Wednesday / AI 101 series: Transformers are starting to treat depth the way they already treat sequence: as an addressable dimension – deep dive
Friday / Series: Interview and deep dive with Ali Kani on Alpamayo, AlpaDream, Cosmos, and everything in between that powers NVIDIA’s self-driving systems (plus my honest opinion after the test-drive)
With LlamaParse (from LlamaIndex), you can skip the complexity and leverage proprietary document understanding agents built for agentic OCR and enterprise-scale AI workflows:
Extract reliable data from PDFs, spreadsheets, images, and scanned or handwritten files
Transform unstructured documents into AI-ready knowledge for RAG and agentic applications
Deploy ingestion pipelines that scale without constant maintenance
Trusted by Fortune 500 enterprises like BP, EY, KPMG, the world’s fastest startups like Lovable and Tabs, and over 300,000 developers, LlamaParse helps teams move from raw documents to production-ready AI faster.
Main topic: 100 000 subscribers! (actually, already more than that)
A few news to share.
Even though I have been writing about ML and AI for the past seven years, Turing Post itself is not even three years old yet, and here we are with more than 100,000 readers on the mailing list alone. Our YouTube channel, a much more recent addition, will turn one in just a couple of months and is about to cross 7,000 followers. Strangely enough, YT became my favorite spot for a thoughtful discussion. Our X account has become known for its research coverage, and we are about to cross 90,000 followers there as well.
We are still a very small team: two full-time people, me and Alyona Vert, plus Will Schenk as a contributor. And I do not take any of this success for granted.
Every day, I see names among our free and premium subscribers with such rigor, seriousness, and depth that I feel genuinely humbled. To each and every one of you, thank you. Thank you for reading, subscribing, sharing, commenting, and also for reading silently but steadily. All of it matters.
I also wanted to share a few changes in direction.
I had been working on a series about open-source AI, but this space is still too young, too fluid, and too early for conclusions that would feel truly solid. Rather than pretend otherwise, I would rather wait and do it properly. So instead, we will publish a State of Open-Source AI later this year.
At the same time, your feedback has made a few things very clear. Right now, the strongest interest is around agentic coding and engineering, and around what began almost accidentally as The Organizational Age of AI series. The response to both has been so strong that, for the next quarter or two, they will likely become two of our main Friday series.
And one more addition: going forward, one of the main new pillars of our AI 101 series will be security, including the emerging best practices for building and deploying these systems responsibly.
That also means that in the coming weeks, you should expect some phenomenally interesting speakers from these worlds.
And I mean that sincerely. There were a couple of strange synchronicities for me at NVIDIA GTC, and they left me with a deeper sense of connection to the part of AI that is actually creating standards, building infrastructure, and opening entirely new worlds.
That, in the end, is also what I want Turing Post to keep doing: helping make sense of a field that is moving fast, speaking clearly about what matters, and bringing a bit more structure to places where there is still mostly noise.
Thank you for helping Turing Post become what it is today.
And thank you for making it worth building further.
To celebrate, we offer a 20% discount for the annual Premium subscription. Ends on March 31.
Upgrade today for only $56 / YEAR
You will immediately get access to our most popular:
We are also seeing growing interest in consulting sessions with us. At the moment, you can choose whom you would like to speak with: Ksenia Se or Will Schenk. Get a consultation here.
PS: We’re working on a few things beyond the content we create. Stay tuned.
No Attention Span episode today. I needed some rest after a wildly intense NVIDIA GTC. But I still have something for you: the second part of my interview with Michael Bolin, on the crucial skills developers need today and how to shift your mindset from coding to managing agentic systems.
Our news digest on Mondays is always free. Click on the partner’s link above to support us or upgrade to receive our deep dives in full, directly into your inbox. Join Premium members from top companies like Nvidia, Hugging Face, Microsoft, Google, a16z etc plus AI labs such as Ai2, MIT, Berkeley, .gov, and thousands of others to really understand what’s going on with AI → Upgrade
Follow us on 🎥 YouTube Twitter Hugging Face 🤗
13 Modern Reinforcement Learning Approaches for LLM Post-Training
Hermes Agent – an open-source agent that refuses to stay in its lane
Nous Research has unveiled Hermes Agent, MIT-licensed autonomous agent designed to live on your own server, accumulate memory, and improve over time. It is not masquerading as yet another IDE sidekick or thin chatbot shell: Hermes spans CLI, Telegram, Discord, Slack, and WhatsApp, with sandboxing, automations, subagents, and browser control. Worth checking out.HyperAgents – Explores recursive self-improvement by allowing agents to modify not only their behavior but also the mechanism that generates improvements, pushing toward open-ended learning systems →read the paper
MiroThinker-1.7 & H1: Towards Heavy-Duty Research Agents via Verification – Introduces verification directly into the reasoning loop, making agents audit and refine their own steps instead of trusting a single forward pass →read the paper
OpenSeeker: Democratizing Frontier Search Agents by Fully Open-Sourcing Training Data – Shows that strong search agents are as much a data problem as a modeling one by releasing the full training pipeline and demonstrating competitive performance with limited data →read the paper
Memento-Skills: Let Agents Design Agents – Turns agents into systems that build and refine other agents through evolving skill libraries, without updating base model weights →read the paper
MiniMax M2.7: Early echoes of self-evolution
Researchers from MiniMax describe M2.7 as a model that helps improve its own agent harness, handling 30%-50% of RL workflows, autonomously iterating over 100 rounds to boost internal programming performance by 30%. It scored 56.22% on SWE-Pro, 55.6% on VIBE-Pro, 57.0% on Terminal Bench 2, 1495 ELO on GDPval-AA, 46.3% on Toolathon, 62.7% on MM Claw, and achieved 66.6% average medal rate across 22 ML competitions.
Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation – Pushes post-training into a structured pipeline where RL and distillation co-evolve across domains, showing how smaller models can reach very high reasoning density →read the paper