I can’t find the podcast but I saw recently someone who (I think) is affiliated with TAG say something to the effect of ‘the hardware is the easy part, the AI grading is the hard part’ and it just made me laugh a little inside.
Every company struggles to reduce damages, dust, misalignments etc and thats with a human doing it. I can’t imagine a machine doing it without tons of consistent issues
edit: found it. He also mentions that PSA apparently tried to acquire them already. This whole segment is quite interesting
Yeah I personally dont want any AI form of grading or complete technological control of cards. I dont grade hardly at all, but the main thing over everything I enjoy about most grading companies is that there is an actual human with a brain looking at the card, and trying their best to determine authenticity, integrity, and giving it a grade deemed suitable for humans to enjoy. Many have already talked about the potential problems, but how to overcome those problems is a hurdle. Even after the hurdle is completely figured out though there is still the “what if” that heavily outweighs modern grading. Its almost like saying “we’ve got maintenance robots to fix the robots!”…like what?!
That’s actually a pretty wild take. If we believe their main value proposition as a company, then the AI is already a solved problem. With the state of machine learning today, a small but expensive research team could put together a comparable grading algorithm.
The hardware is a solved problem too which is obvious because grading companies exist. The unsupervised “podification” of the full process is where my skepticism begins.
On a plane so will endeavor to keep my initial thoughts brief:
This is exciting news to hear - Decentralized grading would be the inevitable next step once automated grading is “perfected.” The fact that TAG is publicly announcing their next stage is an indicator of both their progress toward automated grading tech and corresponding software development.
Regarding the valid concerns around unmanned grading, there would definitely need to be contingency plans in place. I would imagine once they are ready to deploy their first remote grading pod, they will keep it local near their headquarters while they routinely run test and determine maintenance procedures when they are capable of scaling.
Automated encapsulation is nothing new. Automated grading of the raw card has been a work in progress. Automating both into one system is where customers will wait and see to determine success.
I would imagine that if TAG implemented a subscription model or even a tiered fee, they could generate multiple layers of revenue such as a “pre grading” fee for binder collectors or other grading company loyalists.
when u have ig “influencers” willing cracking psa and bgs 10 to regrade with “TAG” u know something is up. what is that TAG 10 illustrator worth then a billion dollas
I mean sure it was a troll but AI doesn’t know stuff a human being can swoobot is a great ai but let’s be honest I’d rather have a real person answer my question ai just doesn’t have enough stimuli for me to justify TAG
I get the without human bias but sometimes human bias is needed
Not all “AI” is the same. The details like how the model was trained and what it’s supposed to do matter. ChatGPT was trained to interpret and produce syntactically accurate strings of text. The truthiness of the text is an afterthought.
In the example provided, what is actually happening is that a model is run to convert the image into a text description and the text description is fed into the GPT prompt. The image interpretation was not trained on detecting real cards vs fake cards so it’s not suprising it doesn’t mention that it’s a fake card. The text component is just summarizing the information given it it including my correction.
The point being that this is not representative of how TAG grades. Presumably the TAG algorithm is trained specifically to do the task of recognizing true damage on cards based on image data and assigning it a numerical value. This is something that machine learning models are well suited to do and to a degree of accuracy that should be satisfactory in general.
The main problem with AI/machine learning today is that if your new data is a weird edge case or doesn’t have suitable representation in the training data, the prediction made can be confidently wrong. To a degree that would confuse a human that such a “smart” algorithm would get something so wrong. The example above is a situation like that. Or how self-driving cars sometimes do something inexplicable.
On a side note, the weird thing about TAG is that every time they update the model to improve any edge cases, they are also effectively making all past grades obsolete, since it’s impossible or at least extraordinarily hard to know whether previously graded cards will get the same grade again.
not a bad idea but I would be hesitant about things like authentication, damage being done to cards, and the system unable to grade/recognize your cards
imagine a dust particle getting stuck on the lens so it thinks every card is a 5
Funnily enough, this just came up on reddit Here. Tag updated their damage report after the user pointed out the issue (i dont know if it clearly documented that it was changed after the fact) but still kept the grade as a 10 with a justification that it would still qualify as a 10, just not pristine. Is that sketchy or true? Hard to say but it certainly highlights some of your original concerns about their style of grading.
Perfect for those that love standing in line for 3-4 hours at a local Gamestop/Walmart/Costco/Target for a single modern ETB or booster box then another 3-4 hours standing in line at a TAG kiosk…then hope no one was scoping you out looking to follow you home with all the junk slabs you just graded…
On a slightly more serious note I appreciate what TAG is trying to do to innovate, err, “disrupt” (their investors absolutely adore that word) the grading industry but this sounds like a solution in search of a problem when thinking about all the things that could go wrong logistically with these “pods.”
But I guess if TAG wants to really gain some traction when their Reddit astroturfing posts aren’t working then giving folks the ability to absolutely inundate the market with a billion TAG slabs isn’t the worst strategy ever (even if it may have the opposite effect of people eventually thinking, ‘why is every slab I see a TAG slab and why are most of them literal junk’)?
Never graded with TAG but people seem to really love them and unlike BGS they are growing not contracting
take a look at a small sample of these 3k comment and witness how coocoo ppl are going for TAG
this video has 1.6million views. I promise you Tag has already seen a decent bump from this short alone
and I’ll restate… for detail oriented customers who love the subscore system and want to understand why a card grades the way it does, I can completely see TAG poaching at BGS customers specifically
Ive never seen such overwhelmingly positive feedback for a grading company before
just wait until the end of the year and compare overall cards graded. Honestly, wouldnt suprise me if TAG overtakes BGS this year in market share
what do you have against TAG?
just click on the users. many of them have their own channels and followers and make their own content. not to mention some of those accounts are ancient and from the earlier part of the century. Not bots
but its clear now they have many haters on here so i need to figure out why. Im curious for those who graded with them what are your gripes. I genuinely ask as i havent graded with them yet
