engineering, healthcare, and design.
🔬 RESEARCH PAPERS
VideoMaker: Enhancing Zero-Shot Video Generation with Video Diffusion Models
A new approach, VideoMaker, leverages the inherent capabilities of Video Diffusion Models (VDMs) to achieve zero-shot customized video generation, eliminating the need for additional feature extraction models. Unlike traditional methods, which often suffer from inconsistencies in subject appearance due to suboptimal techniques, VideoMaker utilizes VDM’s pre-trained intrinsic feature extraction for fine-grained subject features that align closely with its existing knowledge.
The framework also introduces a bidirectional interaction mechanism using spatial self-attention within the VDM. This ensures enhanced subject fidelity while maintaining diversity in the generated video content. Experiments demonstrate the effectiveness of this method for generating customized videos of both humans and objects, setting a new benchmark for zero-shot video synthesis. → Read the full paper here.
📽️ VIDEO
Test-Time Scaling Will Be Much Bigger Than Anyone Realizes
Test-time compute revolutionizes AI by enabling models to "think" during inference, improving long-term reasoning and performance. This approach boosts tasks like image generation and logic-based challenges but comes with high costs. Scaling inference promises transformative advancements across industries as costs decrease and capabilities expand. Get the full scoop in Matt’s latest video! 👇
Reply