Local AI. Building @picogpt Download: https://teapaw.org/t.co/o0PKhVM691 Discord: https://teapaw.org/t.co/KNirjAzl32
San Francisco, CA
Joined April 2008
- Tweets29K
- Following6.5K
- Followers6.9K
- Likes105K
Pinned Tweet
Apple Silicon + Gemma 4 fans: this is for you.
Pico AI Server now supports continuous batching with MLX-Swift.
43 tok/s on 1 stream.
26 tok/s per stream on 2 concurrent streams.
That’s 52 tok/s total. a 21% throughput gain on a six-year-old MacBook Pro M1 Max!
This sounds promising. I will stop making Siri jokes if this works well
There’s more: 27.3 introduces a memory graph.
It’s centered on you: relationships, places, routines, health, workouts, and what motivates you or holds you back.
Plus “memory atoms”: facts, opinions, intentions and events, with time, place, people and topic.
Tidbits:
• The memory graph is filled by two kinds of recipes:
1. Fastpass: Turns already-structured data into insights. For example, Person-v1 reads your Me card’s name fields.
2. Agent: An on-device or PCC model extracts insights by querying Siri chats, messages, photos, health data, and other sources.
• There’s only one agent recipe today, MotivatorBarrierResource. Its prompt asks 60 built-in questions, including “Do extended solo media activities during evening downtime routines highlight time availability barriers for adequate sleep?”
• The memory graph is still being built. For example, there’s no process that creates memory atoms yet.
• Memory atoms can generalize into broader memories. Each is tagged with one of 21 topics, such as food, travel, finance, or services, which can store your dentist or lawyer.
Here’s where to find the DBs on Mac:
• Memory graph: ~/Library/CorePersonalization/insight/MemoryGraph/memorygraph.db
• HybridSearch (incl. explicit-memory records, none yet): ~/Library/Daemon Containers/<UUID>/Data/Library/HybridSearch/index/C/index.db
Both use HybridDatabase, Apple’s fork of Kùzu (storage v57), so the public kuzu package can’t open them. You can use HybridDatabase.framework.
🤖 Made with AI
🚫 @pmarca is no longer following @DoDeportations
Still followed by: @basedmikelee, @palmerluckey
$META
A man of principle. Some hills are worth dying on
Italian chef fired for refusing to put pineapple on pizza bbc.in/4egT8uf
This is so cool. I check my home and my previous homes and it’s all correct
Introducing Isometric SF A pixel-art map of the entire SF Bay Area, over 1,300 square miles, all painted by Klein, a 4B-parameter image-to-image model, on @modal.
Explore the map now!
isometricsf.com
This is what Siri should have been
After Google acquired my AR startup, I started thinking about the next interface, this time for AI. Text and voice work fine, just like DOS did before the Mac. But nothing’s more intuitive than face to face.
After two years of research, today we’re launching Atmee v2: a real-time avatar model that makes live video agents feel truly life-like.
• Any character from 1 image + 10s of audio
• Contextual expressions, gestures, even interacting with the environment
• Live video calls, voice only, or video generation
At its core, Atmee v2 is a video foundation model, it can render anything, but it’s built to excel at characters, which is what video models like Seedance 2.5 and Wan 3.0 are mostly used for. In comparison:
Atmee v2
• Up to 50x faster
• 10-100x cheaper
Build tutors, sales coaches, support agents, companions, game characters, even whole living worlds. The interface after text and voice starts today.
Layoffs suck but if you have to do it, do it like Canva does. Well done (though I’d rather get a 1-year Codex or Claude Code subscription instead of LinkedIn premium, but maybe that’s just me) ;)
Back when Apple still took risks. The iPod Shuffle is such a weird concept I don’t understand how it became a product.
(I bought the first gen Shuffle out of curiosity and ended up using it more than my big iPod)
Ronald Mannak retweeted
We're hosting two Apple Intelligence Group Labs this week to answer your questions about the newly announced Siri AI Developer Preview, as well as any other questions you have about Apple machine learning and AI frameworks.
November 4, 9:00–10:00 a.m. PST Register here: developer.apple.com/events/v…
November 5, 6:00–7:00 p.m. PST Register here: developer.apple.com/events/v…
Or join us on the Apple Developer YouTube channel: youtube.com/@AppleDeveloper
You know what? Let’s forget this ever existed. Let’s wait another year or so until the technology catches up. We’ll get there, but this isn’t it.
Watch Episode 1 of our Odyssey series made with Grok Imagine.
Let us know in the comments if you'd like a full breakdown of how it was done!
Creative collaboration with @WonderStudiosX and director Dev Walker @glumlot_
Maybe a year was too conservative
If your AI characters sound fake, the fix is almost always in how you write and DIRECT the dialogue in your prompt.
I recently started on the dialogue-heavy scenes for my AI feature film about Beethoven, and too many of the lines kept coming out wrong.
They’d sound too theatrical, too monotone, or too much like a practice read-through between actors.
(all this despite the fact that Seedance 2.5 is fantastic when it comes to realism.)
So through a whole lot of trial and error, I've found that these 5 things made the biggest impact on the quality of the dialogue delivery in my generations:
1. Paying close attention to the written punctuation used.
2. Breaking long lines into shorter ones, with intentional pauses.
3. Using bodily actions as punctuation (this one gave the biggest improvement BY FAR! In the attached clip, note the sniffles in between Beethoven's lines. That alone made the delivery feel much more natural.)
4. Using "anchor gestures" (he lifts his finger as if to say "one sec")
5. Just split your generations in two (stupid simple, but it really makes a difference.)
So if your characters keep sounding like they're reading from a teleprompter, give this a try!
