This article introduces the idea that instead of (or as well as) peer-review, scientific papers could be reviewed by Grok, Elon Musk’s AI product. Well, maybe that’s not a new idea, but I haven’t seen it anywhere.

I apologise for the length of this article, but at least is shorter than some of Grok’s responses(!).

My suggestions for Grok-review, drawn from the material below, is that:

• A paper’s author should get a Grok-review prior to submission for publication.
• Grok is as likely to give helpful ideas and corrections as a human reviewer, and it is a lot quicker.
• Grok can and should be used interactively, to drill down into the paper’s arguments and thoroughly test them. In particular, Grok should be asked explicitly if the paper’s arguments are valid.
• Grok does make mistakes, even glaring errors, so care should be taken with it.
• A human reviewer could also benefit by using Grok to augment their review.
• A peer-reviewed journal could consider always using Grok as one reviewer.

A bit of background: I had a conversation recently with a relative of mine who is a great fan of Grok. They are particularly impressed with how competent and unbiased Grok is compared to other AIs (ChatGPT gave USS Enterprise as an example of US military ships with female names). For example, they asked Grok about the probability that something fishy is going on in the Letby trial (Lucy Letby is a British nurse convicted of murdering seven infants and the attempted murders of seven others), and to use bayes math to estimate the probability the Bibas family were killed in an israeli airstrike. The lengthy and finely argued (and IMHO very reasonable) responses were that the probability of Lucy Letby being guilty was 2.46%, IOW that she was clearly wrongly convicted, and the probability of the Bibas family having been killed in an Israeli airstrike was ~10.5% (“Why? Hamas’s failure to leverage immediate aftermath evidence—when they had every incentive to do so—suggests the airstrike claim may be less likely than an alternative (e.g., execution). If they had undeniable proof in November 2023, their silence then and late return now tilt the odds against.” was just a small part of the analysis.). Apart from the decimal places(!), the answers’ reasonableness was easily checkable from the detailed answers.

So I thought I would download Grok (Grok 3) and give it a test run. Each Grok answer took no more than 5 seconds, even ones that required Grok to find, download, read and analyse a paper. I started to think that instead of (or as well as) peer-review, scientific papers could benefit from Grok-review. Apart from Grok’s style being a lot chummier than a typical peer-reviewer (“If you’ve got more papers or ideas to toss my way later, just holler. Cheers!“), my perception is that Grok is an amazing facility and would be very useful for journal review, but it can’t be accepted without question and would have to be used with great care. ie, trust but verify. I would be interested in others’ opinions, especially as I’m sure many would already have used Grok a lot.

Grok appears to be excellent at accessing and analysing information very quickly, but even a super-AI can’t be expected to do a deep analysis in just a few seconds, and there are also obvious errors (just like a peer-reviewer except that Grok doesn’t defend them??). A Grok review would probably be no less valuable on average than a peer-review, but whereas a peer-reviewer could be unduly negative, or a pal-reviewer, or an active gate-keeper, especially if a paper was outside the accepted narrative, Grok would at least be unbiased. But would it be competent? I think so, but it would take a while to learn how to use it effectively.

I began my test run by asking: The IPCC only treats changes in cloud cover as being a reaction to the warming caused by CO2. How possible is it that many if not most changes in cloud cover actually occur independently of CO2?

I got a long answer, which boiled down to:

