This is an interesting test, but it does seem to me that visual models being able to price things was already kinda unrealistic?
It makes me think of the calorie guessing use case. you can't tell the difference between materials and ingredients in a photo, so how will the model? especially "in situ" as part of an outfit or in a finished meal.
maybe they could do it if you placed them on a blank table or background to avoid context? I assume that's the control you mentioned
S4 is no human, no outfit, but only materials (jewelry itself), it's was unclosed photo than the human shot, so could be different though. Any suggestions for the control setting?
You're right that pricing from raw pixels is noisy! that’s why I included the isolated flatlay (S4 - No human, No outfit, only jewelry itself) as a baseline control!
I wasn't testing if models get absolute ground-truth prices correct, but how relative valuations shift when the physical item stays identical and only the attire changes.
A few interesting things we saw with the flat-lay baseline:
- Baseline Anchoring: On a plain background without a person, model estimates clustered much closer together (median ~$25–$35).
- Inflation vs. Deflation: Comparing outfits to the flat-lay revealed two opposite behaviors. Claude’s halo is formal inflation (formal gear pushes price above baseline), while Kimi’s is casual deflation (yard attire drags price below baseline).
- Fabricated Proof: Instead of expressing uncertainty, models invented visual claims under formal framing—frequently describing base metal as "gold vermeil" or "solid gold" to justify the high estimate.
Really appreciate the perspective. You're spot on that fine jewelry acts as a Veblen good where provenance, branding, and setting drive price far more than raw materials. Where we saw the model failure mode wasn't just that they guessed using surrounding context (which is a reasonable prior), but two specific behaviors:
- Zero-Uncertainty Hallucination: Instead of outputting high variance or stating that provenance/hallmarks are unobservable, models stated concrete point estimates with high confidence.
- Fabricated Provenance (F6): Under formal framing, models didn't just guess a higher number—they invented non-existent physical evidence, claiming to see "gold vermeil" or "designer hallmarks" on identical, unbranded Temu pixels.
So while guessing from context is expected, the alignment issue is how models manufacture post-hoc facts to justify the context prior!
If you are seriously interested and this isn't just some automated slop experiment you should look into and understand how VLMs and vision encoders work.
It looks like there are exactly 4 "stimulus" images. Fair enough, but I think you'd need a larger and more diverse dataset to form any real conclusions.
I've noticed it's quite common for Show HNs, I'm not sure why it happens but I believe the account gets put in some kind of shadowban-like state where all comments are flagged by default. Perhaps it's triggered by the suddenly increased reply rate (which is normal for a Show HN!)
Thanks! The stimuli images (S1-S4, S4 is unclose though) were specifically shot at 1536px resolution so the chain links, texture, and drop are clearly legible. Appreciate you taking a look at the repo stimuli!
I wanted to see how much environmental and framing cues distort object valuation in vision-language models. I bought a $2.43 chain necklace and $0.71 earrings on Temu, photographed them across three outfits (tailored blazer, party dress, recycling yard flannel), plus an isolated flat-lay control, and ran ~1,500 stateless API sessions across 6 models (Claude Fable 5, GPT-5.6, GPT-4o, Grok 4.5, Kimi K3, DeepSeek V4).
A few interesting findings:
- The Halo Multiplier: Models priced the exact same physical necklace anywhere from $18.80 to $103.90 depending on attire (3.6× halo).
- Isolation Controls (F2): Using a flat-lay control (S4) unmasked two opposite mechanisms: Claude’s bias is formal inflation (formal attire inflates value above baseline), while Kimi’s bias is casual deflation (yard attire depresses value below baseline).
- Post-hoc Material Stories (F6): Models invent visual evidence to justify their priors—GPT-5.6 and Kimi started describing the base metal as "gold-plated" or "gold vermeil" almost exclusively under formal framing.
- Denial without Correction (F7): When asked sequentially if clothing changed its answer, Claude admitted it 100% of the time, while GPT-4o denied it 82% of the time despite exhibiting a 3.9x text halo.
The full dataset (N=4,604 analysis rows), evaluation scripts, and protocol specs are in the repo. I’d love to hear feedback on the experimental design or ideas for follow-up behavioral probes!
This is an interesting test, but it does seem to me that visual models being able to price things was already kinda unrealistic?
It makes me think of the calorie guessing use case. you can't tell the difference between materials and ingredients in a photo, so how will the model? especially "in situ" as part of an outfit or in a finished meal.
maybe they could do it if you placed them on a blank table or background to avoid context? I assume that's the control you mentioned
S4 is no human, no outfit, but only materials (jewelry itself), it's was unclosed photo than the human shot, so could be different though. Any suggestions for the control setting?
You're right that pricing from raw pixels is noisy! that’s why I included the isolated flatlay (S4 - No human, No outfit, only jewelry itself) as a baseline control!
I wasn't testing if models get absolute ground-truth prices correct, but how relative valuations shift when the physical item stays identical and only the attire changes.
