My personal AI benchmark: "Generate an SVG of a frog with a Habsburg jaw."

3 hours ago

The annotations are mostly structural labels but include editorializing about the jaw feature, such as 'massive protruding mandible' and 'recessed upper lip'.
The first SVG describes the jaw as 'massive protruding mandible' and the upper lip as 'recessed, tucked behind the jaw', with protruding lower teeth.
The second SVG uses 'HUGE protruding Habsburg jaw' and 'droopy regal eyelids', adding implied royal bearing and mood.
The third SVG uses 'massive elongated protruding mandible' and 'underbite mouth: receded upper lip, protruding lower lip', focusing on anatomical deformity.
Overall, the annotations emphasize exaggerated deformity beyond neutral labeling, with varying degrees of editorializing across the three SVGs.

Many AI company logos feature circular shapes with central openings, gradients, and radiating elements, leading to comparisons with buttholes.
Examples include OpenAI's logo redesign, described euphemistically, and Anthropic's Claude logo, which resembles a drawing from Kurt Vonnegut's book.
Reasons for this trend include circular design psychology (representing wholeness, friendliness), unintentional biomimicry, copycat effects, and design by committee.

Critiques adaptations of the mirror test for LLMs as flawed because they translate visual tests into text.
Proposes a better analogy: modify an LLM's own textual output subtly and see if it notices the anomaly.
Describes an experiment with Gemma 4 31B-IT where corrupted text (replacing 'g' with 'sg') was introduced.

Arp 906 days ago

Arp 90 consists of two interacting galaxies in the east.
A weird distorted galaxy to the west may be a tidal fragment or gravitationally bound irregular galaxy.
Total exposure time was 2.6 hours, with 2.3 hours used in stack.

Analysis focuses on NYT crossword layouts: 15x15 daily and 21x21 Sunday versions.
Visualizations show character distribution likelihoods per cell, not overall frequency.
Characters grouped by placement patterns, e.g., 'Great starters' vs. 'Great enders'.

Hasty Briefsbeta