In this thirteenth retrospective, I share personal updates from a dark Danish autumn, plus the tools, reads, listens, and watches shaping my week. I highlight agentic dev tools like Executor, Rex, and Grok Bot, recommend standout reads on confidence, systems thinking, and context engineering, and point to a great Will Larson interview and a sharp Stuart Halloway talk—along with a current streak of UK rap and the new sondaeblu album.
Relevant links:
Original article (https://duarteocarmo.com/blog/retrospectiva-13.html)
Executor (https://executor.sh/)
Rex by Superlogical (https://www.superlogical.com/rex/)
Grok Bot (settings and notifications) (https://docs.x.ai/grok-bot/settings-and-notifications)
Lee Robinson talk on Grok Bot (https://www.youtube.com/watch?v=L1OBRtfPS34)
On Confidence (The School of Life) (https://www.theschooloflife.com/shop/tsol-press-on-confidence/)
The School of Life (https://www.theschooloflife.com/)
Thinking in Systems: A Primer (https://www.chelseagreen.com/product/thinking-in-systems/)
The Ultimate MoE Training Handbook (Hugging Face) (https://huggingface.co/spaces/huggingface/moe-blogpost)
Resisting Cognitive Surrender (https://knlb.dev/essays/surrender.html)
The State of Context Engineering (Drew Breunig) (https://www.dbreunig.com/2026/10/24/the-state-of-context-engineering.html)
The revised rules of engineering management with Will Larson (The Pragmatic Engineer) (https://newsletter.pragmaticengineer.com/p/the-revised-rules-of-engineering)
Will Larson’s blog (https://lethain.com/)
How to Make Things, and Why (Stuart Halloway) (https://www.youtube.com/watch?v=7yeH1DqK984)
8 Oct 2026
#73 Retrospectiva #12
October’s retrospective covers family milestones, a return to running with a second‑hand stroller, and a brief Lisbon trip to speak about European Portuguese large language models. I share favorite tools (like Muesli and the MoErgo Go60), current reads and standout technical papers, plus podcasts on developer training in the age of AI and debates on recursive self‑improvement, and a couple of must‑watch videos.
Relevant links:
Original article (https://duarteocarmo.com/blog/retrospectiva-12.html)
Lisbon AI talk (https://www.youtube.com/watch?v=e7VYjMgMv2U)
Baby Jogger F.I.T. (https://web.archive.org/web/20151208002047/http://www.babyjogger.com/product/fit/)
Muesli (https://muesli.works/)
Superwhisper (https://superwhisper.com/)
Handy (https://handy.computer/)
MacWhisper (https://www.macwhisper.com/)
MoErgo Go60 (https://www.moergo.com/pages/go60)
Happy Hacking Keyboard review (https://duarteocarmo.com/blog/happy-hacking-keyboard-review.html)
Vision Language Models: Building VLMs with Hugging Face (https://www.oreilly.com/library/view/vision-language-models/9798341624047/)
The Machinery of Life (DOI) (https://doi.org/10.1007/978-0-387-84925-6)
Xiaomi MiMo‑V2.6 technical report (PDF) (https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL/blob/main/MiMo_V2_6_technical_report.pdf)
Beyond Per‑Token Pricing: Concurrency‑Aware LLM Infrastructure Cost Estimation (arXiv) (https://arxiv.org/abs/2606.11690)
Dear Congress: Here’s What You Should Ask A.I. Leaders (NYT) (https://www.nytimes.com/2026/09/24/opinion/congress-ai-sam-altman.html)
How Will We Train Developers if AI Does the Routine Work? (InfoQ Podcast) (https://www.infoq.com/podcasts/train-developers-ai-routine-work/)
AI researchers debate how close we are to recursive self‑improvement (Dwarkesh) (https://www.dwarkesh.com/p/john-beren-charlie)
Debating RSI, the US‑China Gap, and Jaggedness (Interconnects) (https://www.interconnects.ai/p/debating-rsi-the-us-china-gap-and)
Better Know a Benchmark: ExploitGym (Linear Digressions) (https://soundcloud.com/linear-digressions/better-know-a-benchmark-1)
How Lauren Tan shipped 2,500 PRs in a month (X) (https://x.com/poteto/status/2102050467505430555)
13 Sept 2026
#73 AMALIA vs. smaller reasoning models
I compare AMALIA against smaller, reasoning-focused models on Portuguese exam-style benchmarks and find that models like Qwen3.5 2B and LFM2.5 can outperform AMALIA—often by using far more tokens via test-time scaling. While AMALIA is strong and the team’s benchmarks are invaluable, practical use cases (like legal Q&A) may benefit from small reasoning models today; this also raises questions about a smaller, reasoning-capable future AMALIA.
Relevant links:
Original article (https://duarteocarmo.com/blog/amalia-vs-smaller-reasoning-models.html)
News: Government delayed grading high school exams (Público) (https://www.publico.pt/2026/07/03/sociedade/noticia/ministerio-educacao-adia-tres-dias-afixacao-notas-exames-2180378)
AMALIA on Hugging Face (https://huggingface.co/amalia-llm)
PT Exams benchmark dataset (https://huggingface.co/datasets/amalia-llm/pt_exams)
Qwen3.5 2B model (https://huggingface.co/Qwen/Qwen3.5-2B)
Interactive chart: Intelligence (https://duarteocarmo.com/images/113/intelligence.svg)
NVIDIA blog: AI scaling laws and test-time scaling (https://blogs.nvidia.com/blog/ai-scaling-laws/)
NVIDIA models on Hugging Face (https://huggingface.co/nvidia)
Qwen team on Hugging Face (https://huggingface.co/Qwen)
Interactive chart: Frontier (https://duarteocarmo.com/images/113/frontier.svg)
LFM2.5 model (https://huggingface.co/LiquidAI/LFM2.5-2.6B)
Interactive chart: Tokens (https://duarteocarmo.com/images/113/tokens.svg)
Qwen3.5 4B model (https://huggingface.co/Qwen/Qwen3.5-4B)
LegalBench PT dataset (https://huggingface.co/datasets/BeatrizCanaverde/LegalBench.PT)
Evaluation code repo (amelia) (https://github.com/duarteocarmo/amelia)
AMALIA-LM-eval repository (https://github.com/AMALIA-LLM/amalia-lm-eval)
Gonçalo on X (Twitter) (https://x.com/goncalossilva)
Contact Duarte O.Carmo's articles
Guest appearances
Books guests
Based on episode analysis; this does not confirm that the show is currently accepting guests.
Host of Duarte O.Carmo's articles?
Claim your podcast to manage its listing and keep your show details accurate.
Pod Engine is an independent podcast discovery and analytics service and is not affiliated with or endorsed by this podcast. Artwork and show content belong to their owners. Full legal notice.
Explore this show Podcast research with Pod Engine
