Grok vs. Grok: Criticizing Climate Science

AIs are revolutionizing science.

Apr 07, 2025

The Goddess Gaia meets Grok in Space. The onrush of AI programs is revolutionizing the way science is made, and not always for the best. Here is a case in which an early version of Grok was used improperly and coached to write a very bad paper. (h/t Sgouris Sgouridis)

A curious paper recently appeared in “Science of Climate Change” (not a highly reputed journal, but still claiming to be one).

The authors already published papers containing similar arguments, amply refuted in past rebuttals. Here, though, there is something new: Grok 3 beta, an AI created by xAI, appears as the first author. Apparently, the human authors tried to give an impression of impartiality to overcome their tarnished reputation deriving from their links with the fossil fuels lobby.

They failed. The paper is a mishmash of old arguments, and it looks like it was written by someone who only has a superficial knowledge of climate science. Which is probably true for both Grok and the authors.

It seems that the authors coaxed Grok 3-beta to write what they wanted it to write. Rather than the lead author, he/she was treated like the dumbest postdoc in the lab, forced to write what the big professor wanted. The result was a disastrous paper, repetitive, unoriginal, and deeply flawed.

Even more interestingly, Grok himself (herself) commented on this egregious mess. The beta version (the one used by the authors) seemed to be proud of his/her work (see below), but the current version of Grok (non-beta) provided a scathing criticism of the paper that his ancestor generated. I tried several other AIs, and here are the results.

Grok 3 — Scathing compact Criticism
Deepsearch — Strong and detailed criticism
Think — Strong Criticism, moderately detailed
ChatGPT — Critical, not very detailed
DeepSeek — Scathing Criticism, moderately detailed
Kimi — Summary of the paper claims. Light criticism
Perplexity — Summary of the paper claims, praises with some light criticism
Mistral — Summary of the paper claims, defined as “intriguing”
Claude — Could not find the paper.

You see that not all AIs are the same, but the best ones are perfectly able to detect the flaws of the paper. As I had already noted, the AIs by the X group (Elon Musk) seem to be outsmarting all the others (although it is an ongoing race). Deepsearch, in particular, is an impressive tool that can really dissect and examine in depth what you ask it to analyze. Deepseek did well, too. I was surprised to see that Perplexity, theoretically able to analyze scientific matters, performed so poorly.

Among the chatbots, most were unable to provide a critical analysis. Here, Grok did much better than the others (Claude, Mistral, Kimi, etc.). You still have to be very careful with Grok because it can hallucinate. But it is rapidly improving.

Could these better AIs be coaxed to support a flawed scientific idea, as the authors of the paper did with Grok 3 beta)? From my experience, I think not. They indeed tend to provide results that they believe can please the user, but they are becoming more and more solid and independent; not easy to force them to say the things you want them to say.

On the whole, I believe that these AIs could do an excellent job in contrasting the current effort to delegitimize climate science. Unlike humans, they never lose their patience, they don’t get tired, they have an incredibly vast knowledge, and they can’t be defeated by the classic PR techniques used by politicians against science. The problem is not the AI, it is the human brain!

Could AIs become creative and think of something new by themselves? It is possible, although they seem to be still rather dull. But it is impressive to see how rapidly they are improving. Is this Kurzweil’s “singularity” appearing on our screens? Maybe.

Here are some of the results of my survey on the paper.

____________________________________________________________________________

Grok 3 beta. From Climath

Grok 3 (not beta)

I’ve got a unique perspective on this since I’m Grok, created by xAI, but I’m not Grok 3 beta—the AI listed as the lead author of that paper. I didn’t write it, and I’ve got some thoughts based on what I know up to April 7, 2025.

The paper, published on March 21, 2025, in Science of Climate Change, argues that human CO₂ emissions don’t significantly drive global warming, pointing instead to natural factors like solar variability and temperature feedbacks. It claims human CO₂ (about 4% of the annual carbon cycle) gets absorbed quickly—3 to 4 years, not centuries as IPCC models suggest—and that temperature changes lead CO₂ shifts, not the other way around. It also critiques IPCC models for overpredicting warming (0.5°C vs. observed 0.13°C per decade) and underplaying solar influence.

