Using 30 years of Congressional floor speeches (1994–2023), we built an AI-powered system to track how — and how often — members of Congress push back on climate science and policy. A few highlights: 🔬 Updated CARDS taxonomy: We revised and extended the CARDS framework (Coan et al., 2021) to better capture solutions-focused skepticism and explicitly link contrarian claims to the “discourses of delay” identified by Lamb et al. (2020) — bridging two major frameworks in the climate communication literature. ⚙️ New method — RECoT fine-tuning: We introduce Reverse Engineered Chain-of-Thought (RECoT) fine-tuning, a procedure that teaches a smaller, scalable model to “reason” through complex, multi-label classification tasks. The result: near frontier-model accuracy at roughly 1/10th the cost — making large-scale claims classification feasible. 📊 Key findings: • Republicans are ~12x more likely than Democrats to make contrarian claims in floor speeches. • “Solutions increase costs” accounts for roughly one-third of all contrarian claims. • Contrarian speech has spiked around major climate policy moments (1997, 2009, 2015), surged during the Biden era, and was driven by a dramatic increase in discourses of delay. • More conservative Republicans, younger members, and those representing fossil fuel-dependent districts are significantly more likely to make contrarian claims.

Using 30 years of Congressional floor speeches (1994–2023), we built an AI-powered system to track how — and how often — members of Congress push back on climate science and policy. A few highlights: 🔬 Updated CARDS taxonomy: We revised and extended the CARDS framework (Coan et al., 2021) to better capture solutions-focused skepticism and explicitly link contrarian claims to the “discourses of delay” identified by Lamb et al. (2020) — bridging two major frameworks in the climate communication literature. ⚙️ New method — RECoT fine-tuning: We introduce Reverse Engineered Chain-of-Thought (RECoT) fine-tuning, a procedure that teaches a smaller, scalable model to “reason” through complex, multi-label classification tasks. The result: near frontier-model accuracy at roughly 1/10th the cost — making large-scale claims classification feasible. 📊 Key findings: • Republicans are ~12x more likely than Democrats to make contrarian claims in floor speeches. • “Solutions increase costs” accounts for roughly one-third of all contrarian claims. • Contrarian speech has spiked around major climate policy moments (1997, 2009, 2015), surged during the Biden era, and was driven by a dramatic increase in discourses of delay. • More conservative Republicans, younger members, and those representing fossil fuel-dependent districts are significantly more likely to make contrarian claims.

Using 30 years of Congressional floor speeches (1994–2023), we built an AI-powered system to track how — and how often — members of Congress push back on climate science and policy. A few highlights: 🔬 Updated CARDS taxonomy: We revised and extended the CARDS framework (Coan et al., 2021) to better capture solutions-focused skepticism and explicitly link contrarian claims to the “discourses of delay” identified by Lamb et al. (2020) — bridging two major frameworks in the climate communication literature. ⚙️ New method — RECoT fine-tuning: We introduce Reverse Engineered Chain-of-Thought (RECoT) fine-tuning, a procedure that teaches a smaller, scalable model to “reason” through complex, multi-label classification tasks. The result: near frontier-model accuracy at roughly 1/10th the cost — making large-scale claims classification feasible. 📊 Key findings: • Republicans are ~12x more likely than Democrats to make contrarian claims in floor speeches. • “Solutions increase costs” accounts for roughly one-third of all contrarian claims. • Contrarian speech has spiked around major climate policy moments (1997, 2009, 2015), surged during the Biden era, and was driven by a dramatic increase in discourses of delay. • More conservative Republicans, younger members, and those representing fossil fuel-dependent districts are significantly more likely to make contrarian claims.

Using 30 years of Congressional floor speeches (1994–2023), we built an AI-powered system to track how — and how often — members of Congress push back on climate science and policy. A few highlights: 🔬 Updated CARDS taxonomy: We revised and extended the CARDS framework (Coan et al., 2021) to better capture solutions-focused skepticism and explicitly link contrarian claims to the “discourses of delay” identified by Lamb et al. (2020) — bridging two major frameworks in the climate communication literature. ⚙️ New method — RECoT fine-tuning: We introduce Reverse Engineered Chain-of-Thought (RECoT) fine-tuning, a procedure that teaches a smaller, scalable model to “reason” through complex, multi-label classification tasks. The result: near frontier-model accuracy at roughly 1/10th the cost — making large-scale claims classification feasible. 📊 Key findings: • Republicans are ~12x more likely than Democrats to make contrarian claims in floor speeches. • “Solutions increase costs” accounts for roughly one-third of all contrarian claims. • Contrarian speech has spiked around major climate policy moments (1997, 2009, 2015), surged during the Biden era, and was driven by a dramatic increase in discourses of delay. • More conservative Republicans, younger members, and those representing fossil fuel-dependent districts are significantly more likely to make contrarian claims.

