site banner

Friday Fun Thread for August 8, 2025

Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.

2
Jump in the discussion.

No email address required.

So these are some curious results, and mirror the issues I was having with the models I tried. For Grim Reaper of Love, it does correctly not that 45Cat lists a May 1966 release date (which every model was able to do), and also correctly notes the May 28 Billboard review, which it was the only model to actually find, since most of the others just defaulted to the first date charted. The curious issue is with the ARSA data. It did indeed appear on the WLS June 10 chart. However, this was not the earliest chart it appeared on. That would be the May 9 KBLA chart, and the prior Monday would be May 2. The even more curious thing about it is that the single appeared in 35 charts documented by ARSA prior to the June 10 WLS chart, so I don't know why it would have picked that one. This is, I guess, somewhat of an improvement; the only other model I tried that even claimed to use ARSA data was Grok, and it simply made up entries that didn't exist! The most interesting thing about this, though, is that it didn't actually follow the instructions. Maybe I could have been a little more clear, but the instructions said:

If ARSA and Billboard data are both available, use the earlier date

Maybe I should have specified that I wanted the earliest date, which would have been the date of the May 28 review, making the correct date based on the data the model actually used to be May 23, 1966. Then again, I thought I specified early that the month of release given by 45Cat and RYM should take priority, so even if this wasn't clear, it should have preferred the May date. In any event, it didn't get the correct ARSA date, so this counts as a fail.

Moving on to Feel the Heat, US Copyright data gives a publication date of June 16, 1980. Maybe this was the search engine it was trying to use, but it nonetheless didn't use it. I give it props for using Cash Box, which I don't even use that much because the available data is fragmentary and not easily searchable (or at least it was when I started doing this a decade ago), and it does point to the correct issue. However, it runs into the same problem of following instructions when it was told to use the date preceding publication but inexplicably picks a date after the date of the issue. Honestly, there must be something up with the pro model, because the free ones I tried didn't seem to have any problem following instructions, and at least gave plausible dates based on the information they had. Here I get two dates that are not only incorrect, but don't actually follow the rule. I had high hopes for this but at this point I can only consider it a failure. If you're interested in running this further, I can try to make the rules a little more explicit and find some other releases to test how it can do different things, but suffice it to say my opinions of AI capabilities haven't appreciably improved.

Thank you for taking the time to look into it yourself! All I did was copy and paste your prompt and egg the model on. It might be interesting to try the Agent mode, either on this prompt or a new one. I do have access, and I can try it when I get the chance.