๐ฆ๐ฎ๐ฝ๐ถ๐ฒ๐ป๐ซ retweeted
Fitts's Law, magnetic attraction, and nodes that can't keep their hands off each other. Building delight is the art attached to the science.
I hate using multiple smart device apps to control the smart devices in my room, so I setup a raspberry pi server, reverse engineered a couple of protocols, added some UX magic, designed an AR enabled app that lets me go โlumosโ and turn the lights on
So far I have integrated: Smart bulbs, Smart strips, AC, Bluetooth speaker, air purifier, and a couple of lamps
Weโve acquired Orchestra Bio, a platform that helps biopharma teams coordinate the complex process of drug development.
Together, weโll build on Rosalind, our AI models and tools for life sciences research, to streamline those operations and help new treatments reach patients sooner.
Welcome @ashokaraj_ and team!
openai.com/index/openai-to-aโฆ
๐ฆ๐ฎ๐ฝ๐ถ๐ฒ๐ป๐ซ retweeted
Perplexity Computer can now run fully local on your Mac.
Work with your native Mac apps and local files, and run tasks without sending sensitive data to the cloud. The model and the agent both run on your Mac.
Available now in the Perplexity Mac app.
๐ฆ๐ฎ๐ฝ๐ถ๐ฒ๐ป๐ซ retweeted
One script. Unlimited emotions. ๐ญ
With Fish Audio Drama 3, you direct every performance, seriously!
Try it on our web app and API ๐ณ
๐ฆ๐ฎ๐ฝ๐ถ๐ฒ๐ป๐ซ retweeted
send your landing page to claude with this prompt
ull get a video like this, made with code for free
thank me later โ
<inputs>
Ask me for: my landing page link, my brand name and logo mark, a two-part closing line (a statement, then a short italic tail), and a 15 second track (optional).
</inputs>
<direction>
A 14.5 second 3D fly-over of my landing page, 1440x1152 at 60fps. The page lies flat on a bright studio floor (#f9fbfb in the middle, #f2f3f6 in the corners) with a soft wide shadow around it and 24px rounded corners. One continuous camera, never a cut: it starts low over the hero and glides backwards down the page at one steady speed while looking up it, rises straight overhead to show the whole page as one strip, dives into the hero and lands it as a card under a two-line title. Focus is razor thin in front of the subject and almost none behind it, so everything near the lens melts. Cards float a little above the page with soft shadows; photos, chips and the testimonial lift off and stand up toward the lens as the camera passes them. Four small tiles of my logo lift off beside the hero button and stay with the film until the end card. The title is Geist 500, the italic tail Instrument Serif.
Banned: cuts, crossfades, stops or surges in the glide, the camera turning or swinging sideways, zooming in and back out, glows, any text laid over the page before the end card.
</direction>
<structure>
14.5 seconds: flight 0-8.2s, rise 7.75-9.25s, overhead 9.25-9.47s, dive 9.47-9.97s, title card 9.97-14.5s.
The camera 56ยฐ off straight-down and 1360px from the point it looks at, centred on the top half of the hero, the phone or product mockup sharp, four 32px logo tiles beside the button โ 0-0.6: the glide eases in from rest โ 0.1-1.35: the tiles lift, grow to 140-200px and hang just outside the page's top corners while the hero's chips rise to 60-80px and stand up 30ยฐ โ 1.25-1.6: the glide doubles to cruising speed as the hero falls behind โ 1.6: the second section's title passes the centre of the frame, its photos blurred near the lens โ 1.35-2.7: that section's main photos lift to 100-130px and stand up 46ยฐ, two of its chips fly up to about 200px, stand up 54ยฐ and pass the left and right edges at 2.5 โ 1.2-2.4: the tiles trail the camera 300px down the page, then drift out of the top โ 3.0-7.3: every following section passes through at the same steady speed, each one sharp as it crosses the middle of the frame and soft below it; a logo tile drifts past the left edge close and blurred at 5.4 โ 5.45-6.3: the testimonial card rises from 22 to 112px, stands up 30ยฐ and shows a 34px white edge โ 7.3-8.2: the glide eases off as the footer arrives, the footer title sharp at 8.0 โ 7.75-9.25: the rise, zooming out at an even rate while the tilt flattens (8.15-8.95) and the page's bottom edge settles 42px above the frame bottom โ 9.25-9.47: the whole page as one centred strip, creeping 3.5% closer โ 9.47-9.97: the dive, the page's top edge pinned to the frame top while it grows, everything below the hero fading out, then the top drops to y 283 so the hero lands as a 1018px wide card with 38px corners and a soft shadow; two logo tiles sit on its top corners, one crisp at the right edge, two huge soft blurs drift in at the bottom corners โ 10.02: a white badge pill (logo + brand, 21px) fades in at y 44 โ 10.12: line 1 at 74px, one word every 0.06s โ 10.36: the italic tail at 82px in grey #8b8b92, one word every 0.05s, each word rising 16px out of a 12px blur in 0.34s โ 10.6-14.5: hold, the card creeping 2% closer, the tiles drifting a few px.
