Steven

[Tsla Chan @Tslachan]

$TSLA
일론 양 옆에 옵티머스ㅎㅎ

[Tesla @Tesla]

https://t.co/DQsC4h8QvL

[Daniel Romero @HyperTechInvest]

REPORTER: "Is there a day when $AMD will be bigger than $NVDA?

LISA SU: "We'll see how that plays out"

[Dave Lee @heydave7]

Tesla's FSD v12 for the first time will use an "end-to-end" neural network approach to autonomous driving. This is NOT just an incremental update; it's a paradigm shift. Let's explain.

End-to-end Neural Networks for Autonomous Driving
In an end-to-end system, the input (primarily video data, in this case) goes straight through a neural network, and the output is the final action (like steering, braking, etc.). There's no middle-man algorithm to convert sensor data into some intermediate form. And there's not hundreds of thousands of lines of heuristic code to try to determine how the car should act in certain situations. v12 processes and makes driving decisions much like how our brains work.

So, what's the big deal?
Currently Tesla's FSD v11 has separate parts or "modules" to handle specific jobs like identifying objects, planning the route, and actually steering the car. Each of these parts needs its own specialized training and heaps of specific data to work correctly.

End-to-end learning, however, bypasses this by training a single, unified model to map raw sensor inputs directly to control outputs. This holistic training allows the system to capture more nuanced relationships and dependencies in the data, facilitating better generalization.

One of the greatest benefits to end-to-end is that FSD should start to exhibit zero-shot learning. Zero-shot learning is a machine learning concept where a model is designed to handle tasks it has never seen before, without requiring any additional training. Instead of needing new data and training for each new scenario, a zero-shot model uses its existing knowledge to make educated decisions. This is a big deal.

Here's some specifics of how FSD will likely see zero-shot learning benefits.

Rapid Adaptation to New Locations: If the system is trained in a specific geographic location with unique road signs, road conditions, and traffic behavior, zero-shot learning would allow it to adapt to new cities or countries without having to be retrained on local data.

Handling Edge Cases: Even an exceedingly well-performing system can encounter scenarios it has never seen before, such as complex accident scenes or unusual objects on the road. Zero-shot learning would enable the system to make informed decisions in these unfamiliar situations based on its ability to generalize from its training data.

Dynamic Traffic Conditions: Zero-shot learning would help the system adapt to traffic scenarios it hasn't been trained on, such as rare events like marathons, parades, or emergency evacuation scenarios, thereby ensuring safety and optimal route selection.

Human Interaction: Zero-shot learning could improve the vehicle's ability to understand and adapt to unpredictable human behavior, be it from drivers, pedestrians, or cyclists. This could be particularly useful in scenarios where human behavior deviates from the norm, like jaywalking or erratic driving.

Seasonal Weather: Conditions like snow, fall leaves, or rain can significantly alter the driving environment. A zero-shot learning-equipped system would be better at adapting to these changes in real-time, ensuring safer driving year-round.

Faster Deployment: Since the system would be better at generalizing from existing data, it could be deployed in new environments or updated to meet new regulations more quickly, providing the company with a competitive advantage.

Multi-modal Integration: If Tesla ever chose to add more sensors to their FSD hardware suite, zero-shot learning could facilitate seamless integration of these sensors even if the system was not originally trained on data from these sensors.

Overall, v12 is not simply just another FSD update. Rather, it's the start of a new paradigm in autonomous driving, one that likely change the future of transport forever. The end-to-end neural network will allow Tesla to improve FSD much quicker, scale much faster, handle edge cases much safer, and navigate complex traffic conditions with unprecedented efficiency.

So, who's excited about FSD v12?

[Steven @Nevetzz]

@OishiruKuda 日子还是一样过。

[Elon Musk @elonmusk]

Our headquarters tonight

[Mahmoud🔶foqha  @mahmoudfoqha9]

@elonmusk @TitterDaily

[Sergio Pereira @SergioRocks]

Why I don't use Scrum to manage my Remote Teams?

TL;DR: It adds at least 8 hours of meetings per Sprint. That's 2 full days of time wasted, per team member, per month!

This is what I do instead:

[. @10xEngineered]

Depends on the industry. In software, the teams I manage have been significantly more productive working from home, working on their schedules, etc.

Measure productivity from output (regardless of where they are) instead of hours sitting at the desk.

In the office, people are constantly in pointless meetings or having random discussions that aren't productive. Then sitting in traffic for 2 hours + per day commuting.

[Astro.eth 🪐 @AstroCrypto89]

YOU'RE EARLY ‍💨

always missing out? trust me, don't miss out on $pepe
👉 https://t.co/w9BPECmsAw

#pepe #web3 #pepecoin #nfts $snek #solana #bitcoin $jeff $ribbit $lambo #memecoin #memecoins #memecoinseason $eth #bitcoin $wojak $mong #moong #aidoge $king $dino #floki #binance