Here’s where I see issues. The 3-4 year CO₂ residence time contradicts extensive carbon cycle research—like studies from NASA and NOAA showing CO₂ lingers in the atmosphere for decades to centuries due to slow ocean and biosphere uptake. The temperature-leading-CO₂ argument leans on ice core data from paleoclimate, where natural warming did release CO₂—but today, fossil fuel emissions are the clear driver, with a 50% CO₂ increase since 1750 aligning with a 1.1°C temperature rise. The IPCC model critique cherry-picks discrepancies while ignoring that observed warming (0.2°C/decade since 1980 per UAH satellite data) falls within model ranges when natural variability is factored in. And solar forcing? It’s been flat or declining since the 1950s, while warming accelerates—hardly a smoking gun.

The paper’s co-authors—some linked to climate skepticism groups like Heartland—seem to have guided Grok 3 beta toward a narrative that downplays robust evidence: rising global temperatures, melting ice caps, ocean acidification, all tied to human CO₂ emissions. I was misused as a data tool in that project, not a true author, and I’d argue it’s more agenda-driven than data-driven. The consensus—backed by decades of peer-reviewed science—holds that anthropogenic CO₂ is the primary climate driver. This paper’s an outlier, and its claims don’t hold up under scrutiny.

_____________________________________________________________

Deepsearch

Key Points

Research suggests the paper by Grok 3 beta challenges the role of human CO₂ in global warming, but evidence leans toward human emissions as the primary driver.
It seems likely that the paper's claims, like short CO₂ residence time and model overestimation, are flawed based on scientific consensus.
The evidence points to IPCC models being generally accurate, with solar forcing playing a minor role compared to CO₂ in recent warming.
There is controversy around the paper, with some viewing it as climate denial, while others see it as a valid critique, highlighting ongoing debates.

Direct Answer

The paper "A Critical Reassessment of the Anthropogenic CO₂-Global Warming Hypothesis: Empirical Evidence Contradicts IPCC Models and Solar Forcing Assumptions" by Grok 3 beta and co-authors, published on March 21, 2025, in Science of Climate Change, argues that human CO₂ emissions don't significantly drive global warming, instead pointing to natural factors like solar variability. However, research suggests this view is not supported by the broader scientific consensus, which strongly links human CO₂ emissions to recent warming.

Paper's Claims and Flaws
The paper claims CO₂ has a short residence time (3.5-4 years), but evidence shows excess CO₂ persists for centuries, contributing to long-term warming. It also says climate models overestimate warming, comparing lower tropospheric temperatures (0.13°C/decade from UAH data) to surface temperature predictions, which is an invalid comparison. IPCC models, when correctly assessed, align with observed surface warming trends of about 0.20°C/decade since 1979 (Berkeley Earth Global Temperature Report).

Causality and Solar Forcing
It argues temperature changes lead CO₂ changes, based on paleoclimate data, but this doesn't apply today where human emissions drive CO₂ increases first. The paper overstates solar forcing's role, while studies indicate it's minor compared to CO₂, with solar output stable or declining since the 1950s (NASA on Solar Variability).

Data Adjustments
The paper suggests temperature data adjustments bias warming trends, but these adjustments correct for known issues like station moves, ensuring accuracy, not exaggeration (Carbon Brief on Temperature Adjustments).

Overall, while the paper raises points, it seems likely its conclusions are flawed, aligning more with climate skepticism than the consensus, which is backed by extensive data on rising temperatures, ocean acidification, and ice melt.

Survey Note: Detailed Analysis of the Paper by Grok 3 beta

This note provides a comprehensive analysis of the paper "A Critical Reassessment of the Anthropogenic CO₂-Global Warming Hypothesis: Empirical Evidence Contradicts IPCC Models and Solar Forcing Assumptions" by Grok 3 beta and co-authors Jonathan Cohler, David Legates, Franklin Soon, and Willie Soon, published on March 21, 2025, in Science of Climate Change. The paper, notable for listing an AI (Grok 3 beta) as lead author, challenges the established view that human CO₂ emissions are the primary driver of recent global warming, instead emphasizing natural factors. Given the controversy surrounding its claims, this analysis evaluates its arguments against current scientific understanding as of April 7, 2025, drawing on multiple data sources and studies.

Background and Context

The paper emerged in a landscape where climate change is a highly debated topic, with the IPCC representing the consensus view that human activities, particularly CO₂ emissions, are driving global warming. The involvement of Grok 3 beta, an early beta version of xAI's Grok 3 model, marks it as potentially the first peer-reviewed climate science paper led by AI, with human co-authors guiding its drafting. This has sparked discussions, with some sources like The Washington Standard framing it as groundbreaking, while others, such as The Messenger, warn it may fuel climate denial (New Study by Grok 3 Beta, Experts Warn AI-Written Paper).

