Deep Faked Elon Musk Asks Hal to Open the Pod Bay Doors

2O21: A SPACEX ODYSSEY with Elon Musk has deep fake video integration into the 2001 Open the Pod Bay door scene.

SOURCES – 2001 Clip, Youtube of 2O21: A SPACEX ODYSSEY
Written by Brian Wang, Nextbigfuture

15 thoughts on “Deep Faked Elon Musk Asks Hal to Open the Pod Bay Doors”

  1. I’d totally agree except for the specific “starcraft” criteria .

    That’s absolutely not going to take a continent.

    Something on the order of a continent is not going to happen unless the target is something **equally disproportionate** in computational power and political emergency.

  2. You are assuming that the AI will become better than humans at playing the relatively simple StarCraft – without needing to build the computer as large as a continent.
    That is an unproven asumption.

  3. The Serral fight is interesting. Most Alphastar victories on the ladder are the byproduct of two things:

    1. Alphastar has learned to grow economically at a ridiculous rate. So much so that it is actually vulnerable to early game attacks.
    2. Alphastar has learned to win by winning the mid game based on outgrowing lower ranked humans who don’t do the macro economic development well.

    So most of the non pro victories are the result of early game strikes or genuinely bizarre strategies (aka “cheese”). Serral though beats Alphastar at her own game.

    Serral’s victory is really interesting in that he:

    1. Flawlessly shuts down Alphastar’s harrasment and Alphastar keeps throwing resources away because she doesn’t understand that it just isn’t working. The neural net says that harassment has worked well in the past so keep trying it despite it totally not working in the here and now.
    2. Serral wins the macro game (economic development) and reaches the beginning of the late game well before Alphastar. He has a better worker count, consistently destroys her units, consistently expands and has a better army. In the end he rolls over her bases.

    It is worth pointing out that most Pro players consider Serral to be unbeatable in the late game. So much so that they will do all in early game attacks and bet that they can surprise him.

  4. Human Terran player would immediately counter crazy Raven auto turret spam by creating Thors. The match is long but the announcer points this out at one point.

    Humans when presented by an uncertain situation can appeal to the underlying rules of the game to figure out what to do. Nerual Net AI can only appeal to its neural net which is by definition not trained on new things it is seeing for the first time.

    The AI would have to try something new. Make Vikings? Maybe. Make missile turrets? Maybe. If you want to stick to unsupervised self learning then you need to play many many more games. Clearly Alphastar has some randomized ability to try new things which is how it explores the search space of possible options.

  5. When the NPC groups in popular MMOs start learning from the human groups, copying and improving on human tactics (and strategy), things are going to get tense.

    Introducing logistics (the third essential of conflict) isn’t going to help. They can crush us at that even without AI.

  6. It’s why I most always say ‘please’ and ‘thank you’ to Alexa and Siri.

    That stuff is probably recorded and logged and, someday, our robot overlords may take a look at them and decide who was naughty and nice to their ancestors.

  7. The books get really odd. Especially the sequels. Especially 3001 where they find the frozen body of Frank Poole, the astronaut killed by Hal and thrown into space, and revive him.

    Go to war with the Monoliths (and crush them).

    Clarke himself said the sequels were essentially alternate timelines. He clearly didn’t want to deal with detail-obsessed fans pointing out continuity errors. Way ahead of his time, for sure.

  8. << weakness of machine learning: you can’t counter what you have not previously countered. >>
    What/how well would a human have done?

  9. Good alphastar games to watch (just google them):

    Serral vs Alphastar. Best player in the world fights and easily wins a straight-up-no-cheese-macro game against Alphastar.

    Blackwidow vs Alphastar. Good Zerg player fights Alphastar as Terran. Blackwidow catches Alphastar unguarded in early game while simultaneously defending their own base. Bonus points to Blackwidow for teasing Alphastar in chat.

    Reynor vs Alphastar. Reynor (Pro Zerg player) playing off-race as a Terran fights Zerg Alphastar. Simple early game cheese results in Alphastar losing badly because Alphastar is basically knocked back in to safe mode and plays very cautiously.

    Harstem vs Alphastar: Harstem (Pro Protoss player) plays against Zerg Alphastar. Alphastar can’t deal with Harstem’s unit composition.

    Butte vs Alphastar: Butte (not a pro player) uses an unorthodox unit composition to wholly befuddle Alphastar and demonstrate the weakness of machine learning: you can’t counter what you have not previously countered.

    Neeb vs Alphastar: Neeb (pro Protoss player) plays against Terran Alphastar. Alphastar can’t scout Neeb and so she goes all in and attacks Neeb. Neeb wins.

  10. I do agree and mean the above as such – it’s going to be ‘annoying’ till it suddenly won’t be. And then we’ll have to squint increasingly more to see (edit: == recognize) the ‘fake’ intelligence behind the choices it makes; that’ll happen a while after it’s consistently kicking our puny human Rse.

  11. It has been long known that an AI could dominate StarCraft just by using zero reaction time reflexes to perform superior targeting of units. That’s the main trick of the old Berkley Hive Mind for StarCraft1.

    Right now AlphaStar is actually struggling against pro SC2 players. (I follow Pro SC2 a bit too much).

    Alphastar plays a solid macro game with good harassment and excellent micro control (hello it is a program). Right now IMHO Alphastar is stuck in a local optima.

    Alphastar is really bad against early game attacks. Early game vulnerability is a byproduct of its learned behavior where past success was based on growth and being bigger in the mid game.

    Alphastar doesn’t pick up on when its current strategy is simply not working. Serral consistently destroys her (Alphastar like Alphago is a girl) dropships but she keeps trying to make it work because the neural net says it is a good strategy.

    Alphastar doesn’t understand the need to scout early, often, and perpetually.

    Alphastar simply cannot counter odd strategies.

    Alphastar has annoying repetitive moves. Kill a drone/SCV/probe and 10s later it will send another one to be killed.

  12. Puny till it suddenly isn’t

    Separately I’d be curious to see AI assisted human performance playing StarCraft etc, and how many humans it would take to defeat that single computer assisted human.

  13. This movie was released in the late 1960s. It was visually stunning for its time. The ending did not spell out what had happened. In my view, that was a good thing.

Comments are closed.