Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

[jerry33]

I did my duty and sold another model 3. Ppl have huge misconceptions about m3 pricing. This time I didn't give out my referral code. I felt it'd taint the experience.

Cool. That's why I've never given out a referral code.

 

[ZeApelido]

Saw 2 more carriers full of Model 3's heading to SF pier this morning.

 

[Tenable]

I haven't been able to follow this thread closely since last week and am now seeing posts about a capital raise -- is there reason to expect one imminently, or are folks just speculating?

EM May have tweeted something about seeking alternative funding avenues. But nothing else I know of.

 

[Tslynk67]

Interresting, so Ford with Rivian, GM on its own. I guess they are waiting for Tesla's Pickup unveil to see if they will have a compelling design..

So here's a cool scenario... Pickup reveal day, Elon drives out onto the stage in the most ridiculous looking thing you've ever seen - as he said, steam-punk meets Bladerunner - titanium, FWD's, motor for each wheel, 400kWh battery pack, spiky things all over, etc. He runs through some outrageous power, range, acceleration and speed specs and then reveals a price-tag that's 2x more than the Roadster. Special order, hand-built, limited volumes, collectors piece essentially. That's his vanity project.

He then throws in a "one more thing" and Franz drives out in the mass-market pickup, with, which nobody expects to see at this moment. It's a killer, well specced and well priced. The crowd goes nuts, the SP goes through the roof, even the MSM struggle to pick faults, Toilet-Boy's head actually explodes, like in Scanners.

This would be something.

 

[ReflexFunds]

Disengagements create only a brief report - no usable training data (e.g. no imagery or whatnot), just a couple kilobytes of "situational" data. They're designed to help Tesla figure out what sort of situations are the most common and problematic reasons for disengagement. Then Tesla launches a campaign with "triggers" (which can be very complex) to gather potential training data for dealing with their most problematic situations. The training data is filtered down to examples that illustrate nuances - including things that should be both positive and negative detections - rather than being included as a flood, which can overreinforce "normal" situations and lead to the exclusion of edge cases. These are manually labeled and then added to the dataset.

Source: a combination of Green's analysis and the Tesla autonomy presentation.

This is my rough guess at how Tesla's Autopilot development process works for each new driving capability. Anyone agree/disagree? @jimmy_d @Fact Checking

1. Tesla decides on a new driving feature. For example Navigate on Autopilot.
2. Tesla writes code to execute the driving policy (may be a mix of heuristics and prediction NNs and in future also imitation learning NNs etc).
3. They test this code on central servers against a standardised test set of different driving scenarios (10k/100k scenarios?). The test may compare the computer's chosen actions (using saved real world sensor data for the encountered scenario) against actual human responses collected from the fleet (here the sensor data is fixed so it can only optimise towards matching the human action). It can also be tested in simulation (here they can also see the consequences of actions that differ from the saved human response).
4. They use this test performance to improve the driving policy, possibly both manually and using reinforcement learning. Repeat until a fixed error rate threshold is reached.
5. The feature is rolled out in shadow mode to a % of cars.
6. Tesla receives a summary of high level data when Shadow Mode executes a manoeuvre. They compare Shadow mode’s planned actions vs human’s actual actions. This is used to create a priority list of situations where the two most often differ significantly.
7. Tesla tells a % of the fleet to collect detailed NN information when they encounter the situation at the top of the priority list. For example bikes on cars. They train a basic NN to recognise these pictures as a trigger to start collecting data.
8. Tesla start getting data back, they annotate this and feed it back to the recognition neural net. This makes the cars better at detecting the situation they are targeting, so the quality of data sent back by the fleet improves.
9. Once they have collected enough good examples (0.1k, 1k, 10k, 100k?), they annotate them, add to the main vision neural net data set and retrain. They also may update the driving policy, either through NN training, or through manual code to correct the mistakes after reviewing the videos.
10. Repeat steps 3-9 until they have worked through their list of priority problems to the point where a certain error rate threshold is reached.
11. Roll out to the development fleet. Get feedback. Get high level data from disengagements. Use to create new priority list of problem situations.
12. Cycle through steps 2-11 until a new error rate threshold is reached.
13. Roll out to the early access fleet. Use data to update priority list. Work towards a new error rate threshold
14. Roll out to the full fleet.
15. Continue to improve the NNs and code as they work down the priority problem list. This is the march of 9s. Note, they may encounter the same problem multiple times as they move down the list. If they reduce error rate on the top priority problem 10x, this will no longer be top of the list, but once they have tackled the next 100/1000 problems on the list, this initial issue may be number 1 again. Also note, as the system gets better and better and is competent with more edge cases, its understanding also gets more generalised and improving performance on one problem may also improve many others on the list.
16. At a certain error rate threshold, remove the requirement for constant driver attention.
17. At the next error rate threshold, remove the driver.