• The IPCC does tend to emphasize clouds as a feedback mechanism—meaning they respond to temperature changes, like those driven by CO2 increases, and either amplify or dampen the warming.
• But clouds are complex, and a good chunk of cloud cover changes could indeed happen independently, driven by other natural processes.
• Clouds are influenced by a ton of factors: atmospheric dynamics (like jet streams or pressure systems), solar radiation, cosmic rays, ocean currents, and even volcanic activity, all completely unrelated to CO2.
• Historical climate data, like from the Little Ice Age or Medieval Warm Period, shows patterns not correlating with CO2, pointing to other drivers like solar variability or internal climate oscillations.
• The IPCC’s focus isn’t entirely baseless. Warming from CO2 can change atmospheric humidity and stability, which does affect cloud formation. [Some detail re models and low-level and high-level clouds].
• Models struggle to nail down clouds precisely because they’re so chaotic and operate on scales smaller than the models can fully capture.
• The uncertainty in cloud feedback is one of the biggest wildcards in climate projections—some estimates suggest 0.5 to 2°C of uncertainty in total warming.
• It’s plausible that most  cloud cover changes could be independent of CO2, especially over short timescales (years to decades), where natural variability often dominates. [Discussion of cosmic rays seeding clouds by ionisation, a process tied to solar activity rather than CO2].
• But over centuries, as CO2 accumulates and warming persists, its influence on cloud patterns probably grows, making independence less likely.

I thought that was a pretty good answer. It did have some “IPCC” thinking, but it addressed a lot of the issues with clouds that certain scientists tend to avoid. So I asked Grok to look at my “Clouds” paper, published in wjarr.com [and written up on WUWT] which concluded that cloud cover changes during the study period were not caused by warming and hence were not caused by CO2. I asked whether Grok agreed that the conclusion is reasonable.

Grok began with a pretty good summary: “Your paper analyzes cloud cover data from ISCCP (1983–2009) and CERES (2000–2020), alongside CO2 radiative forcing estimates, to argue that cloud changes during these periods weren’t driven by warming (and thus not by CO2) and had a radiative impact at least as big as CO2’s. You use cloud cover trends—like a ~2% decrease in total cloudiness—and calculate their shortwave (reflective) and longwave (heat-trapping) effects, stacking them against CO2’s forcing of roughly 0.4 W/m² over the study spans.“. Chummy language, but a neat summary.

But there were also errors. Grok said I tied cloud activity to natural variability (solar activity, ocean cycles), but actually I didn’t. It made a glaring error saying I used CERES data when I didn’t (Grok did). Grok did however check for reasonableness (“CERES data backs this up; it’s shown cloud forcing swings that dwarf CO2’s steady climb.“) and reach (“Your focus on radiation alone keeps it apples-to-apples, avoiding messier temperature feedbacks. The study period caveat—1983–2020-ish—is key. You’re not saying clouds always dominate CO2, just that they did here. That’s cautious and reasonable. The data’s real, the forcing estimates are plausible, and the disconnect from warming holds water. “) and did eventually answer the question: “your conclusion—“clouds independently appear to have as much or greater effect than man-made CO2 on radiative forcing” in that window—stands up to scrutiny based on what you’ve presented.
I’d say it’s reasonable, yeah. The evidence fits, and you didn’t overreach.“.

I was very impressed with the level of understanding that Grok displayed, though the analysis was clearly fairly superficial (but not bad for about 3 seconds). I should have asked Grok to delve deeper and try to find errors, but I had criticisms of another paper of mine, General circulation models cannot predict climate [and written up on WUWT] so instead I presented the criticisms verbatim and asked “Are those criticisms valid?“.

Again, Grok gave a long and detailed reply in just a few seconds, and again there were mistakes but the analysis showed impressive understanding.

First came the summary of the paper: “Your core claim is that General Circulation Models (GCMs), due to their reliance on grid-based calculations iterated over small time steps, hit a prediction horizon wall (around two weeks, per Lorenz), beyond which their outputs lose reliability. You’re arguing that this internal mechanism undermines their ability to predict long-term climate, and that averaging or statistical tricks don’t fix the issue. Let’s see how the criticisms stack up.“.

Then, after a long discussion of chaotic attractors, direction fields, randomness and feedback, determinism and ensembles, Grok says:

“Your argument is laser-focused: the internal workings of General Circulation Models (GCMs), specifically the grid-based, small-time-step calculations rooted in deterministic equations (e.g., Navier-Stokes), hit a prediction horizon (around two weeks, per Lorenz’s chaos limit) that renders them unreliable for long-term climate prediction. You go further, saying these low-level mechanics don’t even drive the final results—parameterizations, higher-level assumptions, and constraints (like CO2 forcings) do the heavy lifting. In fact, you suggest a successful model could skip the grid-stepping entirely and lean on that higher-level logic. So, do the critics’ points knock this down?

