PROTECT YOUR DNA WITH QUANTUM TECHNOLOGY
Orgo-Life the new way to the future Advertising by Adpathway
Generative Artificial intelligence (AI) now appears across nearly all digital data processing, from generating text, images, and videos to coding advanced algorithms; AI agents manage complex tasks and projects. Caution remains necessary with AI-generated content, as a master’s-level example shows. Such work demands careful prompting plus a full review, testing, and correction – across all AI-assisted uses. Figure 3 from Paper 3 by Epp et al. (2010) should be revised for a conference slide. It is especially important that the figure use landscape orientation, legible fonts, and more modern colors with suitable line styles and weights. Ensure any required details – otherwise placed in the caption – appear directly in the figure.

Using ancient DNA from sediment cores, Epp et al. (2010) tracked Brachionus rotifer haplotypes in Kenya’s Lake Sonachi since early 1800s. One haplotype dominated for 150 years, with two turnovers matching environmental upheavals: volcanic ash deposition and a 1940s lake-level lowstand. Two other haplotypes appeared beforehand, one possibly another species.
In January 2011, the designer and co-author of Trauth & Sillmann (2018), Elisabeth Sillmann at blaetterwaldDesign.de, updated the graphic for a research proposal. The image was rotated for the presentation. The color scheme was adjusted to match the proposal, using specific shades of blue, green, and orange. Font sizes and line thicknesses were changed, and missing details – such as Volcanic Ash – were added to the original figure.
In the age of AI, I passed the work to Claude by uploading the vector graphic and issuing a prompt that, admittedly, lacked careful wording:
TRAUTH 12 Nov 2026: “This figure is very ugly. Could you please modify it and make it more beautiful, by adapting for instance line widths, the colors, the size of the pie charts, and everything you may think about?”

Claude spent several minutes on the task, recording each phase with notes that noted the use of the Python package Matplotlib. The completed graph arrived alongside a thorough log beginning with the statement
CLAUDE 12 Nov 2026: “I’m redrawing the figure in matplotlib and checking the first render now. I redrew the figure from scratch (vector PDF plus a 300 dpi PNG), reusing the original core photo.”
So far, so good.
At first glance, the graph seemed improved: the colors felt more modern, and the labels were larger. Claude did not rotate the image because I had not requested it. A closer inspection showed substantial changes. First, the proportions of the Brachionus rotifer haplotypes shifted in the pie charts; the Other haplotype appeared in only one. Stratigraphic assignments were also modified, along with details like smoothing the lake level curve. This adjustment is unacceptable in scientific circles, perhaps due to my vague instructions. Yet Claude should have recognized that a scientific graph follows clear rules. The report claimed quantitative values were estimated by eye from the original. This is surprising, since Claude lacks eyes. More importantly, the vector graphic in PDF format holds the needed data—namely, polygon vertex coordinates—that Claude could have used instead of a non-reproducible estimate. In any case, you can ask Claude to revise it in a second step:
TRAUTH 12 Nov 2026: “Unfortunately, you changed the ratios of the various segments in the pie chart. Furthermore, you smoothed the lake level curve. In science, of course, this is not acceptable. Could you please revise and correct your work?”
which yields the response from Claude
CLAUDE 12 Nov 2026: “You’re right: I estimated the curve and the pie fractions by eye, and that isn’t acceptable. The original PDF contains the exact vector geometry, so I’ll extract the curve and wedges from it directly.”

The revised version improved markedly, showing no obvious flaws and requiring only minor detail work. This case illustrates—like the students’ experiences and their exercise outcomes—how vital careful review of AI content can be, possibly by a person instead of more AI tools. Finally, the outcome confirms that skilled prompting strongly affects results.
Artificial Intelligence Statement
I drafted and wrote the text myself, but I used Apple Intelligence to revise it for better readability before reviewing the revised version and correcting any errors. The only exception is the summary of Epp et al. (2010) beginning with Using … that has been written by Claude based on my prompt, Could you please write a 50 word summary of the paper by Epp et al. 2010, based on the abstract of the Paper?, which provided a summary exactly 50 words long.
References
Epp, L. S., Stoof, K. R. & Trauth, M. H. (2010) Historical genetics on a sediment core from a Kenyan lake: intraspecific genotype turnover in a tropical rotifer is related to past environmental changes. Journal of Paleolimnology, 43, 939–954.
Trauth, M.H., Sillmann, E. (2018) Collecting, Processing and Presenting Geoscientific Information, MATLAB® and Design Recipes for Earth Sciences – Second Edition. Springer International Publishing, 274 p., ISBN 978-3-662-56202-4. (MDRES)






















English (US) ·
French (CA) ·