 

[Artful Dodger]

They GS,MS....will beat the crap out of you and then help facilitate a cap raise....grab a nice fat chunk for themselves. Talk up the stock untill they think it's time to do it all over again.
No wonder EM wants to avoid them.

Lol, Wall St Investment Bankers are like tweakers: they will steal your *sugar*, then help you look for it. Nancy R would say 'No' to Wall St.

LOTSA money out there to fuel Tesla's growth. China is just the beginning, but they will add to the flood. It's all about execution now, starting with InsideEVs numbers for April.

Cheers!

 

[anthonyj]

Ihor Dusaniwsky on Twitter

Short interest is growing

 

[madodel]

Los Angeles sets dramatic new goals for electric cars and clean buildings | HeraldNet.com

LA Mayor Gil Garcetti is proposing requiring all autonomous vehicles to be pure BEV.

That is great. What we need is for NYC to require all commuters going across the bridges and tunnels to be EVs or put an even larger fee on ICE (or minimal/no fee for BEVs) to allow them in. That would be a huge step to carbon neutrality and now with all the compliance cars + Tesla doing over 200 miles it is realistic requirement.

 

[Boomer19]

Ihor Dusaniwsky on Twitter

Short interest is growing

...and growing as of this morning

 

[Causalien]

Ihor Dusaniwsky on Twitter

Short interest is growing

Wow. They are not taking profit at all. Some hard core conviction shorts there.

Have to hand it to them. These shorts are not greedy. They are as faithful as tsla longs.

 

[Remus]

No. That would be incredibly brittle.

These vision NNs just take in pixels and put out labeled bounding boxes. Tesla's labels are ID (car/truck/whatever), distance estimate, direction hint, etc. The NN doesn't know or care about camera location. The heuristic code which builds the 3D view obviously must know camera locations and orientations.

With LIDAR you don't have to "understand". You know there's a large object in a certain location moving in a certain direction. If your vision NNs identify it as an airborne car, terrific. But even if not, you have enough info to take evasive action. (Of course your code has to use that info properly - bad s/w will always defeat good sensors).​

So basically you pickup slight inaccuracies from others to justify your position which has been debunked multiple times. While doing so you continue to pull "facts" out of your behind.

While you have very thick skin to ignore all the posts pointing out your glaring lies.

"heuristic code", really?! If that is heuristic code and the entire system is heuristic. Go back and listen to karpathy about how they constructed 3d model.

And understanding is not needed? What about road signs, lane lines, water puddle or even running water on the road? Is that why they only feasible in Arizona?

 

[Tenable]

Wow. They are not taking profit at all. Some hard core conviction shorts there.

Have to hand it to them. These shorts are not greedy. They are as faithful as tsla longs.

We are below 244 and have an air pocket down to 192-200. Tesla has used up its good news bullets to no avail. If I were short I’d be pretty happy right now.

 

[Tslynk67]

Ihor Dusaniwsky on Twitter

Short interest is growing

I guess the short interest can rise even though the %age activity on any given day is less, if no-one is covering.

