Pouring Resources Into AI — Does It End Well? Will Diminishing Returns On FSD Sink Tesla?

Join daily news updates from CleanTechnica on e mail. Or follow us on Google News!

I wrote an article the opposite day titled “Tesla Proved Many People Wrong For Years — Is This Time The Same, Or Different?” The core factors of the article have been these:

1. Many individuals stated Tesla couldn’t do what it ended up doing at many levels of the corporate’s improvement.
2. I believed Tesla might do these issues.
3. Now, I’m having a tough time believing in Tesla’s strategy to robotaxis and its plan to roll them out inside a couple of years (although I argued the other a decade in the past, and even as much as 5 years in the past).
4. So, I’m questioning, have I turn into the skeptic who might be confirmed incorrect, or am I proper to suppose it’s totally different this time and Tesla received’t obtain its robotaxi targets (once more)?

Down within the feedback, a reader, Doug Sanden, wrote the next:

“FSD coaching — log linear?

• • growing quantity of information wanted for linear progress, or reducing progress for linear increses in information:
    video ‘Has Generative AI Already Peaked?’ – Computerphile
  • diminishing returns?
  • he’s describing the outcomes from this paper:
  • ‘No “Zero-Shot” With out Exponential Information: Pretraining Idea Frequency Determines Multimodal Mannequin Efficiency’, Udandarao et al, 2024
  • inefficient log-linear scaling development
  • exponential want for coaching information”

I seemed up that video and watched it. Right here it’s, and I extremely advocate giving it a watch:

Once more, don’t skip the video, however I’ll attempt to summarize the important thing factors once more right here:

• Generative AI is superb and doing nice issues.
• Due to what it may possibly do, it looks like magic to many and in addition provides folks the idea that including increasingly information, or constructing greater and larger fashions — or doing each — will make determining nearly something attainable and can allow increasingly superb technological options.
• Nevertheless, in easy phrases, there are diminishing returns when the specified output or desired product will get extra difficult, extra nuanced.
• Additionally, as you try to attain these way more troublesome targets, you find yourself pouring an unlimited amount of cash into information assortment and mannequin processing, and you’ll find yourself down a deep cash pit primarily chasing unicorns and butterfly farts.

Because it pertains to Tesla as an organization and Tesla Full Self Driving (FSD) as a product, that is mainly only a higher clarification of one thing I’ve been involved about for some time. I do suppose Tesla dangers chasing unicorns diminishing returns with its strategy to FSD, which I believe might burn an unlimited amount of money with out Tesla with the ability to roll out robotaxis. I additionally suppose this pertains to the “seesaw problem” I’ve mentioned a couple of occasions — that’s simply the impact of making an attempt to resolve very nuanced, difficult issues with the {hardware} and software program strategy Tesla is utilizing.

Possibly I’m incorrect, as I famous in that article linked on the high. However I find yourself similar to this man on the finish of his video, the place he signifies that he expects generative AI to comply with the crimson line on the graph moderately than the yellow one — however that it could be thrilling if it did find yourself following the yellow one:

We’ll see. We’re nonetheless far-off from Tesla robotaxis, however let’s see how the tech progresses.