Using ChatGPT to write comments/posts about dive equipment

[L13]

For example, despite the name, Tesla's Full Self Driving is actually SAE Level 2. The car isn't driving which means Tesla accepts no responsibility for crashes while it is engaged.

I think any fair reading of that chart and actual Tesla performance, it is at least SAE Level 2.9 if not marginally SAE Level 3. It easily accomplishes the "Example Feature" given in your chart (traffic jam chauffeur).

There are no vehicles capable of Level 5, i.e., just taking to the road under any conditions and going anywhere. It's not even clear that we will ever be able to reach this goal without unforseen technological advances.

Given even humans can't achieve SAE Level 5 performance, I agree.

 

[lowwall]

I think any fair reading of that chart and actual Tesla performance, it is at least SAE Level 2.9 if not marginally SAE Level 3. It easily accomplishes the "Example Feature" given in your chart (traffic jam chauffeur)

It's nowhere near 3. Level 2 requires the human driver to constantly monitor what is going on and be ready to intercede with no notice. Level 3 allows the human to not pay attention, but to be able to take over when the vehicle when notified by the vehicle.

Tesla has officially notified California that FSD is Level 2 and the car warns the driver that he or she must pay attention and be ready to intervene at any time

Tesla Tells California DMV that FSD Is Not Capable of Autonomous Driving

After years of touting its long-awaited Full Self-Driving feature, Tesla is telling California regulators a different story about its capabilities.

www.caranddriver.com

Tesla FSD Is Now $15K, but Is It Any Closer to Level 3?

Tesla's so-called Full Self-Driving feature is now $15,000, up from $12,000, but is the system actually meant to evolve from Level 2 to Level 3?

www.autoweek.com

The only Level 3 vehicle that you can actually buy is a Mercedes S-class, which allows autonomous driving on a limited number of highways in Germany, California and Nevada and only during stop and go traffic as it shuts off at 60km/h (36mph).

 

[L13]

It's nowhere near 3. Level 2 requires the human driver to constantly monitor what is going on and be ready to intercede with no notice. Level 3 allows the human to not pay attention, but to be able to take over when the vehicle when notified by the vehicle.

That is just the lawyers talking.

Test drive a Tesla, and compare it to the contents of your chart. It easily exceeds level 2. It marginally meets level 3, with the primary deficiency being Tesla lawyers not being willing to claim liability for it. There is no question that it can "drive the vehicle under limited conditions, and will not operate unless all conditions are met". Almost all of the reported deficiencies with the latest versions are the result of attempting to use it at the margins of its current capability envelope.

Tesla has officially notified California that FSD is Level 2 and the car warns the driver that he or she must pay attention and be ready to intervene at any time

Primarily because otherwise the drivers do not meet the requirement for level 3 "When the feature requests, you must drive". Tesla's lawyers are being defensive. When the "Automated Driving Features" are used as intended/documented (admittedly, not as marketed), they meet the definitions in your chart for level 3.

 

[kimh]

Do you realise that the text about the David Shaw accident basically gets everything wrong?

You missed the 2nd sentence of my post:

I just asked ChatGPT about Dave Shaw's fatal dive in Bushman's Hole. It just invented its own story.

Lingee was good for english to german but horrible for german into english for machine repair manual type stuff and other technical documents.

Long ago I did occasional translation work, but more recently it's just personal use. I generally use Linguee for cross-checking the occasional usage example between English & German. For me, Linguee's side-by-side columns work great for that.

I don't really use the tools for direct translation. More of a check for dumb mistakes, or alternative vocabulary/idioms.

 

[ginti]

I use DeepL for French, too- can you describe what you mean there? Familiar vs. formal is of course also an issue in German.

"what will you do tomorrow?"

Both Google and DeepL will translate as "que est-ce que vous allez faire demain ?" Or something similar, using the "vous". This translation is the right one in three cases:
- asking the question to one person in very formal context
- when speaking with a person you don't know
- when speaking with multiple people

But if you ask it to one of your friends, the translation is different. DeepL has a feature to modify the first translation. On the contrary, I don't manage to modify it on Google (I am not aware of a similar feature - if it has it, please let me know!)

 

[lowwall]

That is just the lawyers talking.

Test drive a Tesla, and compare it to the contents of your chart. It easily exceeds level 2. It marginally meets level 3, with the primary deficiency being Tesla lawyers not being willing to claim liability for it. There is no question that it can "drive the vehicle under limited conditions, and will not operate unless all conditions are met". Almost all of the reported deficiencies with the latest versions are the result of attempting to use it at the margins of its current capability envelope.