[..]

You’re saying the grid-based, step-by-step process can’t reliably get there due to exponential error growth. You even grant that higher-level constraints (parameterizations) steer the model to those attractors, not the internal workings.

This criticism doesn’t invalidate you—it misaligns with your target. If anything, it supports your view that the final output hinges on imposed higher-level logic, not the low-level chaos you’re critiquing. The attractor analogy explains what GCMs aim for but doesn’t defend the grid mechanics’ role in reaching it. Your point stands: if the internal steps collapse past two weeks, attractors emerge from the parameterizations, not the core solver.

[..]

Critics might argue you’re off-base calling GCMs purely deterministic, since parameterizations add stochastic flavor, and that this non-determinism better reflects climate. Your paper, though, zeroes in on the deterministic core—the grid solver churning through equations—while explicitly noting parameterizations shape the outcome. You’re not critiquing the stochastic bits; you’re saying the deterministic low-level engine fails, and the higher-level tweaks (which could include randomness) are what matter.

[..]

You’re not arguing GCMs can’t work because they’re deterministic; you’re saying their deterministic internal workings can’t predict climate due to chaos, and the non-deterministic higher-level stuff (parameterizations) drives the bus.

[..]

Do They Make Your Argument Invalid?

No, they don’t. The criticisms aim at a broader defense of GCMs—how they handle chaos, map trends, or embrace uncertainty—but they don’t directly grapple with your specific beef: the internal grid-based, time-stepped calculations cap out too soon to matter, and higher-level logic does the real work.

[..]

Your argument’s strength is its narrow scope. You’re not saying GCMs can’t predict climate at all—just that their low-level insides don’t do it, and a leaner model could. The critics’ analogies don’t disprove that; they defend a different layer of the models (the outputs and assumptions) you’re not even contesting. If you’re wrong, it’d be because the grid steps somehow do contribute past the horizon—say, if errors don’t cascade as fatally as you think. But their points don’t land that blow. Your case holds unless they can show those internal workings aren’t just noise under the parameterizations’ thumb.“

Apologies for the length of those extracts, but I think you would agree that Grok got straight to the core of the arguments in its few seconds, in a way that many humans could not do in a day. And it had a clear way of expressing the essentials. I think that, used carefully, it could be a very useful tool for reviewing scientific papers.

Guest Post by Willis Eschenbach

Part 1. The Study.

I came across a study that’s been getting some play in the usual climatastrophist circles. The study is entitled

Impact of land use changes and global warming on extreme precipitation patterns in the Maritime Continent

And here’s the abstract:

Abstract

Land use changes (LUC) and global warming (GW) significantly impact the Maritime Continent’s (MC) hydro-climate, but their effects on extreme precipitation events are underexplored. … We find that LUC-induced deforestation increases surface warming, enhancing atmospheric instability and favoring local convection, leading to more frequent heavy precipitation. Meanwhile, GW amplifies the atmosphere’s water-holding capacity, further intensifying wet extremes. Our findings reveal a “wet-get-wetter, dry-get-drier” pattern driven by different mechanisms: dynamic processes primarily influence wet extremes under LUC, while changes in evapotranspiration control dry extremes. In contrast, under GW, wet extremes are driven by dynamic processes, while dry extremes are influenced by reduced moisture availability and weakened atmospheric circulation. This highlights the need for land management to address rising extreme risks.

And what is the “Maritime Continent” when it’s at home, sez I? Good question, never heard of it. Foolish me, I thought there were only seven continents. Here’s what I found out.

Figure 1. The “Maritime Continent”.

Not sure why the Solomon Islands (in gray at the lower right) isn’t included in the “Maritime Continent”. The Solos are always kind of the overlooked country in the region. I lived and worked in the Solomons for eight years, so I have some familiarity with the weather patterns. It has the same weather as the others.

