Pi School
Condividi
We help organisations turn AI into real business impact through applied research and practical training.
07/07/2026
𝗪𝗲𝗹𝗰𝗼𝗺𝗲 𝗘𝗺𝗮𝗻𝘂𝗲𝗹𝗲 𝗮𝗻𝗱 𝗟𝗼𝗿𝗲𝗻𝘇𝗼 𝘁𝗼 𝗣𝗶 𝗦𝗰𝗵𝗼𝗼𝗹
We're pleased to welcome two new team members to Sébastien Bratières's team:
Emanuele Micieli is joining us as AI Platform Engineer. He brings full-stack development skills to the team and will start working on the Meetween and EVE projects. His background as a startup founder gives him experience in customer exploration and business modelling.
Lorenzo Loconte is joining us as a Senior Deep Learning Scientist. He has recently defended a PhD in Probabilistic Machine Learning at the University of Edinburgh. He will lead Pi School's contribution to DVPS, working on multimodal foundation models.
Welcome to the team, Emanuele and Lorenzo!
22/06/2026
🛰️ 𝐏𝐢 𝐒𝐜𝐡𝐨𝐨𝐥 𝐚𝐭 𝐭𝐡𝐞 𝐄𝐒𝐀 Φ𝐧𝐧𝐨𝐯𝐚𝐭𝐢𝐨𝐧 𝐒𝐮𝐦𝐦𝐢𝐭.
Visit us at the Magellan room on both days!
This week, we will attend the ESA Φnnovation Summit at ESA-ESRIN in Frascati, where we will run three side events showcasing our research and applied AI work.
🔹 On Tuesday, 23 June at 17:00, we will be presenting 𝑇ℎ𝑒 𝐶𝑟𝑜𝑠𝑠-𝑑𝑜𝑚𝑎𝑖𝑛 𝐺𝑎𝑝, which explores multimodal foundation models and how they bridge data across domains with Enzo Fabiani and other team members of the DVPS consortium.
🔹 Wednesday 24 June at 14:30, 𝐵𝑢𝑖𝑙𝑑𝑖𝑛𝑔 𝐴𝑔𝑒𝑛𝑡𝑖𝑐 𝐸𝑎𝑟𝑡ℎ 𝐼𝑛𝑡𝑒𝑙𝑙𝑖𝑔𝑒𝑛𝑐𝑒, with Àlex R. Atrio. A hands-on tour of EVE, the open-source assistant for Earth observation built by Pi School in collaboration with Imperative Space and Mistral AI for ESA Φ-lab.
🔹 Wednesday 24 June, 17:00 ESA Φ-lab Collaborative Innovation Network presented by Cristiano De Nobili. Discover how the ESA Φ-lab CIN connects researchers across the hashtag ecosystem.
🚀 𝐏𝐢 𝐀𝐈 𝐖𝐞𝐞𝐤𝐥𝐲 𝐓𝐫𝐞𝐧𝐝𝐬 #𝟖𝟗 𝐢𝐬 𝐡𝐞𝐫𝐞!
It’s Friday! Get ready to stay ahead with the latest AI breakthroughs, handpicked by our Deep Learning Scientist, Riccardo Corrente.
This week’s highlights:
⚡ 𝐌𝐞𝐫𝐜𝐮𝐫𝐲: 𝐔𝐥𝐭𝐫𝐚-𝐅𝐚𝐬𝐭 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐌𝐨𝐝𝐞𝐥𝐬 𝐁𝐚𝐬𝐞𝐝 𝐨𝐧 𝐃𝐢𝐟𝐟𝐮𝐬𝐢𝐨𝐧
Inception Labs introduces Mercury, a new generation of LLMs built entirely on diffusion. Parameterised via the Transformer architecture, these models are trained to predict multiple tokens in parallel rather than one by one. The specialised Mercury Coder models shatter the speed-quality frontier, achieving massive throughputs while maintaining top-tier quality on coding benchmarks and Copilot Arena.
🌐 https://pischool.link/f204fd
🖼️ 𝐈𝐦𝐚𝐠𝐞 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐨𝐫𝐬 𝐚𝐫𝐞 𝐆𝐞𝐧𝐞𝐫𝐚𝐥𝐢𝐬𝐭 𝐕𝐢𝐬𝐢𝐨𝐧 𝐋𝐞𝐚𝐫𝐧𝐞𝐫𝐬
Can image generators actually understand what they create? Google DeepMind proves they can. By instruction-tuning the Nano Banana Pro (NBP) model into "Vision Banana," researchers seamlessly reframed perception tasks (like segmentation and depth estimation) as RGB image generation. It achieves state-of-the-art 2D/3D visual understanding, rivalling specialists like SAM 3, without losing its creative edge.
🌐 https://pischool.link/7e4c1f
💻 𝐃𝐫𝐚𝐟𝐭-𝐓𝐡𝐢𝐧𝐤𝐢𝐧𝐠: 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐄𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐭 𝐑𝐞𝐚𝐬𝐨𝐧𝐢𝐧𝐠 𝐢𝐧 𝐋𝐨𝐧𝐠 𝐂𝐡𝐚𝐢𝐧-𝐨𝐟-𝐓𝐡𝐨𝐮𝐠𝐡𝐭 𝐋𝐋𝐌𝐬
Scaling inference compute enables LLMs to use long chains of thought (CoTs) to backtrack and correct errors. This study explores how these capabilities emerge during training. It reveals that while supervised fine-tuning (SFT) isn't strictly necessary alongside reinforcement learning, it drastically boosts efficiency.
🌐 https://pischool.link/6cc2d0
Was this helpful? Let us know by liking and sharing!
Clicca qui per richiedere la tua inserzione sponsorizzata.
Digitare
Sito Web
Indirizzo
Via Indonesia 23
Rome
00144