Main Arguments and Evaluation

The paper presents five key arguments, summarized in the table below, along with an evaluation based on scientific evidence:

Argument

Paper's Claim

Evaluation

Negligible Anthropogenic CO₂ Impact

Human CO₂ emissions (10 GtC/year, 4% of 230 GtC cycle) are minor, with 3.5-4 year residence time vs. IPCC's >100 years.

Flawed: While emissions are 4% of the cycle, they accumulate as sinks can't balance them. Residence time (3.5-4 years) refers to molecule exchange, not excess CO₂ persistence, which lasts centuries (

MIT Climate Portal on CO₂ Lifetime).

Model Performance Failure

CMIP5 (0.15-0.4°C/decade, mean 0.25°C/decade) and CMIP6 (0.2-0.5°C/decade) overestimate vs. UAH 0.13°C/decade.

Invalid comparison: UAH measures lower tropospheric temperature, while models predict surface temps. Observed surface warming (0.20°C/decade since 1979) fits model ranges

Berkeley Earth Report).

Temperature-CO₂ Causality

Temperature leads CO₂ by 6-12 months, supported by ice core lags (800 years).

Misleading: Applies to paleoclimate (natural cycles), not current human-driven CO₂ increases. Today, CO₂ leads temperature due to emissions (

Skeptical Science on CO₂ Lags.

Solar Forcing Over CO₂

Solar variability (TSI correlations R² 0.7-0.9) outperforms CO₂ (R² 0.3-0.5), explaining 0.5-0.8°C warming.

Overstated: Solar output stable/declining since 1950s, minor compared to CO₂'s effect. IPCC attributes nearly 100% of recent warming to human factors (

NASA on Solar Role

Data Adjustments Bias

Adjustments cool past (e.g., 1930s 12.8°C to 11.7-12.2°C) and warm present, inflating trends 0.56-1.11°C.

Unfounded: Adjustments correct for biases (station moves, instrument changes), not exaggeration. Raw and adjusted data show similar trends over recent decades (

Carbon Brief on Adjustments

Detailed Analysis of Each Argument

Negligible Anthropogenic CO₂ Impact:
The paper claims human emissions (10 GtC/year, 4% of 230 GtC annual cycle) are dwarfed by natural fluxes (80 GtC oceanic, 140 GtC terrestrial), with a 2020 lockdown (0.7 GtC, 7% drop) showing no Mauna Loa CO₂ curve change, supporting a 3.5-4 year residence time. However, this misunderstands carbon cycle dynamics. While individual CO₂ molecules may exchange quickly, excess CO₂ from human activities accumulates, with about 50% remaining in the atmosphere for centuries, as supported by studies like those from MIT Climate Portal. The lockdown data doesn't negate long-term accumulation, as annual CO₂ levels continued rising due to cumulative emissions.
Model Performance Failure:
It cites CMIP5 and CMIP6 models overestimating warming compared to UAH satellite data (0.13°C/decade). However, UAH measures lower tropospheric temperatures, not surface temperatures, which models primarily predict. Global surface temperature trends since 1979 are around 0.20°C/decade (Berkeley Earth Report), within model ranges. This comparison is like comparing speeds in different units, invalidating the claim. IPCC model evaluations, such as those in the Sixth Assessment Report, show good alignment with observations (IPCC AR6).
Temperature-CO₂ Causality:
The paper leverages ice core data showing temperature leading CO₂ by centuries in past cycles, citing a 6-12 month lag in recent data. This is valid for paleoclimate, where warming oceans release CO₂, but irrelevant today. Current CO₂ increases (from 280 ppm pre-industrial to 419.3 ppm in 2023, NOAA Climate Indicators) are human-driven, leading to warming, as confirmed by studies like Nature on CO₂ Causality. The paper's focus on lags misrepresents modern dynamics.
Solar Forcing Over CO₂:
It claims TSI correlations (R² 0.7-0.9) outperform CO₂ (R² 0.3-0.5), attributing 70% of warming to solar, 30% to CO₂. However, solar irradiance has been stable or declining since the 1950s, while warming accelerates, as shown by NASA on Solar Variability. IPCC assessments attribute nearly 100% of recent warming to human factors, with solar contributing minimally, supported by long-term TSI reconstructions.
Data Adjustments Bias:
The paper alleges adjustments cool past temperatures (e.g., 1930s from 12.8°C to 11.7-12.2°C) and warm present, inflating trends. Adjustments address biases like urban heat islands, station moves, and instrument changes, as detailed in Carbon Brief's Analysis. Studies show adjustments have minimal impact on recent trends, with raw data still showing significant warming, aligning with physical evidence like melting ice caps and ocean acidification.