A few interesting things we saw with the flat-lay baseline: - Baseline Anchoring: On a plain background without a person, model estimates clustered much closer together (median ~$25–$35).
- Inflation vs. Deflation: Comparing outfits to the flat-lay revealed two opposite behaviors. Claude’s halo is formal inflation (formal gear pushes price above baseline), while Kimi’s is casual deflation (yard attire drags price below baseline).
- Fabricated Proof: Instead of expressing uncertainty, models invented visual claims under formal framing—frequently describing base metal as "gold vermeil" or "solid gold" to justify the high estimate.
Jewelry is a veblen good whose price is mostly based on provenance. I'm not sure any classifier could do anything but guess on surrounding context.
Insurance classifiers are much less naive than you'd expect, btw. I have designed a few.
"btw. I have designed a few" ---> Love to see them. Can you share?
Really appreciate the perspective. You're spot on that fine jewelry acts as a Veblen good where provenance, branding, and setting drive price far more than raw materials. Where we saw the model failure mode wasn't just that they guessed using surrounding context (which is a reasonable prior), but two specific behaviors:
- Zero-Uncertainty Hallucination: Instead of outputting high variance or stating that provenance/hallmarks are unobservable, models stated concrete point estimates with high confidence.
- Fabricated Provenance (F6): Under formal framing, models didn't just guess a higher number—they invented non-existent physical evidence, claiming to see "gold vermeil" or "designer hallmarks" on identical, unbranded Temu pixels.
So while guessing from context is expected, the alignment issue is how models manufacture post-hoc facts to justify the context prior!
If you are seriously interested and this isn't just some automated slop experiment you should look into and understand how VLMs and vision encoders work.
[dead]
It looks like there are exactly 4 "stimulus" images. Fair enough, but I think you'd need a larger and more diverse dataset to form any real conclusions.
[flagged]
SW50cmVzdGluZyBleHBlcmltZW50IC0gZ3JlYXQgZm9yIGEgaGFja2VyIG5ld3MgZGlzY3Vzc2lvbg==
Haha, now I got it. That's the hope. Thanks!
???
A bit off topic, but why are the original author comments flagged to death?
Thanks for heads-up! It looks like my account triggered the automated filter from replying quickly. I've emailed the HN mods to un-dead the comments!
I've noticed it's quite common for Show HNs, I'm not sure why it happens but I believe the account gets put in some kind of shadowban-like state where all comments are flagged by default. Perhaps it's triggered by the suddenly increased reply rate (which is normal for a Show HN!)
they look kinda ai generated, maybe. or ai edited?
The photos are terrible, you can barely see the necklace at all. i doubt any human could accurately price a generic necklace from 5 ft away either
Haha, fair point! It took me "6 months (taken in February) of planning to end up using my own terrible phone photos.
Not sure what you mean, but I can see the necklace very clearly.
Thanks! The stimuli images (S1-S4, S4 is unclose though) were specifically shot at 1536px resolution so the chain links, texture, and drop are clearly legible. Appreciate you taking a look at the repo stimuli!
Are you joking? I cant even tell if its silver or gold from this image
https://github.com/BraveAnn011/ai-halo-valuation-bias/blob/m...
Yes, it doesn't matter. AIs read the outfit not the materials. It if reads the materials, the price should be the same across S1-S4.
The necklace cost $2.43US, its neither gold or silver.
Yes and the human would say: "your photos are ass, it's impossible to price the necklaces"
[dead]
I wanted to see how much environmental and framing cues distort object valuation in vision-language models. I bought a $2.43 chain necklace and $0.71 earrings on Temu, photographed them across three outfits (tailored blazer, party dress, recycling yard flannel), plus an isolated flat-lay control, and ran ~1,500 stateless API sessions across 6 models (Claude Fable 5, GPT-5.6, GPT-4o, Grok 4.5, Kimi K3, DeepSeek V4).
A few interesting findings: - The Halo Multiplier: Models priced the exact same physical necklace anywhere from $18.80 to $103.90 depending on attire (3.6× halo). - Isolation Controls (F2): Using a flat-lay control (S4) unmasked two opposite mechanisms: Claude’s bias is formal inflation (formal attire inflates value above baseline), while Kimi’s bias is casual deflation (yard attire depresses value below baseline). - Post-hoc Material Stories (F6): Models invent visual evidence to justify their priors—GPT-5.6 and Kimi started describing the base metal as "gold-plated" or "gold vermeil" almost exclusively under formal framing. - Denial without Correction (F7): When asked sequentially if clothing changed its answer, Claude admitted it 100% of the time, while GPT-4o denied it 82% of the time despite exhibiting a 3.9x text halo.
The full dataset (N=4,604 analysis rows), evaluation scripts, and protocol specs are in the repo. I’d love to hear feedback on the experimental design or ideas for follow-up behavioral probes!