I've been testing Grok for a few days, and it feels like a major step backward. It randomly deleted some of my code - something I haven't had happen in a long time.
While the top coding models have become much more trustworthy lately, Grok isn't there yet. It doesn't matter if it's fast and/or free; if you can't trust a tool with your code, you can't use it.
Kilo Code has a free trial of Grok Code Fast 1 and I've had very poor results with it so far. Much less reliable than GPT 5 Mini, which was also faster, ironically.
To me, "full self driving" means you can hop in the back seat and have a nap. If you have to keep your hands near the wheel and maintain attention to the road then... shrugs not really the same. IMHO we're in the "uncanny valley" of vehicular automation.
I would add that "full self-driving" also means that the car company or the self-driving development company holds all liability in a car accident that the owner has none. Even Tesla right now states that the owner holds the liability in any accident. [0]
There are no proper retention laws with car manufacturers and self-driving development companies that I know of.
everything a layman would call "AI" is in the "uncanny valley" at the moment!
- Boston Dynamics' Atlas does not move as gracefully as a human
- LLM writing and code is oh-so-easy to spot
- the output of diffusion models is indistinguishable from a photo... until you look at it for longer than 5 seconds and decide to zoom in because "something's wrong"
Maybe it's because we get use to it and therefore recognize it easier, but it does seem to get more and more recognizable instead of the opposite, doesn't it?
I think I could recognize a ChatGPT email way easier in 2025 than if you showed me the same email written by gpt-3.5.
I think cursor was using it on my behalf and it's been making many tiny edits and also doing stuff I never asked it to do. I can rely on claude to limit itself to what I ask.
Because it's an obvious waste of time that should just never happen.
Not to mention that accidents happen, not everyone always has the good habit of using version control for every change in every project, and depending on the source control software and the environment you work in, it may not even be possible to preserve a pending change (not every project uses git).
I have heard real stories of software bugs causing uncommitted changes to be deleted, or causing an entire hobby project to be wiped from disk when it has not been pushed to remote repositories yet. They are good software engineers, but they are not super careful, and they trust other people's code too much.
While the top coding models have become much more trustworthy lately, Grok isn't there yet. It doesn't matter if it's fast and/or free; if you can't trust a tool with your code, you can't use it.