Edit: stupid graphics uploading bug on this site!!!

 

[Doggydogworld]

Does anyone know what the new inventory level of S+X is ? At the end of Q1 there were 20k Model 3s in inventory.

I don't know if anyone replied, but my estimate for 3/31 Finished Goods Inventory is:
1050m - New S/X (14k @ 75k COGS, +2k and +150m in Q1)
900m - New Model 3 (20k @ 45k COGS, +12k and +540m in Q1)
150m - Used cars (3k @ 50k each, flat during Q1)
50m - Solar, Powerwalls, etc.
--------
2150m Total

BTW, "New" means never titled. It includes showroom cars, demos and a lot of loaners that may have miles on them. They moved some loaners to PP&E, though, so it doesn't include those.

 

[jbcarioca]

Cool. That's why I've never given out a referral code.

I have yet to give a referral code, although two of the people I referred asked fro a code, so I helped a more needy friend get them both. His daughter needed her baby Tesla more than I needed something then available.

OTOH, if I'd been greedier I'd have a Roadster reservation for free instead of $250,000. No wonder Tesla reduced the richness of spiffs!

 

[TheMrX]

So here's a cool scenario... Pickup reveal day, Elon drives out onto the stage in the most ridiculous looking thing you've ever seen - as he said, steam-punk meets Bladerunner - titanium, FWD's, motor for each wheel, 400kWh battery pack, spiky things all over, etc. He runs through some outrageous power, range, acceleration and speed specs and then reveals a price-tag that's 2x more than the Roadster. Special order, hand-built, limited volumes, collectors piece essentially. That's his vanity project.

He then throws in a "one more thing" and Franz drives out in the mass-market pickup, with, which nobody expects to see at this moment. It's a killer, well specced and well priced. The crowd goes nuts, the SP goes through the roof, even the MSM struggle to pick faults, Toilet-Boy's head actually explodes, like in Scanners.

This would be something.

Je n'ai pas l'habitude de généraliser, mais je dirais que ça prend bien un Belge pour nous faire un scénario de BD avec le Pick Up reveal

 

[lklundin]

So here's a cool scenario... Pickup reveal day, Elon drives out onto the stage in the most ridiculous looking thing you've ever seen - as he said, steam-punk meets Bladerunner - titanium, FWD's, motor for each wheel, 400kWh battery pack, spiky things all over, etc. He runs through some outrageous power, range, acceleration and speed specs and then reveals a price-tag that's 2x more than the Roadster. Special order, hand-built, limited volumes, collectors piece essentially. That's his vanity project.

He then throws in a "one more thing" and Franz drives out in the mass-market pickup, with, which nobody expects to see at this moment. It's a killer, well specced and well priced. The crowd goes nuts, the SP goes through the roof, even the MSM struggle to pick faults, Toilet-Boy's head actually explodes, like in Scanners.

This would be something.

Do you realize, you just delayed Tesla's Pick-up Truck reveal by six months?

Because now, they _have_ to make both versions...

 

[InDaClub]

We are below 244 and have an air pocket down to 192-200. Tesla has used up its good news bullets to no avail. If I were short I’d be pretty happy right now.

No worries, I call that a buying opportunity - starting this week I'm buying TSLA every single week.

 

[Carl Raymond]

Wow. They are not taking profit at all. Some hard core conviction shorts there.

Have to hand it to them. These shorts are not greedy. They are as faithful as tsla longs.

Of course they are greedy. Which suggest external funding, seeking to harm.

 

[StarFoxisDown!]

Pretty crazy that we're approaching all-time high short interest with so many catalysts on the one year front (Pick up reveal, Giga 3 operational, FSD feature complete, probably additional S/X refresh-interior) and initial production trials of Model Y, Semi, and Roadster at this time next year. There's sooooo much stuff coming in the next year. Shorts are basing their entire basis on one bad quarter which was mostly bad because of production stalls and battery cell shortages.......