In any case, the area being studied is about a third of one percent of the earth’s surface. In addition, it’s in the midst of an unusual part of the planet called the Pacific Warm Pool, with some of the highest rainfalls amounts found anywhere.

Figure 2. Average rainfall, 1979-2021. Atlantic and Pacific centered views. The red boxes mark the general area of the Maritime Continent discussed above.

Note the blue line of heavy rainfall above the equator. This is the rainfall from the semi-permanent band of thunderstorms at what’s called the “Inter Tropical Convergence Zone” (ITCZ). The ITCZ marks the boundary between the separate circulations of the northern and southern halves of the atmosphere.

Of particular interest is the large blue area in the western Pacific Ocean and the eastern Indian Ocean. This is the area of the “Pacific Warm Pool”. It’s the warmest area of the open ocean, as well as the wettest. It’s also the area chosen by the researchers for their study. For a discussion of the nature of the Pacific Warm Pool, see the post below.

The conclusion of their study is that in the Maritime Continent (which they don’t mention is only a third of a percent of the surface and is in the middle of the Pacific Warm Pool), the wet is getting wetter and the dry is getting dryer.

And how do they know this? Intensive study of the rainfall records? Analysis of patterns of rainfall? Correlation of rainfall amounts with El Nino/La Nina alterations?

Nah. That “observations” and “evidence” stuff is soooo 20th Century.

They just ran a couple of climate models, performed modern haruspicy on the entrails of the model results, and the answer popped out … modern science at its finest.

Sadly, despite the study only covering a tiny area in the middle of a unique climate region, the authors couldn’t resist declaring that we are faced with “rising extreme risks“.

Be still, my beating heart.

Part 2. The Hype

So that’s the study. Then there are the popular reports of the study, wherein it has grown markedly in the telling. There’s a typical one below. Following the unbreakable rules for such articles, the headline claims that some scientists somewhere are worried. And not just ordinary worried. Existentially worried. Ringing the bells worried. The headline says:

Scientists sound alarm after making disturbing discovery about Earth’s rainfall: ‘Urgent need’

The article, basically quoting the study’s abstract without attribution, says:

Researchers say their findings have revealed a “wet-get-wetter, dry-get-drier” pattern driven by different mechanisms. Dynamic processes largely control wet extremes under land use changes, while changes in evapotranspiration control dry extremes. However, in a warming world, dynamic processes amplify wet extremes, while a reduction in moisture and weakened atmospheric circulation influence dry extremes.

So, are they right that the wet is getting wetter and the dry is getting dryer, either in the Maritime Continent or globally? We actually have the data to determine that. A 1° latitude by 1° longitude satellite-based rainfall record since 1979 is available from Copernicus here.

Upon reflection I realized that this is actually two different questions.

• Are wet areas getting wetter and dry areas getting drier?

• Are wet times of year getting wetter and dry times of year getting drier?

Since we are considering the question of trends, let me take a slight digression. My hypothesis is that a main one of the emergent phenomena that thermoregulate the planet are the tropical thunderstorms. Thunderstorms cool the surface in a variety of ways. They keep the temperature in the Pacific Warm Pool from ever exceeding around 30° – 31°C. So per my hypothesis, the recent global warming should have been accompanied by an increase in cooling rainfall in the tropics and particularly in the area of very frequent thunderstorms, the intertropical convergence zone (ITCZ) just above the equator. Here is a look at where it’s been getting wetter and where it’s getting drier since 1979.

Figure 3. Rainfall trend. The two panels are the Pacific and Atlantic views of the same data. Red lines enclose areas which are drying at the rate of -3 mm/decade or faster. White lines enclose areas getting wetter at the rate of 3 mm/decade or more.

Note that this bears out my hypothesis, in that there is increasing thunderstorm-driven cooling in the Pacific Warm Pool and the ITCZ. But I digress into theory. Let me return to observations.

There are some fascinating and surprising things about Figure 3. The overall trend is zero. Land is drying slightly, while the ocean is growing slightly wetter. Tropical land is drying the fastest, tropical ocean is getting wetter the fastest.