</structure>
<build>
1. Open my link in headless Chromium (Playwright) at 1440px wide, device scale 2, and scroll to the bottom once so lazy images and scroll animations finish. Mark the pieces that lift (cards, phone or product mockups, chips, feature photos, the testimonial, the hero button) with data-layer. Screenshot the full page with them hidden, then each piece alone on a transparent background with 28px padding, and save every piece's page position to a JSON.
2. One WebGL2 film page, 1440x1152, every frame computed from time inside seek(t): no timers, no CSS transitions. The page is a textured quad on the floor and each piece its own quad at its height, drawn lowest first with no depth buffer. Use mipmaps and 16x anisotropic filtering or the far text shimmers.
3. Drive the camera from one speed curve, never from a position key every half second. Focal length 1880px, 56ยฐ tilt, distance easing out from 1360 to 1700px and never back in, no turning. Speed down the page as a share of cruise (cruise is about 770 page px/s): 0 at 0s, 0.5 by 0.6s, held at 0.5 to 1.25s, 1.0 by 1.6s, held between 0.94 and 1.1 to 7.3s, then easing to 0.12 by 8.3s. Integrate it into the page position.
4. Depth of field per pixel: blur radius = 75 x (focus - depth) / depth in front of the focus and 6 x (depth - focus) / depth behind it, capped at 120px, 40 taps on a golden-angle disc, each tap read from a mip level matched to the blur. Keep the focus on whatever crosses the middle of the frame.
5. A lifted piece stands up by turning about its own middle, top edge rising toward the lens. Each casts a shadow on whatever is under it: the same picture in flat #1a2133, offset (0.10z, 4 + 0.36z), blurred 7 + 0.5z, 26% opacity.
6. Rise: zoom out in log scale with a smooth trapezoid speed (speeding up 7.75-8.35s, even 8.35-8.95s, slowing to a stop at 9.25s) and hand the glide over to a camera that holds the page's bottom edge at that 42px line from 8.3 to 8.65s. Dive: log scale on a quadratic ease in-out, then solve the look-at point from the pinned edge. A rounded-rect mask morphs from the full page to the hero, 24px to 38px corners.
7. Render with Playwright on the real GPU (--use-angle=d3d11 --enable-gpu --ignore-gpu-blocklist): 60fps, 16 subframes across 0.7 of a frame for motion blur, H.264 MP4. The track's hit goes on 9.97s, whooshes on 8.66 and 9.80.
</build>
<gotchas>
If you can't run code on my computer, tell me straight away that you can only make a live preview here and that the MP4 needs Claude Code. Phones or cards that fade out with transparency show whatever is behind them, so fade them into the haze colour first and only cut to transparent at the very end. Keep most lifts tiny (20-40px) and lock the page before timing anything, because any change in a section's height shifts everything after it. Before the final render, measure a quick 30fps preview with optical flow: the speed down the page must never dip and the zoom must never turn into a zoom-in during the flight.
</gotchas>
<start>
Ask me for the inputs, capture my page, then show me stills at 0, 1.6, 2.5, 5, 6.5, 9.3 and 11 seconds before you render.
</start>
Grok Voice 3.0 is now the worldโs leading conversational voice model
SpaceXAIโs Grok Voice 3.0 is the new leading Speech to Speech model, debuting at #1 on the Artificial Analysis Speech to Speech Index at 86.9, while also being competitive on latency
Grok Voice 3.0 is @SpaceXSI 's successor to Grok Voice 2.0, a Speech to Speech model available through the SpaceXAI Realtime API at $0.08 per minute. We evaluated Grok Voice 3.0 with the Rex voice through the SpaceXAI Realtime API.
