Recent content by mrmage

Go is complicated for a game, but not compared to predicting appropriate behaviors in the real world. Alphago was also a blend of algorithms and ML, whereas Tesla gave up on rules / algorithms because they didn't work in the real world. As far as complexity, Alphazero has about 20M parameters...

I asked ChatGPT4 about the conversion rate for my Tesla convertible bonds, taking into account splits. Last year, it responded authoritatively with an answer that was wrong. Today, ChatGPT4 gave the right answer. It's getting better, but it takes awhile. AlphaGo is predicting moves in a game...

Training creates the model parameters, which are then used for inference to predict the best behavior. The amount compute required for inference is closely related to model size. A small LLM of 10 to 20 GB parameters generates only a few tokens with a fast CPU, and perhaps dozens of tokens on a...

Although fast compared to training, inference is still compute and memory intensive compared to algorithms (ie hand coding). The code saved is small compared to model computing and size. Tesla's first priority and top challenge will be to get FSD to work, which includes reducing interventions...

What I meant is that datacenters can increase model sizes orders of magnitude larger by simply adding more computers. For edge computing applications like FSD, the model size is limited to available memory without expensive retrofits in the field. It's worth noting that real time edge models...

I just tried FSD 12 and was very impressed. Only 2 interventions needed over 30 miles and much more natural behavior, which is orders better than anything Tesla released before. As a potential investor though, I need to determine whether FSD can achieve an intervention rate of 100x better for...

My other investments are likely to grow too, and empirically I made the right decision. My former TSLA investment has grown enough in other assets that I can almost double the TSLA shares if I converted back. If TSLA has the potential to leap past $700 a share in 2024, I'm happy to accept my...

There's certainly an echo chamber effect. Many with unpopular views end up leaving. It's too bad, because investments are as much about headwinds as tailwinds. Although I sold off the last of my TSLA stake earlier this year, I would reopen a position if FSD or Teslabot gets close. For now, it...

Guess my thoughts about TSLA not breaking 170 without blowout earnings didn't age well. I'll admit that Tesla is not boring. Opened 170 cc for next Friday yesterday after making a few cents on 160, but will probably let them get assigned next week if TSLA continues up. Then write puts for...

It's usually 2 or 3 trading days after earnings where it gets really interesting. Aka it isn't over until the Friday lady sings.

Was going to do something similar because the results didn't seem to warrant the bump, but my CCs would be naked.

In case I want to try some AH action, is there a better way to check for the ER summary besides refreshing the Telsa IR page for the next hour? Edit: That was fast. NM...

Not following the logic, so maybe I'm misunderstanding the question. Theoretically, assignment can occur anytime the option holder (the buyer of the options) wants, but in practice it usually happens after expiration which is Friday (two days from now). Most option holders don't assign early...

Wrote Jan27 160c for 1.83 @ SP ~145.5 Should have been 1.88, but rushed and went long before realizing my mistake and lost a few cents to correct. :( Going to play this two ways: if SP pulls back, I'll close it before end of day. If SP goes up, I'll see what happens after earnings. Bought my...

I closed my options earlier this week because I didn't have a good sense of the direction at this point, with ER coming out. The profit estimates on here put Tesla 10% higher than the street. If that happens, I don't see the SP blowing past the 150 call wall this week after the recent run...