maltz
About
- Username
- maltz
- Joined
- Visits
- 188
- Last Active
- Roles
- member
- Points
- 1,450
- Badges
- 1
- Posts
- 552
Reactions
-
Apple rolls out ninth developer beta for vision OS 2
"outs"?I do not think that means what you think it means...
-
ISS astronauts can watch 'Ted Lasso' and 'For All Mankind' in space
Why would they be up-linking anything? I get that weight is a premium for launches, but so is bandwidth. A single hard drive these days can hold over 11,000 hours of 1080p video at Netflix streaming rates, if my math is right. Seems like they could manage a pretty decent on-orbit library?
-
Hands on: Siri starts to get better thanks to Apple Intelligence
To be fair, I'd have been right there with them a month or two ago. I fell into a rabbit hole exploring LLMs after DuckDuckGo announced their privacy layer for a few models, including ChatGPT, and then I went even deeper when I discovered that running an entire LLM on your own hardware was possible. Since then, and especially with the latter, I've had my mind blown over and over and over again as I've explored what LLMs can (and can't, and sometimes will/won't) do.But I do want to draw a distinction between an LLM and the much broader term "AI". Siri almost certainly does use some form of AI right now, just in translating the sounds you make into words, if nothing else. Given all the differences in the way people speak, even within a single country, that alone is a feat - and one that Siri has become more and more adept at over the years. But Siri is clearly only using AI for that part, and is otherwise mostly using keywords for parsing and taking action on your commands. Employing an LLM for that would allow Siri to actually understand your meaning and not just parse for keywords. That's really game-changing, and actually can be done on-device in a way that solidly protects privacy.
-
Hands on: Siri starts to get better thanks to Apple Intelligence
DAalseth said:Text Faith and ask what she wants me to make for dinner, then set a reminder to start cooking at 5," gave us a single text message with the contents "what do you want me to make for dinner, then set a reminder to start cooking at 5.In fairness this is a tricky request. We’ve all heard the joke where the wife says, “Stop and pick up a quart of milk and if they have eggs get twelve,” and the guy comes back with twelve quarts of milk. Knowing where to parse out these requests can be very hard. People get it wrong all the time. Indeed, a lot of this can be more cultural, than simple language management. Perhaps saying “…and also set a reminder…” would have made a difference.
People say how bad Siri is, and admittedly it could be better. But I’m very careful with how I word my requests, and seldom have a problem.I'm not sure people realize how capable modern LLMs are for things like this. That's not a difficult request at all, even for a tiny LLM that would easily fit in a cell phone. Here's a response from Gemma2:2b, for example:>>> You are an AI assistant on a cell phone. Your user asks you to "Text Faith and ask what she wants me to make for dinner, then set a reminder to start cooking at 5" Describe the actions you take to fulfill the request.Okay, I can help with that! Here's how I'll approach this:
1. **Text Faith:** I will craft and send the text message to Faith: "Hey Faith! What are you craving for dinner tonight? I want to cook
something delicious!"
2. **Set Reminder:** To remind myself of the cooking time, I will set a reminder on your phone that says: "Dinner Prep: 5 PM - [Insert Name of
Dish/Type of Meal]". You can customize this further if you like!
-
Google Ads used to distribute Mac malware disguised as 'Loom' app
This is a very common, and not new thing. But it's good to see a reminder now and then - Don't click on google ads. Scroll down to the search results and click on those.
