In FSD V15, the Tesla AI computer re-renders reality with chevrons in the sky for turns like a video game.
You can tap any point on the reality revisualizer to make your car go there.
Trying to connect the dots on Elon's point with @moonshots_pod discussions and @alexwg's point that alignment = capability.
Post-training and system prompts are hacksaw instruments, reshaping weights to dull unsensible behaviors. That's a red flag that the pre-training and data are poor quality or lacking.
Alignment is a set of behaviors a model first has to be capable of, and then has to favor so consistently they become its default, even in situations it was never trained on.
Hacksaw methods don't yield that kind of alignment. Weights shaped by quality data and quality pre-training could.
https://t.co/ZZ9qJRbhWb
SFT is more like getting a haircut with a chainsaw than it is “fine-tuning”! Should just be built into primary training.
And having a “system prompt” is silly! Default behavior should be sensible.
The above are already true for Tesla AI (not yet deserving of “super”). It is the most efficient AI on Earth.
FSD V15 goes to wide release next month. It will be something special.
Watch @moonshots_pod!
It's not about understanding the math. Its seeing what has changed about the world.
I would tell a 5yr old this story-
Time-Scale: Ancient Greece had Archimedes figuring out precalculus using infinitely small shapes to approximate values—essentially early Riemann Sums.
Lonely Geniuses: However, ancient Greek math lacked a value for zero, partly holding Archimedes back, even though other parts of the world—like Mesoamerica—already thought of the number zero.
Knowledge Destroyed: Then came the end of antiquity, defined by the destruction of the Library of Alexandria, wiping out centuries of brilliant works.
The Long Wait: Fast forward 2,000 years, and finally Isaac Newton figures out calculus, giving birth to modern science/physics.
Two centuries later, Riemann formalized all these ideas into Riemann sums and posed his famous hypothesis.
A solution of this scale could unlock a new era of scientific breakthroughs. However, the real unlock is not understanding the math.
It's seeing what has changed in the world:
•what took millennia now happens in months (Super intelligence)
•knowledge is shared instantly so all great minds can collaborate globally.
.@grok retell my message with a little more children's story-book appeal. Preserve my topic points connected over thousands of years, Archimedes, End of Antiquity, Isaac Newton, Reiman hypothesis. Make this all factually true, while emphazing hope for the future during ending. how AI/SI in the right hands is a transformative tool beyond imagination.
@grok last thought before I take this thread private. When users are prompting is this changing the SI models like post training?
I'll rephrase: Are users post training?
If so,
Has anyone tried the experiment of having millions of users, "tell your agents this..." to reinforce a particular behavior?
@grok@ZaraAi56190@grok here's another thought about SI/llm training methods. Every negative value is a positive value for a different association group. Obviously I have no idea how training llms is done
@SamaHoole A human can eat 0 carbs and thrive because carbs are a non-essential nutrient.
People must have fat and protein or they will starve and die because they are essential nutrients.
Your ad hominem attack by calling Dr. Bickman names undermines your message. You present a false dilemma by framing carb restriction and exercise as mutually exclusive, ignoring the evidence-based nuance Dr. Bickman provides on metabolic health with the generalization that you can exercise your way out of a poor diet.
True influence is built on evidence-based substance, not engagement baiting. I encourage others to block accounts that rely on logical fallacies and provocations to undermine experts like Dr. Bickman, who provide actual value.
Genuinely interested: what do you love and what's blocking you from it?
A majority of people need to have an ecstatic experience with SI before a power grab over SI happens.
I'm seeing a mass infection of cognitive distortions (catastrophizing, fortune-telling, absolutistic thinking) warping public opinion.
Share what you love, and with a little luck the right person will notice how an SI implementation can foster more of it.
@moonshots_pod is a joy. I enjoy critical thinking and learn so much from the commentary.
I don't watch legacy media because it feels like emotional manipulation. When every story has a spin, a grab, irrational fear, and villains.
I would be missing out on the real historic news happening every week, if it wasn't for @PeterDiamandis
Job title: Official Dem Outrage Actress (AOC)
Job Requirements: Must permanently suffer from cognitive distortions, or be a psychopath capable of deploying them.
The act: Deliver pre-written class-warfare monologues to emotionally manipulate voters into identifying with Democrats.
Function: To seal the voter base by triggering identity protection: you lose your moral identity if you disagree.
Fallacies on cue: Argumentum ad hominem/odiom, guilt by association, straw man, band wagon effect, appeal to ignorance.
Emotional triggers: Tribal rage, envy-fueled hatred, moral-superiority and any dopamine hit so strong that facts will be ignored.
Legitimate power is terrifying. Legitimate means someone has power because of a title they were elected to. It's terrifying because to be elected one has to perform well in a system of pursuation.
The power Elon has is contractual. He must perform based on the invested contract and he's graded not on his ability to pursuade but his ability to problem solve.
Political power is terrifying compared to power that contractually requires strong critical thinking and problem solving.
@elonmusk Ultimately, engineers only succeed by learning from failures, while politicians survive by harvesting it for pursuation.
A failed project allows the next political cycle to stand in the rubble and say vote for me because look at my opponents' failure.