Key takeaways:
โค Speech to Speech Index: Grok Voice 3.0 debuts at #1 at 86.9, ahead of GPT-Live-1 (Astra, medium) at 83.2, Gemini 3.8 Live Extended Thinking (High) at 82.6 and GPT-Live-1 (Sol, low) at 80.1. It is up on Grok Voice 2.0 at 81.3 (+5.6 points) and Grok Voice 1.0 at 73.7 (+13.2 points), released just under six months ago. The Index averages Speech Reasoning (Big Bench Audio), Agentic Performance (Tau Voice), Arena Preference, and Task Success Rate
โค Speech Agent Arena: Grok Voice 3.0 ranks #3 in preference at Elo 1070, behind Gemini 3.1 Flash Live Minimal (1100) and Gemini 3.8 Live (1088) and ahead of GPT-Live-1 (Sol, low) at 1057, and #1 on Task Success Rate at 94.8%, highest on the board, just ahead of Grok Voice 2.0 at 94.6%. Grok Voice 2.0 trails at Elo 1011
โค Tau Voice: Grok Voice 3.0 takes the #2 spot on our Tau Voice benchmark implementation at 71.6%, behind GPT-Live-1 (Astra, medium) at 74.5% and ahead of Gemini 3.8 Live Extended Thinking (High) at 68.6% and GPT-Live-1 (Sol, low) at 59.3% - up from 56.5% for Grok Voice 2.0
โค Big Bench Audio: Grok Voice 3.0 scores 98.5% on audio reasoning, ahead of Gemini 3.8 Live Extended Thinking (High) at 97.7% and Grok Voice 2.0 at 97.2%
โค Speed: Average Time to First Audio on Big Bench Audio is 0.81 seconds for Grok Voice 3.0, well ahead of GPT-Live-1 (Sol, low) at 1.24s, GPT-Live-1 (Astra, medium) at 1.34s and Gemini 3.8 Live Extended Thinking (High) at 1.35s, though slower than Grok Voice 2.0 at 0.70s
โค Cost: Grok Voice 3.0 costs $4.80 per hour of input audio, the same as Grok Voice 2.0. That is above GPT-Live-1 (Sol, low) at $4.47 and Gemini 3.8 Live Extended Thinking (High) at $3.50, cheaper than GPT-Live-1 (Astra, medium) at $5.83, and ~2.2x cheaper than GPT-Realtime-2.1 High at $10.75
Congrats to @elonmusk and the @SpaceXSI team ๐
See below for more detail โฌ๏ธ
๐ฆ๐ฎ๐ฝ๐ถ๐ฒ๐ป๐ซ retweeted
We've been working hard on making the Claude Code desktop app great for running lots of sessions side by side.
It's now much quicker to start sessions, easier to split work across a few, and you can always see what each of them is doing. Hereโs what's new:
๐ฆ๐ฎ๐ฝ๐ถ๐ฒ๐ป๐ซ retweeted
MUSE WONDER
this is the culmination of over a year of research and infrastructure work in msl
since we originally shipped muse spark in april, we have continued to scale our models across pre-training and RL, now reaching the frontier ๐
see our model's improvements on the artificial analysis index:
๐ฆ๐ฎ๐ฝ๐ถ๐ฒ๐ป๐ซ retweeted
OpenClaw v2026.10.2 patch release is out ๐ฆ
๐ง Upgrade and Doctor fixes
๐ฌ Missing and duplicate reply fixes
๐ง Long-conversation fixes
๐ Credential-masking fixes
๐ More bug fixes and tweaks
107 contributors. Thank you! docs.openclaw.ai/releases/20โฆ
๐ฆ๐ฎ๐ฝ๐ถ๐ฒ๐ป๐ซ retweeted
A developer says he used Opus 5.5 to recreate seven essential Adobe apps, including Photoshop, as free, open-source alternatives (no more subscriptions).
Built in Rust, free to use, with public source code.
These are early versions, PhotoCraft is not yet ready to replace Photoshop for daily professional work completely, bout about 90% they say.
What a fantastic time to be alive. I love it.

























