Alisson Soares
@alissonmasoares.bsky.social
460 followers 900 following 200 posts
Computational social scientist; PhD in Sociology at UFMG, former postdoc researcher in political sciences at UFPE-Brazil #Disinformation; conspiracy theories; #rstats #python, text mining, #text-as-data My open projects https://github.com/SoaresAlisson
Posts Media Videos Starter Packs
Reposted by Alisson Soares
tevoelker.bsky.social
New paper out with @dasalgon.bsky.social: “Far-Right Agenda Setting: How the Far Right influences the Political Mainstream” doi.org/10.1017/S1475676525100066 #openaccess in @ejprjournal.bsky.social🧵
Abstract
alissonmasoares.bsky.social
"U.S. President Donald Trump was the dominant source and
spreader of #disinformation in our content analysis.... There is a correlation between the rise of Trumpism on one hand and the loss of trust in journalists as ‘watchdogs’/‘truth seekers’ on the other, "

www.icfj.org/sites/defaul...
alissonmasoares.bsky.social
At least in Brazil, the far right began by criticizing the humanities and quickly moved on to claims that “universities are woke” and “public universities are drug factories,” cutting funding for all higher education. The humanities are the first step, then they go to biology, health sciences...
alissonmasoares.bsky.social
The "validate, validate, validate" (GRIMMER, 2014) principle of Text Analysis/NLP never gets old.
Reposted by Alisson Soares
jbgruber.bsky.social
If you feel uneasy using LLMs for data annotation, you are right (if not, you should). It offers new chances for research that is difficult with traditional #NLP/#textasdata methods, but the risk of false conclusions is high!

Experiment + *evidence-based* mitigation strategies in this preprint 👇
joachimbaumann.bsky.social
🚨 New paper alert 🚨 Using LLMs as data annotators, you can produce any scientific result you want. We call this **LLM Hacking**.

Paper: arxiv.org/pdf/2509.08825
We present our new preprint titled "Large Language Model Hacking: Quantifying the Hidden Risks of Using LLMs for Text Annotation".
We quantify LLM hacking risk through systematic replication of 37 diverse computational social science annotation tasks.
For these tasks, we use a combined set of 2,361 realistic hypotheses that researchers might test using these annotations.
Then, we collect 13 million LLM annotations across plausible LLM configurations.
These annotations feed into 1.4 million regressions testing the hypotheses. 
For a hypothesis with no true effect (ground truth $p > 0.05$), different LLM configurations yield conflicting conclusions.
Checkmarks indicate correct statistical conclusions matching ground truth; crosses indicate LLM hacking -- incorrect conclusions due to annotation errors.
Across all experiments, LLM hacking occurs in 31-50\% of cases even with highly capable models.
Since minor configuration changes can flip scientific conclusions, from correct to incorrect, LLM hacking can be exploited to present anything as statistically significant.
alissonmasoares.bsky.social
Transformers Models Timeline
Interactive timeline to explore models supported by the Hugging Face Transformers library

#transformers #LLM #bert #IA
huggingface.co/spaces/yonig...
alissonmasoares.bsky.social
"We define workslop as AI generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task."
"The insidious effect... is that it shifts the burden of the work, requiring the receiver to interpret, correct, or redo the work."

hbr.org/2025/09/ai-g...
alissonmasoares.bsky.social
Discurso de Lula na Assembleia Geral da ONU. Palavras mais frequentes. #rstats
alissonmasoares.bsky.social
... há performance. No entanto, ele funciona com fenômeno social, muito novato terá seu contato com estes assuntos através deste formato e seria problemático deixar falando sozinhos. Um caso interessante no Jubilee foi do Jordan Peterson www.youtube.com/watch?v=lrdO...
How Jubilee Accidentally EXPOSED Jordan Peterson’s 1 Trick
YouTube video by Rationality Rules
www.youtube.com
alissonmasoares.bsky.social
o principal é convencer a plateia, plantando sementes de dúvidas aqui e ali. A conversão não é automática diante de um bom argumento, não atinge todos e leva tempo. Mas ela pode sim ter efeito. Para meu consumo próprio, acho estes debates muito pouco informativos e concordo que não há debate real...
alissonmasoares.bsky.social
Sim, eles "weaponizam", mas nessa de disputar narrativa, ao menos aqui no Brasil, houve diversos casos de sucesso, de desmoralização de debatedores como vazios e fúteis. O objetivo principal não é convencer o oponente, que geralmente monta um modelo de business baseado em moralismo, o principal é...
alissonmasoares.bsky.social
Deputados Josimar Maranhãozinho (PL-MA),
Bosco Costa (PL-SE) e
Pastor Gil (PL-MA) foram denunciados pela PGR

acusados dos crimes de corrupção passiva e organização criminosa.

noticias.uol.com.br/politica/ult...

"acusados de vender emendas e pedir R$ 1,67 milhão em propina" (G1)
noticias.uol.com.br
alissonmasoares.bsky.social
noticias.uol.com.br/politica/ult...

senador Eduardo Gomes (PL-TO)
esquema de desvio de emendas parlamentares.
noticias.uol.com.br
alissonmasoares.bsky.social
Violência política nos EUA, por ideologia. Alguns estudos:

Fonte: www.start.umd.edu/news/start-r...