Lucid assistant is lagging further and further behind

The silver lining is that those with actual expertise and experience will be in demand because someone's gonna have to clean up the mess AI leaves behind.

And smack some sense back into those who drank the AI kool aid

Now that's job security 😁
I'm thankful that my engineering work hasn't really been affected by AI the way we see in so much of the tech industry. I do a lot of multi-disciplinary research-style work. But as the technology continues to improve, it's just a matter of time. The junior engineers I work with are already seeing big upheaval in what their work life looks like. It comes with productivity improvements, but I have yet to meet someone who enjoys cleaning up after an AI. For now my frustrations remain concentrated on the corporate insanity that comes from human leadership. 🤣
 
I trust the devs know what to look for and how to instruct the agent, it'll just speed them up. Our own members here can vibe code fully functional Lucid ios watch apps with half baked python apis within minutes or hours. Surely would be also helpful to a resource-stretched dev team.
Depends on the severity and materiality of the scope of the work. Especially when things go wrong.

For an ios watch app failing, worst case scenario the user gets inconvenienced.

An ADAS system failing? Someone could get killed. Which means this would require someone to be as hands-on as possible on the software to ensure the system is robust and safe.
 
Last edited:
I'm thankful that my engineering work hasn't really been affected by AI the way we see in so much of the tech industry. I do a lot of multi-disciplinary research-style work. But as the technology continues to improve, it's just a matter of time. The junior engineers I work with are already seeing big upheaval in what their work life looks like. It comes with productivity improvements, but I have yet to meet someone who enjoys cleaning up after an AI. For now my frustrations remain concentrated on the corporate insanity that comes from human leadership. 🤣
Trust me, it's bad. Really bad.

I'm a data scientist so unfortunately I'm right in the thick of it. Before ChatGPT I would be neck-deep into learning the material and understanding the physical problem, the data, and the modeling from the inside out.

Ever since ChatGPT, it's been the biggest gift to the most infuriating "I'm-an-ideas-guy" middle manager. Now they can give the perception or the appearance of competence without actually doing any of the work or having any understanding. By the time it all falls apart they've thrown their hands up in the air and moved on to the next thing.

Now, I'm drowning in fixing AI slop. Even worse, I feel as though I'm watching peoples' critical thinking skills melt and dissolve in front of my eyes and it's sad.
 
Trust me, it's bad. Really bad.

I'm a data scientist so unfortunately I'm right in the thick of it. Before ChatGPT I would be neck-deep into learning the material and understanding the physical problem, the data, and the modeling from the inside out.

Ever since ChatGPT, it's been the biggest gift to the most infuriating "I'm-an-ideas-guy" middle manager. Now they can give the perception or the appearance of competence without actually doing any of the work or having any understanding. By the time it all falls apart they've thrown their hands up in the air and moved on to the next thing.

Now, I'm drowning in fixing AI slop. Even worse, I feel as though I'm watching peoples' critical thinking skills melt and dissolve in front of my eyes and it's sad.
Oh dear, I had no clue it was going so fast.

I don’t know what the future brings if people don’t learn how to just look at the data.

I found lots of interesting things just scrolling through the data
 
Oh dear, I had no clue it was going so fast.

I don’t know what the future brings if people don’t learn how to just look at the data.

I found lots of interesting things just scrolling through the data
Well, there's looking at data, and then there's looking at data. There's always a big population of people that can take a naive but sane look at data. The number of people who really understand statistics is much smaller, and most people don't appreciate how much statistics has changed over the last three decades. When you start talking about teasing patterns out of multi-dimensional data sets, it can be hard to even describe the techniques used to those who have no background in statistics. And as the techniques grow more sophisticated, so too does the judgement needed to know how to properly apply them. If you spend enough time with a spreadsheet, you can get it to tell you anything you want to hear. Add in the modern proliferation of dirty datasets, and the problem gets even worse. I was just reading a news article this morning about someone getting screwed over by Flock's shoddy data handling practices, which they can't figure out how to do a good job on even though it's theoretically core to their business.
 
