Carlos Baquero

Carlos Baquero

@xmal.bsky.social

Professor at FEUP University of Porto and researcher at INESC TEC. Distributed Systems and Data. Co-creator of CRDTs. Founder eurotux.com. Still searching for unknown unknowns. (🦋 July 2023) Web: https://cbaquero.github.io/web/

2.2K followers1.6K following1.1K posts

Preparing for an Informatics Europe session where I plan to address both the past publishing patterns and the AI transformation of research. Not an easy topic, since we are all figuring out how to cope with the exponential. Will probably frame it as questions for audience.

Beautiful morning in Madrid. Visiting the statistics department at UC3M for a PhD committee. A large body of work in Network Scale-up survey method.

Very happy to see this result published today @springernature.com Digital Finance. Takeaways: In short time horizons, up to 6 months, BTC price evolves randomly; at medium term, checked up to 2 years, the underlying power law trend dominates; on-chain data does not help predictions.

There are only two options: a) AI only produces sloph and we must double down on detection of AI errors; b) AI produces significant science breakthroughs, and we should double down on detecting human errors.

After four revisions, the "The Naive–Power Law Blend as a Robust Baseline for Bitcoin Price Forecasting" paper was accepted for publication in Springer-Nature Digital Finance Journal. Reviewers and editorial team were fundamental to improving the depth of the evaluation that supports the results.

I learnt about Actors in the 90s, and years later had the privilege to meet Gul Agha and Joe Armstrong, on separate occasions. The third Intellectual Genealogies report tells the story of Actors.

Third day of DARE 2026. I will present tomorrow a line of work on reconciliation of divergent states and some war stories on AI based protocol discovery.

Intelectual Geneologies is a series of reports that (unlike surveys) track forward in time the spread of computer science concepts. The calibration report was with CRDTs, and the first report is on Lamport clocks’ paper. Open to community feedback on the idea.

Suggestions are accepted for more seed papers, seminal papers with 10+ years, and I plan to issue a couple of reports per month since the process is almost fully automated.

I start a sabbatical research year this September, will be doing contacts for short visits (still open to invites). As automation of reasoning expands, the contributions expected from humans will also be more ambitious. August, calm before the storm.

These vacations I don’t need a social media detox, I have mostly been parsimonious with that vice since a competing one has creeped in. What I will pursue is a AI assisted projects detox. Must get back to have ideas that just stay ideas and are not immediately executed.

Carlos Baquero reposted

Pfff. You young whippersnappers and your fancy machines. I run Kimi K3 locally, the old fashioned way: with pencil and paper, one matrix multiplication at a time.

When multiple agents reason in parallel it is easy to merge their textual outputs, but very hard to merge their, richer, latent abstractions. We provide a first solution to this problem that also delivers CRDT properties.

Can we merge latent spaces as we merge CRDT replicas? It works for two replicas with a CRDT-like design, and we can do federated querying without exposing the query to the LLMs that process the source data. Abstract and link in thread.

Carlos Baquero reposted

⏳ Long block — #950,100 found after 36 min. Pressure built up during the wait. ~3,750 txs cleared. Mempool: 55,546 txs · 12.9 vMB · median 0.3 sat/vB. Miner reward: 3.1775 BTC. btcweather.report

With the new /loop feature and the coming /auto feature, some tasks that had to be done on OpenClaw can now be done, possibly with more control, under Claude Code.