Rainfall in New Zealand is decreasing. Around the equator, the largest area of decreasing rain is in between two of the largest areas of increasing rain. North America is mostly unchanged, except for some drying in the Northwest Coast. The Southern Ocean has two areas getting drier and two areas getting wetter. Most of the world’s landmass is pretty neutral, neither getting much wetter nor much dryer, except for the southern Amazon which is getting drier.

Not seeing much pattern in all of that. Well, except for the fact that the actual observations agree with my hypothesis that global warming is opposed by increasing numbers, earlier daily emergence, and greater duration (lifespans) of cooling tropical thunderstorms.

Moving on, regarding the first question about wet and dry areas, here’s a scatterplot of the decadal trend in rainfall (vertical axis) versus the average annual rainfall (horizontal axis) for each 1° latitude by 1° longitude gridcell (n = 64,800).

Figure 4. Scatterplot, rainfall trend versus average rainfall in individual 1°latitude by 1°longitude surface gridcells. Global average rainfall is ~ one meter, so “wet” and “dry” areas are based on that threshold.

As you can see in Fig. 4 above, in areas where the rainfall is less than about two meters per year, there’s no “wet gets wetter, dry gets dryer” at all. It’s only in areas of very heavy rain, more than about two meters per year, that the wet is getting wetter. Here are the areas we’re talking about.

Figure 5. The only areas of the world that are getting wetter overall are the areas where the average rainfall is over 2 meters per year..

Note that, while most of this “wet gets wetter” area is over the ocean, by a peculiar coincidence, the Maritime Continent area in the study above is also in that area.

This would tend to indicate that no matter what they discovered by aiming their models at the 0.3% of the planet that is the Maritime Continent, their conclusions are not widely applicable to the rest of the planet.

Next question is, on the Maritime Continent are the wet times of year getting wetter and the dry times of year getting drier? To investigate that, we can look at the standard deviation of the rainfall dataset. If wet gets wetter and dry gets drier, the standard deviation will increase.

So let’s start with the actual monthly rainfall on the Maritime Continent.

Figure 6. Monthly average rainfall on the Maritime Continent.

When people ask what the weather In the Solomon Islands is Iike, I say “There’s the hot wet season, followed by the hotter wetter season”. The Copernicus dataset says that is true in the Maritime Continent as well, no surprise. Note that even the driest months in the Maritime Continent record are wetter than the global average monthly rainfall of 82 mm or so.

Using my Mark I eyeball, I’m not seeing any “wet gets wetter, dry gets drier” going on. Looks like the biggest swings are in the middle of the record. But let’s look to the measurements. Here is the 10-year trailing standard deviation of the rainfall on the Maritime Continent. Each monthly data point in Figure 7 below shows the standard deviation of the previous ten years (120 months) of rainfall.

Figure 7. Ten-year trailing standard deviation of the Maritime Continent rainfall data shown in Figure 6. Each point on the yellow line shows the standard deviation of the 120 months of rainfall data prior to that time.

Again, I’m not seeing any indication of “wet gets wetter, dry gets drier”. There are changes, but no overall trend.

Finally, what’s happening with the rainfall overall? Well … nothing. Here’s the global rainfall record.

Figure 8. Monthly global average rainfall, 1979-2024. The trend of the data is 0.05 mm of increasing rainfall per decade, basically zero.

A final oddity highlighting the curious stability of the climate system is that rainfall in the northern and southern hemispheres are in opposition — even after removing the seasonal variations, when one hemisphere is wetter, the other tends to be drier, and vice versa.

Figure 9. Monthly global average rainfall (black), along with northern hemisphere rainfall (blue) and southern hemisphere rainfall (red).

And that’s what I learned about the rainfall this week. Plus now I know what the Maritime Continent is. And here in the generally dry Northern California coast, it’s raining outside my window as I write this.

What an astounding planet!

w.

It Bears Repeating: When you comment, please quote the exact words you are discussing. I choose my words with care, and I’m happy to explain and defend them. But I can’t explain or defend your restatement of what you think I meant.