Gustavo Barra

[Gustavo Barra @bugarra]

Premiado cientista Uberabense https://t.co/4q7Vl6sl2n

[Gustavo Barra @bugarra]

EXTRAI, part 3: I built a meta-analysis pipeline out of local AIs auditing each other — and the quality gate did more damage than the sabotage it was https://t.co/HtEdjmgDkG via @LinkedIn

[Gustavo Barra @bugarra]

EXTRAI, part 2: no model got a single confidence interval right in its head — then I handed them a calculator, and one closed a perfect meta-analysis https://t.co/eRhJoZuqeW via @LinkedIn

[Gustavo Barra @bugarra]

EXTRAI, part 1: I made four local models redo a meta-analysis's data extraction — they found more errors in it than it found in them https://t.co/TDI4TmNp3d via @LinkedIn

[Gustavo Barra @bugarra]

Summarization benchmark, part 13: the most faithful model I've ever tested — and it lost to both smaller ones https://t.co/cNjDODTOl9 via @LinkedIn

[Gustavo Barra @bugarra]

Summarization benchmark, part 12: the big model didn't fit my GPU — and ran anyway https://t.co/BDuawtFlQ9 via @LinkedIn

[Gustavo Barra @bugarra]

Summarization benchmark, the wrap-up: what 432 graded measurements taught us — and where a local model can be trusted https://t.co/0u0XuvGmLx via @LinkedIn

[Gustavo Barra @bugarra]

Summarization benchmark, the wrap-up: what 432 graded measurements taught us — and where a local model can be trusted https://t.co/fgy70CNFzn via @LinkedIn

[Gustavo Barra @bugarra]

Summarization benchmark, part 11: I tried to replace the human reviewer with software — and what I got was a triage tool https://t.co/J9kqch1gD1 via @LinkedIn

[Gustavo Barra @bugarra]

Summarization benchmark, part 10: the outreach article — the long format dethroned the champion https://t.co/7J0ayPJx60 via @LinkedIn

[Gustavo Barra @bugarra]

Summarization benchmark, part 9: the study note — the format local models get most right (and the gap that became a hallucination) https://t.co/I8AVuFQ5Ra via @LinkedIn

[Gustavo Barra @bugarra]

Summarization benchmark, part 8: six real scientific papers became 144 LinkedIn posts — can you trust them? https://t.co/Lb29O8cDrF via @LinkedIn

[Gustavo Barra @bugarra]

Summarization benchmark, part 7: is the collapse the model's fault or Ollama's? I swapped the engine to find out https://t.co/H7zkHjTsch via @LinkedIn

[Gustavo Barra @bugarra]

Summarization benchmark, part 6: I tried to rescue the thinking with prompts https://t.co/eWEEiDwYXD via @LinkedIn

[Gustavo Barra @bugarra]

Four studies and 250 graded runs later: which local LLM should you actually use for summarization? https://t.co/cDkY9mIMMy via @LinkedIn

[Gustavo Barra @bugarra]

I gave my local LLMs a real word-counting tool — nine models reacted in five very different ways https://t.co/hPdNkQpFqK via @LinkedIn

[Gustavo Barra @bugarra]

I let my local LLMs think before answering — and the benchmark champion stopped responding https://t.co/wnMi0FOrTG via @LinkedIn

[Gustavo Barra @bugarra]

I benchmarked nine local LLMs on faithful summarization — on a mini PC with no graphics card https://t.co/Kfnu7uh6j1 via @LinkedIn

[Suresh Reddy @AmbSKReddy]

Visited Sabin Labs Central Processing center! Sabin truly represents the success of the #women #entrepreneurs of Brazil! Success built on innovation, dedication to customers & commitment to quality! Parabens to Team Sabin! #healthcare #healthtech https://t.co/iCeH6nJK15

[Gustavo Barra @bugarra]

My article is featured on the @ResearchGate #COVID19 community page. Read it here: https://t.co/v8d4efFuSb