Alejandro Saucedo | KubeCon 2025 Europe AI Keynote
@axsaucedo.bsky.social
Zalando Director of Eng, Science, Product & Analytics | Advisor at the UN, EU Commission, ACM, Institute for Ethical AI & others
Google DeepMind released Gemini 4 Argon this week to compete against OpenAI's Astra and Claude's Fable, with a 1M token output limit (finally)! blog.google/innovation-a...
Diogo Almeida, co-author of RLHF and ChatGPT and now behind TypeSafe's Jev model, gave a talk at AI Council on the philosophy behind it: www.youtube.com/watch?v=o-y1...
China published their AI Safety Governance Framework 3.0, and as an OWASP Agentic Security reviewer I'm impressed with how well it covers the real vulnerabilities behind agentic risks. www.cac.gov.cn/rootimages/u...
Uber Eats ranking models serve 8 million predictions per second. They have build an impressive flywheel for training data from their inference pipelines, and they shared their learnings: www.uber.com/gb/en/blog/t...
What if the agent harness could be trained into the model itself, instead of shipping alongside it at runtime? Peking University, Google and HKUST propose distilling what a specialised harness does into the model weights. arxiv.org/abs/2609.24974
We all know that getting structured outputs from LLMs is a pain... Ex-OpenAI / RLHF co-inventor Diogo Almeida says "hold my beer" and tackles this with Jev: typesafe.ai/blog/introdu...
We all know that getting structured outputs from LLMs is a pain... Ex-OpenAI / RLHF co-inventor Diogo Almeida says "hold my beer" and tackles this with Jev: typesafe.ai/blog/introdu...
It's time to build our own agentic HARNESS! NVIDIA published a paper that goes after the agent harness instead of the model: arxiv.org/abs/2609.20519
Forecasting foundation models are taking over, with a new top model from Europe this week 🇪🇺🚀🇫🇷 The Forecasting Company released an Open 256 million parameter time series foundation model, going straight for the top 3. www.theforecastingcompany.com/blog/t0-beta
Netflix's Multimodal Asset Personalization at Massive Scale; or how they use AI to pick the artwork that really gets us to click on the new titles: netflixtechblog.com/maps-netflix...
Did you ever wonder how OpenAI builds software today internally? If you are interested on the emerging trends of "Software Factories" this is a great breakdown: newsletter.pragmaticengineer.com/p/openai-sof...
Cognition have shipped their new SWE-2, and it seems new records are set every week... they are just behind Fable 5.1 but at 64% less cost. cognition.com/blog/swe-2
Google Research published ToolGrad, a new way to build the datasets optimize tool-use for foundation models, and it's intersting how it inverts the usual approach. research.google/blog/toolgra...
Pinterest shared how they are evolving their Billion-Scale Embedding Retrieval models, and there's quite a few great lessons. medium.com/pinterest-en...
OpenAI published a 2-part series on scaling to 70 million requests a second over 500 petabytes, across almost 40 regions and more than a billion people a week. openai.com/index/scalin...
Raschka is out with another mega write-up on transformer architectures, this time on GPT-6 Astra's looped transformer. magazine.sebastianraschka.com/p/gpt-6-astr...
Google Research have released TimesFM-3. Everything up to TimesFM-2.5 was univariate, while almost every real forecasting problem has covariates hanging off it. research.google/blog/timesfm...
Netflix on how they build, align and monitor an LLM judge at scale. If the judge quietly starts approving bad output, nothing downstream will flag it. netflixtechblog.medium.com/the-lifecycl...
GPT-6 Astra is out, and at this point the benchmark tables are getting quite ridiculous! openai.com/index/gpt-6-...
Can anyone actually challenge NVIDIA on inference? OpenAI and Broadcom said "hold my beer" with "Jalapeño", OpenAI's first Intelligence Processor - these names are getting out of hand. openai.com/index/openai...
Uber shares how they run their Software Factory: AI agents across the whole SDLC, with over 70% of pull requests now attributed to agents and 30K+ agent skill executions per day. www.uber.com/gb/en/blog/e...
The story of the week needs little introduction: NVIDIA has agreed to buy Hugging Face for $12.9 billion, as first reported by The Information and echoed by CNBC's own sources confirming the talks. www.cnbc.com/2026/08/27/n...
This week we published a new post on how to actually write agent skills! ethical.institute/blog/how-to-...
OpenAI (self) reports that an internal version of its forthcoming Astra model generated solutions to ten long-standing problems across geometry, coding theory, group theory, circuit complexity, quantum complexity, lattice + extremal combinatorics: openai.com/index/ten-ad...
Netflix has developed an LLM-native Recommender System, and they share some of the learnings they gathered along the journey: netflixtechblog.com/genrec-towar...