Wrote several machine language programs for the HP-41CV/CX calculator. Ever produced a couple of ROM modules that sold reasonably well (including printing barcode on printers directly from the calculator to printers at 5X to 10X the speed which HP was doing.
 
All these forums are big circlejerk parties where everyone upvotes positive comments.

If you read through the tesla forums it is enlightening because it's full of people who actually care about the products and whether they are good or not.

You see plenty of detractors on there and it is welcome because it provides a different view. Then you find people who actually care about the tech in the cars, battery degradation, best charging practices, etc.

It's also important to not be ignorant about what others have because it shows what is possible.

Good example of this is how on every forum people talk about more sensors, more compute, etc. leads to better driving and surpassing tesla in record time when there is zero evidence of this. There's a reason mercedes drive pilot has not release their P2P system in the U.S. this year. There's a reason that the chinese systems have a safety level of FSD 12. There's a reason that tesla increased memory bandwidth of AI5 to have 4-5x what is found in nvidia thor.

There was a guy on that tesla forum who used to talk about how FSD had no moat and that all these systems from nvidia, wayve, chinese, etc. were on the same level. It became increasingly clear they are nowhere close and "mr. always right" basically left the forum.

What also became clear over the past year is that tesla's penny pinching and using the least possible cameras to cover the largest FOV possible has a side benefit in that it allows them to use very low memory bandwidth and have very fast reaction times.
Hogwash! Only reason Musk got rid of other sensors is because it would take much longer and more difficult using more compute. Don’t believe everything that comes out of his mouth. He always cuts corners and does not put safety first.
 
Hogwash! Only reason Musk got rid of other sensors is because it would take much longer and more difficult using more compute. Don’t believe everything that comes out of his mouth. He always cuts corners and does not put safety first.

This is 100% false. The only sensor tesla had was cruise control radar. Cruise control radar is the same radar found in almost every car in the U.S.

Those are the same cars that will willingly hit stationary objects at 80mph because cruise control radar doesn't work for self driving

When tesla actually had accidents with autopilot where they'd hit stationary vehicles it was because they were using radar for depth detection
 
This is 100% false. The only sensor tesla had was cruise control radar. Cruise control radar is the same radar found in almost every car in the U.S.

Those are the same cars that will willingly hit stationary objects at 80mph because cruise control radar doesn't work for self driving

When tesla actually had accidents with autopilot where they'd hit stationary vehicles it was because they were using radar for depth detection
The use of radar has nothing to do with vehicles running into stationary objects. Radar returns a fairly low resolution of distance to objects and better radars also return the speed of those objects. Cameras return a higher resolution picture with no distance or speed information. Multiple cameras can determine distance to objects and multiple pictures with multiple cameras can determine speed of an object. The issue with stationary objects is compute power and the number of objects/targets tracked, not radar or cameras. When an ADAS system cannot track all objects/targets it begins to ignore some objects/targets. Moving objects are a threat so they are not ignored, many stationary objects are not a threat so many of them get ignored. This is why ADAS systems can run into stationary objects.

Cameras do have an advantage over radar at target identification that can help decide if a target should be tracked or ignored. However, radar has an advantage of seeing objects in any lighting and through a lot of weather. Hence, multiple types of sensors will give you the best target identification and tracking. All types of sensors have their advantages and disadvantages. Your comments regarding radar indicate a fundamental lack of understanding of radar systems and ADAS sensors.
 
The use of radar has nothing to do with vehicles running into stationary objects. Radar returns a fairly low resolution of distance to objects and better radars also return the speed of those objects. Cameras return a higher resolution picture with no distance or speed information. Multiple cameras can determine distance to objects and multiple pictures with multiple cameras can determine speed of an object. The issue with stationary objects is compute power and the number of objects/targets tracked, not radar or cameras. When an ADAS system cannot track all objects/targets it begins to ignore some objects/targets. Moving objects are a threat so they are not ignored, many stationary objects are not a threat so many of them get ignored. This is why ADAS systems can run into stationary objects.

