No clue, but sounds like Boney M's Rasputin (also a great-but-obscure "trivia question" goes with that. lol.)
The only reason I knew it was I accidentally rediscovered that song a few years ago. I used to love it as a little kid for some reason. It was a case of "oh yeah! I remember this!", so I started researching the artist and its history. I mean, it's not often you hear a black dude singing about a Russian historical figure (who I was also mesmerized by as a kid after reading all the failed attempts at his life). Then, to top it off, he's dancing like a Russian on stage. Anyhoo... the "trivia". The guy who managed/created the group Boney M was also the guy who actually sang the song. The black guy dancing and getting all the notoriety as being the lead singer just lip-synced and danced on stage like he was on crack (seriously - I don't know how he kept that dancing up). But everybody I think knew he was lip-syncin'ing and wasn't the actual singer. Despite this, they were popular world-wide, including in Russia, from what I recall. Fast-forward to the days of Milli Vanilli who got torched and humiliated for hiding the fact they weren't actually singing or dancing on their hits. This is the trivia - the same guy who created/managed Boney M is the guy who created/managed Milli Vanilli. The irony is that the same thing he did with Boney M having the front men just be lipsyncers with someone else singing the actual vocals that made Boney M a worldwide hit is what crushed Milli Vanilli because they tried to hide it (I guess). The manager also went on to produce other songs/groups, like I think the gorup La Bouche (of "Be My Lover" fame) was created by him, too. Second bit of trivia I learned while researching Boney M back then. Their biggest hit was probably that song "Rasputin". The lead singer of Boney M died on the birthday of the actual historical figure Rasputin. I'm glad I finally got to use this bit of useless information in this off-topic post.
Great writeup on AI by one of the best thinkers of our era, Jason Pargin: Everyone is Wrong About AI Except Me
Apple sues OpenAI over trade secrets The lawsuit alleges that a former senior electrical engineer at Apple kept a work-issued Apple laptop and discovered a bug that allowed him to access Apple's cloud file storage after leaving and while employed by OpenAI. "Recently, significant evidence has emerged suggesting individuals employed by OpenAI wrongfully took Apple's secret and confidential information regarding our unreleased technologies, processes, and products," Apple said in a statement.
... but they fight like they're drunk and blind. Especially the one that lost its head. BattleBots >>> drunken blindbots
Shower Thought so, if CPUs use prediction tables to evaluate our conditionals, to guess the branch code and start executing instructions before the condition is fully resolved, does AI do the same thing: prepare their answers before fully understanding our questions? [insert thinking T-Rex.gif]
Nope, when a cpu makes a guess... it's a binary prediction point. If it's wrong, the cache is flushed, the pipeline emptied and there is a performance penalty. In a LLM, at each step, the model is computing probability distributions across tons of possible next tokens simultaneously. It's not picking a single "branch" to speculatively execute. It is a black box in that.... it isn't possible to run in reverse to see where it went wrong. It commits to output one token at a time, and the previous token influences the next. When a cpu guesses wrong, the branch resolves and the hardware knows... a LLM has no branch resolve that either finds/doesn't find the data it needs like a cpu pipeline, no way to know if it "understood" the question... it just generates what the weights predict is most likely. It will confidently ride down that road until until you make the corrections... or frankly, once you see the LLM is going down the wrong path, it's best to start over and reframe/rephrase. This is because most models are using something of a global weighting in a secession, your whole conversation is weighing the token predictions in a certain way and once it's on the wrong track.... start over. Ultimately, YOU are the B.S. limiter/detector.