Primarily because otherwise the drivers do not meet the requirement for level 3 "When the feature requests, you must drive". Tesla's lawyers are being defensive. When the "Automated Driving Features" are used as intended/documented (admittedly, not as marketed), they meet the definitions in your chart for level 3.

Level 3 means the automaker takes legal responsibility for the driving. Tesla's engineers have determined that FSD is not safe enough for them to take on that responsibility under any set of real world conditions, no matter how tightly controlled. Tesla's lawyers are the ones tasked with explaining that reality. When the legal department contradicts the marketing department, you should always assume the former are the ones telling the truth.

And I don't need to drive it. Even if I wasn't willing to take Tesla's own word that FSD is only level 2, I've seen the videos and read/listened to the reviews. Intervention without warning is always necessary at some point. It doesn't matter if it's once in 100 hours (and it's nowhere near that good currently), if you have to monitor what the car is doing, it is not Level 3.

BTW, there are some Level 4 test vehicles out there. These are fully autonomous within a controlled set of circumstances. They all have in common that they only operate in a limited geographic area that has been fully mapped by the vehicle provider. Fully mapped goes far beyond the standard concept of mapping, for example the vehicles are programmed to know the boundaries of every lane and street edge and the location and rules for every traffic signal and sign. And they are typically limited to low speed areas and times of good weather and visibility.

 

[kimh]

Quote of the day:

"No one cares if LLMs aren’t AGI, as long as they do the job."​

​

@Hello_World​

(LLMs: ChatGPT etc.)

 

[berndo]

You missed the 2nd sentence of my post:

So what's your point? You're calling something AI that recognizes context and than goes to jumble up random text. People claim it's doing reseach but is doesn't. The openai people claim they build a reasoning engine. The language tool looks for what's most likely the next word should be in a given context.
Too me, it sounds like you're buying into marketing BS. AFAIK, Musk is involved in openai in some way... he's been claiming to have a full self driving and robotaxis next year... for how many years now?

I generally use Linguee for cross-checking the occasional usage example between English & German. For me, Linguee's side-by-side columns work great for that.

I don't really use the tools for direct translation. More of a check for dumb mistakes, or alternative vocabulary/idioms.

I wouldn't to that. Lingee's all over qualitiy is really bad unless you looking for very basic stuff. Especially for checking vocabulary I wouldn't use it for the reasons I mentioned in an earlier post. Something like dict.cc is way better for checking vocabulary.

 

[Angelo Farina]

FWIW, unless we are talking about a brain-in-a-box generalized AI, I don't think the term AI is useful in the real world. What is useful is defining your task and then setting parameters for levels of success.

A good example of this are the SAE autonomous driving levels:

This approach lets you get past the marketing puffery. For example, despite the name, Tesla's Full Self Driving is actually SAE Level 2. The car isn't driving which means Tesla accepts no responsibility for crashes while it is engaged.

As an aside, the furthest anyone had gotten is SAE Level 4. There are no vehicles capable of Level 5, i.e., just taking to the road under any conditions and going anywhere. It's not even clear that we will ever be able to reach this goal without unforseen technological advances.

Please note that our local spinoff company Vislab reached level 3 in 2013 and is currently way beyond level 4.
They have now an experimental car which is targetting level 5. This should be reached in a few months, possibly this summer.
More info here: VisLab | Extend Your Vision – Extend Your Vision

 

[kimh]

So what's your point?

…you asked me if I realized ChatGPT got the Dave Shaw story wrong, but I'd already pointed that out at the start.

You're calling something AI that recognizes context and than goes to jumble up random text.

Did you really read anything I wrote? I never described anything as "AI" — without immediately qualifying that term. In fact, I spent half the thread describing how vague "AI" is as a concept & criticizing its oversimplification.

I wouldn't [use Linguee for cross-checking]. Lingee's all over qualitiy is really bad unless you looking for very basic stuff. Especially for checking vocabulary I wouldn't use it for the reasons I mentioned in an earlier post. Something like dict.cc is way better for checking vocabulary.

Apparently you don't understand what Linguee does. You compared it directly to dict.cc, but they're totally different tools, for different purposes. I use both.

Linguee is neither translator nor dictionary. It's a database of parallel text segments, in two languages that have already been translated elsewhere e.g. on the web. It only provides these potential examples, which is why I use it for cross-checking, ideas etc. Some sample pairs are relevant, and many are not.

Maybe you use these tools differently. My German is already quite fluent, and I'm not looking to just translate "A" to "B." It's more about double-checking grammar details and seeing potential ideas/idioms. The writing is always mine in the end. Btw- are you German? Your profile is blocked, so I can only guess.