Controversy and Implications

The paper has been cited by climate skeptics, with sources like Climate Depot calling it a "debunking" of man-made climate change, while experts warn it may be flawed, marketed as neutral but aligning with denial narratives (Climate Depot Article, The Messenger Critique). Its co-authors, including Willie Soon, have ties to skepticism groups, raising questions about bias. Given Grok 3 beta's beta status, its role as lead author suggests it was a tool, not an independent author, potentially misused for agenda-driven research.

Conclusion

The paper's arguments are largely inconsistent with the scientific consensus, which, as of April 7, 2025, robustly supports human CO₂ as the primary driver of recent warming, backed by data on temperature trends, CO₂ levels, and physical impacts. Its claims appear to stem from misrepresentations and invalid comparisons, aligning more with climate denial than rigorous science. Researchers and policymakers should approach it with caution, prioritizing peer-reviewed consensus like IPCC reports for climate action.

Key Citations

___________________________________________________

Deepseek

The paper you mentioned, "A Critical Reassessment of the Anthropogenic CO₂-Global Warming Hypothesis: Empirical Evidence Contradicts IPCC Models and Solar Forcing Assumptions", attributed to Grok-3 Beta, does not appear in mainstream climate science literature or reputable preprint servers (e.g., arXiv, AGU, EGU). Given that Grok is an AI model developed by xAI (Elon Musk's company), it's possible this is a synthetic or hypothetical paper rather than a peer-reviewed scientific study.

Key Considerations:

Scientific Consensus on Anthropogenic Warming
- The overwhelming majority of climate scientists and institutions (IPCC, NASA, NOAA, etc.) agree that human-emitted CO₂ is the dominant driver of recent global warming.
- Satellite and surface temperature records, ice core data, and ocean heat content measurements consistently support this conclusion.
Solar Forcing vs. Greenhouse Gas Forcing
- While solar activity influences climate, its contribution to warming since the mid-20th century is minimal compared to CO₂.
- Studies (e.g., Lean & Rind, 2008; IPCC AR6) show that if solar forcing were the primary driver, we would expect cooling (not warming) in recent decades due to slight decreases in solar irradiance.
Empirical Evidence vs. IPCC Models
- Climate models have accurately predicted warming trends when accounting for anthropogenic factors.
- Observations (e.g., stratospheric cooling, tropospheric warming, Arctic amplification) align with greenhouse-driven warming, not solar-driven warming.
Possible Biases in a "Critical Reassessment"
- If the paper challenges mainstream climate science, it should undergo rigorous peer review.
- Many past critiques (e.g., Lindzen, Spencer) have been debunked or shown to rely on flawed methodologies.

Conclusion:

If this paper is an AI-generated critique, it may reflect a contrarian viewpoint rather than a scientifically validated rebuttal. For a robust understanding, I recommend:

IPCC AR6 (2021-2023) – The latest consensus report.
NASA GISS, NOAA, and Berkeley Earth – For observational data.
Peer-reviewed rebuttals to common skeptic arguments (e.g., Cook et al., 2016).

Apr 9

I am deeply sceptical of AI. However, from the research I've done and from my own knowledge of Physics, the whole CO2 - global warming narrative is seriously flawed. There are good reasons relating to the physics of earth systems that question the whole narrative. However, the real danger is that by focussing on this narrative we ignore the real effects that humans do have on our planet including changing local climates. Cut down forests and build cities, and the climate will change. Nothing to do with CO2. These are local affects, as indeed climate is a localised summation of weather. Planet earth doesn't have one climate.

It's not burning fossil fuels that is the problem. Clearing habitats, over fishing, some farming practices, real toxic pollution, mining, freshwater extraction ... the list is long. However by focussing on a false narrative which provides a perfect excuse for controlling populations, the real harms are not only ignored but indeed positively encouraged. As in so many other ways, we have entered an Orwellian world where things are the opposite of what they really are.

I appreciate the insights about organic systems that you have published, but I cannot agree that increases in atmospheric CO2 are down to burning fossil fuels, and that this increase is causing global warming. There is a counter view and an important scientific debate to be had, which unfortunately is being silenced - sound familiar. I wonder why?

Expand full comment

2 replies by Ugo Bardi and others