Google revolutionizes ML engineering

Dear Readers,

How fast does the future sometimes feel? While yesterday we were still laboriously writing lines of code or training robots in hours of trials, today we see systems that do this work in minutes – and often perform better than humans. With agents such as Google's MLE-STAR and the first truly universal robot brains from Skild AI, the door to a new era of automation is wide open. The question is no longer whether machines can relieve us of some of our workload, but how radically they will change our work and our everyday lives.

In this issue, we take a look at the most exciting fault lines of this development: from MLE-STAR, which automatically wins 63% of Kaggle medals, to Skild's general-purpose robot brain that climbs stairs and assists humans, to humanoid robots preparing for the first World Games. We also show how companies such as Figure.02 are already filling washing machines autonomously and which trends are currently driving the S&P 500 almost exclusively with tech companies. If you want to know where AI and robotics are really headed, start here.

In Today’s Issue:

All the best,

The days when machine learning engineers had to spend months writing code and experimenting are over! With MLE-STAR, Google has developed a groundbreaking AI agent that has won medals in 63% of Kaggle competitions – and it does so fully automatically.

MLE-STAR (Machine Learning Engineering via Search and Targeted Refinement) is no ordinary code generator. The agent searches the internet for state-of-the-art models, creates a solid foundation from them, and then refines them in a targeted manner through intelligent ablation studies. Instead of blindly relying on outdated libraries, as previous approaches have done, MLE-STAR automatically finds the latest, most powerful models for each task.

The ensemble strategy is particularly fascinating: MLE-STAR creates several candidate solutions and combines them not by simple voting, but by developing innovative ensemble strategies itself. In addition, the system monitors itself through integrated debugging, data leakage, and data usage checkers.

What does this mean for the AI community? For the first time, smaller teams and individual developers can also access enterprise-level machine learning. The open source version is already available through Google's Agent Development Kit, heralding a democratic revolution in ML engineering.

Imagine if every data scientist had access to a virtual senior ML engineer who knows and applies the latest research findings around the clock.

Why it matters: MLE-STAR dramatically lowers the barrier to entry for high-quality machine learning and could accelerate innovation across all industries. Because the system automatically integrates the latest models from the web, it continuously improves as AI research advances—a self-reinforcing cycle of innovation.

Sources:

🔗 Google Research Blog: https://research.google/blog/mle-star-a-state-of-the-art-machine-learning-engineering-agents/

🔗 ArXiv Paper: https://arxiv.org/abs/2506.15692

🔗 MarkTechPost: https://www.marktechpost.com/2025/08/02/google-ai-releases-mle-star-a-state-of-the-art-machine-learning-engineering-agent-capable-of-automating-various-ai-tasks/

🔗 Medium Analysis: https://medium.com/data-science-in-your-pocket/mle-star-a-deep-dive-into-machine-learning-engineering-with-search-and-targeted-refinement-faed1adaae29

TLDR covers the best tech, startup, and coding stories in a quick email that takes 5 minutes to read.

No politics, sports, or weather (we promise). And it's read by over 1,250,000 people!

Subscribe for free now and you'll get our next newsletter tomorrow morning.

The performance of the S&P 500 is driven almost exclusively by tech, and in particular by CapEx. Without NVIDIA, Microsoft, Apple, Tesla, Meta, etc., there would hardly be any growth for the most important US index.

Skild AI recently announced on its blog the development of “Skild Brain” – a groundbreaking robotics foundation model that works across tasks and hardware, from humanoid, quadrupedal, and mobile manipulators to table arms. Trained on simulations and human videos, then fine-tuned with real-world data, it masters complex actions such as climbing stairs, balancing, and interacting with humans – and includes built-in safety mechanisms. This true omni-body AI promises a new era of physical general AI with great potential for automation in logistics, manufacturing, and inspection.

China is preparing for the first edition of the World Humanoid Robot Games in Beijing (August 15–17, 2025), where more than 20 countries will pit humanoid robots against each other in competitions such as soccer, athletics, dance, industrial, and medical tasks. Among the participants is the humanoid robot T1 from the Hephaestus team at Tsinghua University, gold medal winner of the RoboCup Humanoid League in Brazil. The event promotes perception, decision-making, and control in physical agents and serves as a testing ground for technology with industrial and domestic potential.

A revised version of their last industrial robot: this one is now much lighter and has a range of around 20km!

Figure is known to be one of the most promising robotics companies in the world. Alongside Tesla and Unitree, Figure regularly demonstrates what it can already do today. Now its CEO Brett Adcock has presented a clip showing how Figure.02 independently fills a washing machine - a task that is more difficult than you might think, as it requires a great deal of sensitivity. A sensational achievement!

Are you building the future of robotics powered by AI? We’re featuring projects at the intersection of artificial intelligence and robotics in Superintelligence, the leading AI newsletter with 200k+ readers. If you have exciting product presentations about your robotics products or significant breakthroughs, please show them to us.

Submit your research or a summary to [email protected] with the subject line “Robotics Submission”. If selected, we’ll contact you about a possible feature.

OpenAI employees create mood and anticipation for presumably GPT-5 this week

Just as OpenAI's Aidan also increases the anticipation.

AdQuick unlocks the benefits of Out Of Home (OOH) advertising in a way no one else has. Approaching the problem with eyes to performance, created for marketers with the engineering excellence you’ve come to expect for the internet.

Marketers agree OOH is one of the best ways for building brand awareness, reaching new customers, and reinforcing your brand message. It’s just been difficult to scale. But with AdQuick, you can plan, deploy and measure campaigns as easily as digital ads, making them a no-brainer to add to your team’s toolbox.

You can learn more at AdQuick.com