Roberto Barroso Luque

Buzzword, jargon, buzzword, jargon.

[Lin Qiao @lqiao]

https://t.co/rMVRgBvwEI

[Winston Weinberg @winstonweinberg]

With @FireworksAI_HQ we’re researching agent routing to more intelligently navigate quality-cost trade-offs for long-horizon work

[Roberto Barroso Luque @BobBarroso]

Very cool work we did with Harvey. Training OSS models with your data + engineering on your harness >>> generic frontier models

[Fireworks AI @FireworksAI_HQ]

Frontier models are powerful advisors.

On @harvey's Legal Agent Benchmark, a GLM 5.1 worker using Claude Opus 4.7 as a sparse advisor reached 18/100 all-pass versus 14/100 for Opus alone, at 39% of the cost.

More on the harness design, advisor pattern, and training results: https://t.co/ozxFycdzcT

[Fireworks AI @FireworksAI_HQ]

At 30 trillion tokens a day, we're the largest inference provider outside the frontier labs' own APIs.

We're hiring.

Come build the infrastructure behind the top AI workloads: https://t.co/bUweMxMq9L

📸 @Wing VC's 2026 Enterprise Tech 30 list (thanks @WingVC @ericnewcomer)

[Dmytro Dzhulgakov @dzhulgakov]

Kimi has been the most popular model on Fireworks, both out of the box and as a fine-tuning base (including Composer 2)

Now Kimi K2.6 is live with huge jumps (10+%) in coding, long-running agents and deep research

Try it out for inference or fine-tuning!

[Fireworks AI @FireworksAI_HQ]

Fireworks Training is now in preview.

You can now full-parameter fine-tune Kimi K2.5 (1T params, 256k context) with custom loss functions (GRPO, DRO, DAPO, or bring your own) on managed infra.

@genspark_ai built their proprietary model stack in four weeks. @vercel hit 93% error-free generation with RFT. @cursor_ai runs their RL rollout fleet on Fireworks.

Full-parameter from 8B to 1T. Multi-LoRA serving. Managed or bring your own training loop.
Your model is your product. Your data is your moat.
https://t.co/kyz7HzihC1

[Fireworks AI @FireworksAI_HQ]

Training trillion-parameter MoEs is an infra problem disguised as a modeling problem.

So we built the infra solution. Cursor used it to train Composer 2. Now it's available for Kimi K2.5, Qwen3.5 397B, MiniMax M2.5, and more:

→Fused RL loss (~2x faster PPO)
→MXFP8 expert kernels on Blackwell
→Composable 4D parallelism
→1M+ token context training validated

Here's how it all works ↓
https://t.co/PA20I8EFaD

[Fireworks AI @FireworksAI_HQ]

Most teams don’t hit a fine-tuning bottleneck because they picked the wrong algorithm.

They hit the bottleneck because the loop around training is too slow:
-data plumbing
-eval setup
-deployment friction
-too much guesswork between runs

[Roberto Barroso Luque @BobBarroso]

@gonzalo__nunez retweet to amplify some quality science please

[Roberto Barroso Luque @BobBarroso]

People of the virtual sphere, check out the recently published "Shared and specialized coding across posterior cortical areas for dynamic navigation decisions", quality work from the Harvey Lab's @ShihYiTseng1 @selmaanchettih @CharlotteArlt!

https://t.co/Aqt0xxew5B

[bioRxiv Neuroscience @biorxiv_neursci]

Cognitive experience alters cortical involvement in navigation decisions https://t.co/DA49zmcAps #biorxiv_neursci

[Roberto Barroso Luque @BobBarroso]

Could not be happier for @CharlotteArlt and the Harvey Lab @harvardmed. Some quality science I had the fortune of working on out on pre-print now! Check it out!

[Charlotte Arlt @CharlotteArlt]

Excited to share some of my postdoc work as a @biorxivpreprint from the Harvey lab @harvardmed. We show that mice rely on different cortical areas for navigation decisions, depending on what other complex tasks they have previously learned. Training history matters!

[Chi Hack Night @chihacknight]

Dig into digital influence operations with us *tonight* @ 7pm CST. @BobBarroso & @nick__tallant of Omelas will discuss the map they've created of the online information environment to detect digital influence campaigns used to undermine democratic values.

https://t.co/WshtIrm84F

[Roberto Barroso Luque @BobBarroso]

Interested in information operations and geopolitics? Would recommend subscribing to the Omelas Newsletter https://t.co/yRhb6rEbWP

[Roberto Barroso Luque @BobBarroso]

Also big shout out to @_mcnutt_ who instructed us through this journey.

[Roberto Barroso Luque @BobBarroso]

People of the virtual world, check out some quality work from the students of @UChicagoCAPP using data viz to understand a multitude of relevant policy issues.

https://t.co/IZUCiwLBKx

[Roberto Barroso Luque @BobBarroso]

A tad late but thought I'd share. Worked with the brilliant @HanaPassen
investigating methods to demystify results from dimensionality reduction. Check it out!
https://t.co/3AJdSfrefn

[UChicago CAPP @UChicagoCAPP]

@jesicaramirezto, Oscar and @StephieRamos explored crime statistics in Mexico City to better understand trends, patterns, and assess the safety of citizens. See: https://t.co/Tl7HEoVF35

[UChicago CAPP @UChicagoCAPP]

Maps tell stories, and these maps tell familiar stories. See how public health indicators in Chicago correlate to demographic indicators along the CTA. See what @maerunes, @HanaPassen and @BobBarroso created: https://t.co/1yBqsiMmbP