I got semi average score of 2.4k not trying super hard, but the "moving click" exercise seems dubious, got literally zero off it. Not sure if it's comparable to any actual game?
Not about the bugs, but feature removals. Known issues is a whole different thing from "we don't care about doing this properly so we just removed a bunch of stuff we've had for a long time because it's easier than maintaining it"
I've used Kasa plugs for a long while and was not surprised that their API allowed relay control and basic info of them as long as you manage to get in the same internal network. It's local, so IMHO that is not just reasonable, it's desirable. I don't need to give my friends permission to toggle the lights manually either.
Routers having abnormal amount of zerodays, and not being fixed on the other hand is actually serious, unlike this.
Just a week ago I actually set up one of TP-link's new line of smartplugs (Tapo instead of the old Kasa), and for that I had to make an account. For actual security, I'd rather have an option to control them locally with zero additional authentication when you're already inside the network, instead of the cloud stuff. But I HAD to make an account even though the custom code I control said plug with only accesses the plug locally.
An arbitrary data error I found is that https://en.wikipedia.org/wiki/Jekadefari is shown on the year 760251. That number is obviously wrong, it instead appears to be the postcode of the place.
Cool, though I assume there's some accuracy shortfalls, when not close to zero the UI breaks when years are ~1 pixel in size.
Probably doesn't matter for much of the content, though immediately comes to mind that for after big bang there is "known events" that happened at second-scale. Don't know if there's exact Wikipeida articles of those, but with an appropriately accurate timestamp storage & handling (128bit? more?), one could well zoom in to those if they did exist.
Not conflating anything - Gemini developers _absolutely_ knew that their biasing will only increase the amount of untruth yet they still did it. To hell with the truth as long as it hit their whatever "diversity/inclusity" goals.
No liars lie for the sake of lying - they always have a goal in mind. Google had theirs well in sight here.
xAI _actually_ cares about the truth. And puts value to it. So situations where 50% chance of being truthful is traded for 10% less chance of it saying a bad thing, xAI will not choose it. Like any model developer with moral fibre wouldn't.
About your comment - It is sometimes hard to speak things you know don't make sense to technical and truth-adhering people on HN. You should take that as a sign that maybe you should rethink your stances.
Don't need an ounce of corporate bloat to be efficient in your own free-time projects. Writing some basic C/C++ OpenGL stuff, especially when using a ready-made engine, is now easier than ever.
Using a browser here is like buying an expensive, professional tier Stihl chainsaw just to cut two small twigs in your garden. Stupid overkill is stupid.
A 256MB Raspi Zero can comfortably drive a HDMI-connected display at 1080p60, and not even drop frames when rendering textured+shaded 3d if you just reasonably account for overdraw.
Drawing something as simple as this with a browser is peak overkill. If the watch had a unreal-like PBR graphics diorama for example, gigabytes would at least be argued for.
This. I've recently been porting my ECS-based 3d game/app engine for my 256MB Raspi Zero AND ESP32-S3, so reading that you need gigabytes to comfortably render a clock appears just ... stupid.
For reasons I'd use an LLM to learn about things / find relevant sources? A lot of things have politics ingrained in to them, and from experience I know some models will absolutely lie to avoid even being a lightyear from saying anything offensive.
Your arguments defeats itself, as Grok is the only one that let the user make it say such things by command. People act like Grok did that on it's own lol.
I want a car that If I decide to run it in to a wall to avoid running over people, it does that. People arguing that Grok doing what the user asks for, being actually impartial, being somehow "biased" is crazy.
The training data and the end result should reflect the real world. No you don't need to "add bias to fight bias", just keep it truthful. Don't force SF-brained forced diversity/inclusion/whatever, we know well some models do.
X serves as a good real-time additional source of training / RAG data for Grok. But it's always been built to be a competitor for what OpenAI was founded for, after they went rogue with the "Open" part. And today, Grok 4.5 is equal to Claude Sonnet while being a lot faster and cheaper to run.
My original comment disproves your argument about the "junk" you talk about. If one is so scared of offending someone that intentionally they make their model lie, they deserve to go bankrupt.
This. It's like people collectively forgot about the "misgendering worse than thermonuclear war, founding fathers were black" stuff.
Always go for Grok first for political questions. Other models have such a bad history of being so crudely aligment-hacked, I'd feel like a fool trying to get an impartial answer out of them on some political figure for example.
How is changing history and making a a group of objectively white people black "removing bias" ˋ? What it is is literally bias. Like thinking Pi could be 4. Removing bias ends with truth, not these crazy wonky results.
What a crazy thing to say explicitly about the model that _avoids_ that kind of stuff. Did we already collectively forget about the ChatGPT "misgendering is worse than thermonuclear war" case, or Gemini picturing the Founding Fathers as black?
Anything political, I always go to Grok first. It's the only one that has bled for not just trying to play it super safe with political correctness, but trying to be impartial.