Ben Snyder

Today in shitty machine learning startups, this company claims to predict IQ, personality, and violent tendencies by applying deep learning to facial features and bone structure. That's phrenology. You just made the ML equivalent of a racist uncle.

@mason_compsci @__tinygrad__ 8b will fit on a single 4090, so probably pretty good inference. Training though, would be terrible having to shard across 6 GPUs without NVLink. That's likely why they're showcasing a tiny model like Resnet.

@__tinygrad__ How many token per second does this thing crank out running a small model like Llama3-8b

Every tinybox trains a ResNet-50 before it leaves the factory. Screen software is system level so it works with tinygrad, PyTorch, and everything else!

@TSLAFanMtl Remember when Mobileye's CEO (the guy with the doctorate in AI who has been studying the problem for almost 40 years) said exactly this would happen, that Tesla's brute force approach would hit a ceiling? Maybe the experts actually know what they're talking about.

My long-time followers will know that I have been bearish on Tesla's FSD ambitions for years. This post is from May 2020, when I was all-in 100% $TSLA stock and have every bias in the world to believe Elon's bull 💩on the FSD timeline. The data did not support it, despite Elon's assurances and despite the noisy FSD hyperbulls busting a nut after every increasingly less terrible (but still terrible) iteration. How did I know? How could a cat who still doesn't use FSD daily or even drive a Tesla be right 4 years later? Answer = I don't have my head up my ass, unlike some people on this app. Since this time, FSD has improved a lot; it is now a functional, useable and valuable driver-assistance software. But the gap between driver-assistance software and unsupervised FSD is huge. Tesla needs to go from 1 intervention every 250-500 miles to 1 intervention every 17,000 miles. The pace of progress is still too slow, and until proven otherwise, I would be weary of watching 15-20 minute intervention-free drives and believing the hype.

What a piece of shit.

@editingemily I still remember elementary school, how kids would try to whisper to the cashier in the lunch line that they're on the free lunch program.

One of the reasons I believe ALL KIDS should receive lunch free of charge is we don’t talk about how brutal it is to grow up poor and have everyone else know your family relies on public assistance. Embarrassing doesn’t even cover it. I wish that experience on no one.

@wholemars Sure. Give me a date when you expect Tesla to deliver actual attention off no liability driving.

So let me get this straight — your thesis is that the same companies that could never catch Tesla in EVs are going to somehow beat them in self-driving? I'm happy to take the other side of that bet. $TSLA

@ericmossman_ I have, but you didn’t understand them, and so just ignored them.

@jbensnyder Would love for you represent some of those in your arguments. All you have is credentials, evidence evades you

“$28 trillion” That’s more than the entire US GDP. How are these chucklefuck scammers not in prison?

@ericmossman_ Yeah, and do you not see the problem with that assumption over time? Also, what’s your AI experience? That might explain why you’re struggling with the math.

@jbensnyder That expresses a variable growing or regressing exponentially…

@ericmossman_ So again, what’s your AI experience?

@jbensnyder 40% growth per annum, linear? 🤔

@ericmossman_ And he bounces back, assuming linearity. What’s your experience with Ai?

@jbensnyder 35-40% would get you there. Tesla averaged 32% growth in earnings over the past 5 years

@ericmossman_ “Insinuating” So I never said that, you imagined it. What’s your experience with AI?

@jbensnyder insinuating that the number we were talking about was revenue. It’s not, it’s a market cap based anticipated future earnings. I never flipped, you didn’t read again. I said that with 50-70% margins they could grow to get to $100b in earnings WITHOUT needing a 50% growth rate.

@ericmossman_ Yeah, I compared to GDP as a measure of the addressable market.

@jbensnyder “28T. That’s more than the entire US GDP.” You only talked about TAM when everyone called you out. And yeah the TAM is big AF. The $8T is Ark’s predicted market cap for 2029, you then said “that would be 10% of the entire economy”

@ericmossman_ Again, you did the gish gallop through terms you didn’t understand. So what’s your technical experience with AI?

@jbensnyder Based on what? You couldn’t even decern between margin and growth rate. Technobabble, you’ve literally only used buzz words in your defenses and not demonstrated a lick of technical knowledge. You think that legacy OEMs have the most advanced tech, what are we talking about?

@ericmossman_ Well no. I never said it was market cap or revenue. I never even referred to the 8 trillion amount. You read all that in yourself. And you flipped from growth rate to margins. What’s your technical experience with AI?

@jbensnyder You thought $28T was the market cap, i then explained to you it was TAM. You acted like you knew that, then thought $8T was Ark’s revenue prediction, I showed you that was market cap. I said software typically had 50%+ margins, you thought that was growth rate. Twist and turn

@ericmossman_ Well no, I'm deeply embedded in the fact that I actually know what the fuck I'm talking about, while you're deeply embedded in technobabble you don't understand.

@jbensnyder I can tell that you’re deeply embedded within the biases that you harbored from your own hubris. This is not over my head and I don’t doubt that you know a lot but to assume I don’t because we disagree shows the depth of your coherence.

@ericmossman_ I’m telling you selling a promise of “next year” isn’t the same as actually developing the tech. But, being that you pretty clearly only know buzzwords, the tech is way over your head.

@jbensnyder Wait, you’re telling me they need to spend money and man hours on developing a new technology? That’s crazy, I thought it was gonna be perfect right off the bat, you got me

@ericmossman_ While not being at all autonomous.

@jbensnyder And yet somehow FSD is currently pulling in $1-3B in revenue and mobile eye has a total MARKET CAP equal to Tesla EARNINGS per year. Why isn’t mobile eye dominating as the most valuable company in the world? They’re the key to autonomous driving right?

@ericmossman_ In your fantasy world that doesn't understand how AI system actually work.

@jbensnyder You don’t(can’t) read, 50% MARGINS. Tesla could get to $100b earnings with less than 50% growth per annum

@ericmossman_ Technical details of the chips, not an appeal to authority.

@jbensnyder Do you own a Tesla? Have you tried FSD? Autonomous driving in any other vehicle is laughably bad. These guys are still using expensive lidar systems which have been dropped by many companies. It’s too expensive, not scalable. The people designing Tesla’s chips are former apple