That's not correct. Cruise control radar has no vertical resolution so they turn of detection of stationary objects to reduce phantom braking

The reason tesla engineers wanted to keep the radar is because it is useful for data collection

the reason moving objects are detected is because it's easy to detect a moving object
 
That's not correct. Cruise control radar has no vertical resolution so they turn of detection of stationary objects to reduce phantom braking
Typical low cost automotive radar has 15 degrees of vertical resolution. Radar sees both stationary objects and moving objects, neither is easier or harder to see. I wish you would speak with facts rather than You Tube propaganda.
 
Typical low cost automotive radar has 15 degrees of vertical resolution. Radar sees both stationary objects and moving objects, neither is easier or harder to see. I wish you would speak with facts rather than You Tube propaganda.
Yes but you can't brake for stationary objects without falsely braking for objects that are above you

Radar measures objects using doppler shift which makes it easier to detect objects moving at a similar speed

Therefore in this case it is easier to detect objects moving at a similar speed to you
 
Yes but you can't brake for stationary objects without falsely braking for objects that are above you

Radar measures objects using doppler shift which makes it easier to detect objects moving at a similar speed

Therefore in this case it is easier to detect objects moving at a similar speed to you
Radar measures both time of flight and frequency shift. It is neither easier or harder to detect stationary objects versus moving objects.
 
Radar measures both time of flight and frequency shift. It is neither easier or harder to detect stationary objects versus moving objects.
you can't detect stationary objects without too many false positives. But for data collection purpose radar is useful because you know the location and speed of the objects from video and therefore it is easier to link it to radar returns
 
Typical low cost automotive radar has 15 degrees of vertical resolution. Radar sees both stationary objects and moving objects, neither is easier or harder to see. I wish you would speak with facts rather than You Tube propaganda.
You didn’t get the memo? YouTube is the source of truth these days. Influencers know more than the engineers that design these vehicles and systems.

Add a Tesla fanboy into the mix and they know so much about FSD and what other automakers should be doing that it’s astonishing that they’re not designing their own ADAS suite of software and hardware to sell to these automakers who seem to have no clue what they’re doing. At least according to a FSD fanboy!
 
You didn’t get the memo? YouTube is the source of truth these days. Influencers know more than the engineers that design these vehicles and systems.

Add a Tesla fanboy into the mix and they know so much about FSD and what other automakers should be doing that it’s astonishing that they’re not designing their own ADAS suite of software and hardware to sell to these automakers who seem to have no clue what they’re doing. At least according to a FSD fanboy!
Well you are chatting with MOre BAd DAta
 
You didn’t get the memo? YouTube is the source of truth these days. Influencers know more than the engineers that design these vehicles and systems.

Add a Tesla fanboy into the mix and they know so much about FSD and what other automakers should be doing that it’s astonishing that they’re not designing their own ADAS suite of software and hardware to sell to these automakers who seem to have no clue what they’re doing. At least according to a FSD fanboy!
What's sad is everything I'm saying is not new. This was never new. All of these limitations were known 15 years ago before tesla was using camera only, before tesla even started developing a self driving system. Nothing has changed.
 
Well you are chatting with MOre BAd DAta
punchout-tyson.gif

@Aka-none by first round KO 😅
 
You didn’t get the memo? YouTube is the source of truth these days. Influencers know more than the engineers that design these vehicles and systems.

Add a Tesla fanboy into the mix and they know so much about FSD and what other automakers should be doing that it’s astonishing that they’re not designing their own ADAS suite of software and hardware to sell to these automakers who seem to have no clue what they’re doing. At least according to a FSD fanboy!
Youtube the source of truth? That's old stuff!

ChatGLP

ChatBBQ

ChatBobaTea

ChatGPT said I was right!

I mean sure ChatGPT disagreed with me at first but after enough prompting the AI agreed with me and called me brilliant!
 
Back
Top