iiinova — where your digital self keeps evolving

Niche, unlikely problem? Maybe. But if reviewers on panels are using AI to review, then they're probably also using AI to write. And scientists who are reviewing are exactly the people who know the niche area well enough to ask this sort of niche prompt.

2201

so if a grant or unpublished manuscript is part of training data, then I see no reason an LLM wouldn't spit out those novel ideas at some point. And the user would see it as a novel idea, since it isn't *publicly* available elsewhere.

1200

AI proponents say a potential contributions is putting together information in new ways, or developing novel research ideas. If your prompt asks for a novel research idea in [discipline], then it seems logical that the LLM will give increased weight to training text suggesting novel research ideas.

1150

So now my ideas and words and data can be used by the corporations to feed and train their future gen AI. So my ideas are no longer mine. They are available to anyone with a computer and an internet connection. Which would be less concerning if the manuscript were accepted and grant were funded.

29412

Manuscripts and grant proposals submitted for *peer* review are typically meant to be confidential. I'm fairly confident the text of these confidential documents have now been uploaded to an LLM. I don't know which one, but I'm guessing it wasn't a 'local' LLM sandboxed on someone's computer.

211518

Richard Waite @waiterich.bsky.social · 19/08/2026

I've now recieved a rejected manuscript review where a reviewer submitted LLM-generated content (100% confident), and a rejected grant review where a reviewer submitted LLM-generated content (80% confident). On it's own, this is somewhere between frustrating and infuriating. But my main concern:

310124

Reposted by Andrew Kniss

How many times did you realize this - once or twice?

011

Andrew Kniss @wyoweeds.bsky.social · 19/08/2026

I have travel bottles of toothpaste and shampoo that are similar in size and shape. This morning in my hotel room I realized this is not ideal.

1101

Not immune, indeed. I might suggest more susceptible, even.

000

Proliferation of AI slop makes it more difficult every day to figure out what is actually known about different topics. What a goddamn mess.

230

A lot of the discussions about AI in academia have focused on students, grading, etc. But I think there is even more of a problem in academic research... Just had a manuscript clearly "peer" reviewed by an LLM. And who knows how many AI written and AI reviewed grants are in the system.

130

Chinese farmer kills 25 acres of crops after following AI-generated weed and pest control advice — farmer trusted pesticide recipe after months of successful advice

Reports like this are almost certainly going to increase as these tools become more trusted. "Farmer kills 25 acres of crops after following AI-generated weed and pest control advice." share.google/uDDq4j9nOXJw...

share.google

AI is useful, but you should never blindly trust it.

041

Reposted by Andrew Kniss

Roope Kaaronen @roopekaaronen.net · 13/03/2025

Pleased to see our work published: The Ties That Bind: Computational, Cross-cultural Analyses of Knots Reveal Their Cultural Evolutionary History and Significance We analysed knots across 12,000 years and 82 societies. Time to tie a thread 🧵 about why knots matter. doi.org/10.1017/S095...

A circular dendrogram (phenetic tree) of knots, made using the ggtree package in R. A text-readable and high-resolution PDF-version of the tree is available as a supplementary material.

420277

There are a LOT of products out there being marketed to farmers. Many of these products have little to no independent testing to evaluate whether they work. A couple years ago, we tested two products marketed as "soil detoxifiers" - and we found they didn't seem to do anything.

0105

Obviously it isn't quite that simple, but we found no evidence in this study that the 'soil detoxifier' products we tested worked as claimed.

030

Betteridge's law of headlines generally applies to academic papers also. If the title is phrased as a question, then the question can be answered with "no"

160

The Onion @theonion.com · 31/07/2026

New publication out today: "Can soil detoxifiers mitigate crop injury from residual herbicides?" doi.org/10.1002/agg2...

270

Reposted by Andrew Kniss

Iran War Enters 6th Justification

393412462

Andrew Kniss @wyoweeds.bsky.social · 20/07/2026

When grant programs are posted only 6 weeks before the deadline, but then funding decisions aren't announced for another 9+ months, it really makes managing a research program stressful...

050

Reposted by Andrew Kniss

Laurel Gale @laurelgale.bsky.social · 12/07/2026

A great blue heron coming in for the landing. Captured this morning. #birds #nature #wildlife 🐡📷

A photo of a great blue heron with its wings spread wide and its long legs extended down as it prepares to land on a rocky riverbank.

18611

Reposted by Andrew Kniss

Bethany Brookshire @beebrookshire.bsky.social · 10/07/2026

There's an OMB rule change that would require political oversight of scientific grants, and could pull any that don't "reflect presidential priorities" Another country has tried this before! It was the USSR and it was called Lysenkoism. By @thsaey.bsky.social www.sciencenews.org/article/omb-...

sciencenews.org

Here’s what happens when you put politicians in charge of science

Proposed federal rules would rely on political appointees to decide how a lot of U.S. science gets done. History shows the consequences of such actions.

22871431

Andrew Kniss @wyoweeds.bsky.social · 28/05/2026

Potato Oles from Taco John's. Preferably with nacho cheese, but not necessary to be the best.

120

Andrew Kniss @wyoweeds.bsky.social · 27/05/2026

AI also tells me that a Turkey Tap can also be another name for a bib cock. So there's that.

010

From yesterday - a thread about a new publication where we look at seed germination ecology, and how two species might respond to cover crops for weed management.

020

This means that common lambsquarters will usually be suppressed by a vigorous crop canopy, and you can use a cover crop to suppress new emergence of this weed. BUT: common lambsquarters has the genetic potential to evolve resistance to cover crops!

010

For common lambsquarters, though, things are different. Germination of most of the genotypes were heavily suppressed by far-red light. But not all of them!

100

So what does this mean from a practical perspective? Well, for kochia, it means that the light environment does fuckall to suppress kochia germination. Bottom of a corn crop? Kochia will germinate. Using cover crops for weed suppression? Good luck. Kochia will germinate.

100

Plants have evolved to NOT germinate when they sense a low R:FR light environment, because they 'know' that they will be entering a very competitive environment. Better to wait until they sense a higher R:FR, indicating there are fewer plants to compete with.

100

The reason we're looking at far-red light is that in the bottom of plant canopies, the environment is depleted of red light (used by the plants for photosynthesis). So at the bottom of a forest or a corn canopy, there is a very low ratio of red light to far-red light (R:FR).

110

And kochia? Well, kochia don't give a shit about your silly theories about light quality. If there's water, kochia will germinate.

120

The coolest part (IMO): although germination of most annual weeds (including common lambsquarters) is suppressed by far-red light, we found a few lambsquarters genotypes that germinate very well under this light condition.

100