Yes, the update process frequently fails on my EQEV, no doubt because I live in an area with extremely poor cell signal. Basically the updates have to run when I'm away from home. I was not referring to issues with the update process, but regressions introduced in their content. Bugs that affect the infotainment, various apps, bluetooth, etc. GM does not have the best firmware engineers. This is also on Google, who develops Android automotive. But ultimately, the buck stops with GM.

It looks like the 2026 Leaf finally switched to liquid cooling. Can't believe it took them that long.

I had a 2012 Leaf which was leased. I agree with you. It sucked. Mine was not totaled. I had $0 down 39 months lease. I returned it 9 months early.

My husband drives a 2017 Bolt. It had a battery replacement due to the recall under warranty. New battery increased range from 238 to 259 miles. Car has been flawless. I suggested we replace the car. He is happy with it still and doesn't want to.

My Equinox has great range and price was hard to beat. But lots of software bugs, which OTA updates don't fix, or worsen. I wish they made a Bolt with the same range as the Equinox as I find the Equinox a bit too wide and challenging to park.

Yes. We are agreement on that.

Not all EVs. Look at Nissan Leaf, Chevy Bolt , Equinox EV, Hyundai Kona. All of them under $35k. The median US vehicle purchase price is now just under $50k. That includes many vehicle sizes, though.

My Equinox EV was $27k out the door including about 10% CA sales tax last August before the $7500 incentive expired. I was sti able to take it as point of sale despite having only 3 digit federal tax liability last year.

Exactly. If your output is identical to copyrighted work, it doesn't matter if it came from an LLM or whether you dreamt it.

Exactly. And many AI companies have settled cases, for example with authora, to prevent court rulings they might have lost in order to stop creating legal precedent.

Eventually, one of those cases will reach a judge. It is only a matter of time.

AI is not sentient or a legal entity, and can't make claims. Humans can already successfully prompt LLMs and get it to regurgitate word for words excerpts from books and movies. I have done so myself. There is no reason it couldn't be made to regurgitate exact code the same way. Whether it has ever seen the training data is not even relevant in the eyes of the law. It doesn't matter what mechanism it is using to infringe. If it is producing an exact copy, that is clear infringement.

This new button on other posts. Or so I'm told.

Actually it really is terrible at checking facts. It's a probability engine. Try asking it anything about music. This is the kind of result you'll get. Pure hallucination. It can't even figure out what key the works is in. And recommends a recording that doesn't exist.

https://chatgpt.com/share/6a6873a5-b874-83e8-bc1d-5e21a302769a?ogimg=plain

Qwen 3.6 35B-A3B and Qwen 3.6 27B Coder were the largest 2 I tried to run. Neither really "runs" on my 3060 Ti / 8 GB or GTX 1660 Super. Crawl is more like it. But even when waiting a long time for jobs to complete, the result on the codebase analysis was worse than useless - only hallucinated findings.
For other simpler tasks, I have seen Qwen write some useful code. But it often rewrites unwanted code as well, and breaks that one in the process.

Lol. I wish. It matches with my own findings, unfortunately.

Local LLM is good enough as Home assistant conversation agent, though.

Ford, not GM. And they are only rehiring some engineers, not all. AI is still killing many jobs.

Have you tried small models ? I have. They hallucinate way too much. I had codex run a whole bunch of tests over the weekend to analyze a relative small code base. It automatically downloaded many models from huggingface, with various quants, and tried them all. Few completed within a timeframe that a human would be willing to wait for. We are talking 1 hour timeout. But for the ones that did, the report was mostly incorrect, with the local models hallucinating 5 different flaws that didn't actually exist at all - according to the large models.

It's fair to say that small models are far less useful than large ones. They have some use, but ultimately are unsuitable for many tasks.

I do agree we need orders of magnitude improvements in performance. Not sure if we will see them.

Facebook knows too much about too many people. They have enough kompromat to be able to survive as a mass blackmailing business.