Using 30 years of Congressional floor speeches (1994–2023), we built an AI-powered system to track how — and how often — members of Congress push back on climate science and policy. A few highlights: 🔬 Updated CARDS taxonomy: We revised and extended the CARDS framework (Coan et al., 2021) to better capture solutions-focused skepticism and explicitly link contrarian claims to the “discourses of delay” identified by Lamb et al. (2020) — bridging two major frameworks in the climate communication literature. ⚙️ New method — RECoT fine-tuning: We introduce Reverse Engineered Chain-of-Thought (RECoT) fine-tuning, a procedure that teaches a smaller, scalable model to “reason” through complex, multi-label classification tasks. The result: near frontier-model accuracy at roughly 1/10th the cost — making large-scale claims classification feasible. 📊 Key findings: • Republicans are ~12x more likely than Democrats to make contrarian claims in floor speeches. • “Solutions increase costs” accounts for roughly one-third of all contrarian claims. • Contrarian speech has spiked around major climate policy moments (1997, 2009, 2015), surged during the Biden era, and was driven by a dramatic increase in discourses of delay. • More conservative Republicans, younger members, and those representing fossil fuel-dependent districts are significantly more likely to make contrarian claims.

Using 30 years of Congressional floor speeches (1994–2023), we built an AI-powered system to track how — and how often — members of Congress push back on climate science and policy. A few highlights: 🔬 Updated CARDS taxonomy: We revised and extended the CARDS framework (Coan et al., 2021) to better capture solutions-focused skepticism and explicitly link contrarian claims to the “discourses of delay” identified by Lamb et al. (2020) — bridging two major frameworks in the climate communication literature. ⚙️ New method — RECoT fine-tuning: We introduce Reverse Engineered Chain-of-Thought (RECoT) fine-tuning, a procedure that teaches a smaller, scalable model to “reason” through complex, multi-label classification tasks. The result: near frontier-model accuracy at roughly 1/10th the cost — making large-scale claims classification feasible. 📊 Key findings: • Republicans are ~12x more likely than Democrats to make contrarian claims in floor speeches. • “Solutions increase costs” accounts for roughly one-third of all contrarian claims. • Contrarian speech has spiked around major climate policy moments (1997, 2009, 2015), surged during the Biden era, and was driven by a dramatic increase in discourses of delay. • More conservative Republicans, younger members, and those representing fossil fuel-dependent districts are significantly more likely to make contrarian claims.

Using 30 years of Congressional floor speeches (1994–2023), we built an AI-powered system to track how — and how often — members of Congress push back on climate science and policy. A few highlights: 🔬 Updated CARDS taxonomy: We revised and extended the CARDS framework (Coan et al., 2021) to better capture solutions-focused skepticism and explicitly link contrarian claims to the “discourses of delay” identified by Lamb et al. (2020) — bridging two major frameworks in the climate communication literature. ⚙️ New method — RECoT fine-tuning: We introduce Reverse Engineered Chain-of-Thought (RECoT) fine-tuning, a procedure that teaches a smaller, scalable model to “reason” through complex, multi-label classification tasks. The result: near frontier-model accuracy at roughly 1/10th the cost — making large-scale claims classification feasible. 📊 Key findings: • Republicans are ~12x more likely than Democrats to make contrarian claims in floor speeches. • “Solutions increase costs” accounts for roughly one-third of all contrarian claims. • Contrarian speech has spiked around major climate policy moments (1997, 2009, 2015), surged during the Biden era, and was driven by a dramatic increase in discourses of delay. • More conservative Republicans, younger members, and those representing fossil fuel-dependent districts are significantly more likely to make contrarian claims.

Using 30 years of Congressional floor speeches (1994–2023), we built an AI-powered system to track how — and how often — members of Congress push back on climate science and policy. A few highlights: 🔬 Updated CARDS taxonomy: We revised and extended the CARDS framework (Coan et al., 2021) to better capture solutions-focused skepticism and explicitly link contrarian claims to the “discourses of delay” identified by Lamb et al. (2020) — bridging two major frameworks in the climate communication literature. ⚙️ New method — RECoT fine-tuning: We introduce Reverse Engineered Chain-of-Thought (RECoT) fine-tuning, a procedure that teaches a smaller, scalable model to “reason” through complex, multi-label classification tasks. The result: near frontier-model accuracy at roughly 1/10th the cost — making large-scale claims classification feasible. 📊 Key findings: • Republicans are ~12x more likely than Democrats to make contrarian claims in floor speeches. • “Solutions increase costs” accounts for roughly one-third of all contrarian claims. • Contrarian speech has spiked around major climate policy moments (1997, 2009, 2015), surged during the Biden era, and was driven by a dramatic increase in discourses of delay. • More conservative Republicans, younger members, and those representing fossil fuel-dependent districts are significantly more likely to make contrarian claims.