Some things only make sense once the last line is drawn.
0:48 0 0 0 0 36 0
Paolo Rosson
@redp314
Head of Applied AI at Dext. Building menuphotoai.com | Quantum physics PhD from Oxford | Former Italian blindfolded Rubik’s cube record holder
2KFollowers
2.3KFollowing
1.4KPosts
208Media
2.5KLikes
0:48Time heals everything
0:30 0 0 0 1 217 0
"But can it run Doom?" My glasses can now! @EvenRealities G2
0:13 1 7 4 20 6.2K 2
got @Muse to do some customer emails for menuphotoai.com and it got it's first conversion! @alexandr_wang
0 0 0 0 84 0
testing the new world model by @odysseyml with a "safari in tanzania". Crazy what can be done today!
Still not perfect with some inconsistencies but the results are so promising. Also command latency is just a couple of seconds!
Still not perfect with some inconsistencies but the results are so promising. Also command latency is just a couple of seconds!
1:02 2 2 0 39 1.7K 13
When 11 people join a zoom call
1 0 0 3 175 0
Time heals everything. Even sand that was never there. Not filmed: every grain is code, by Claude Code with Opus 5.5!
0:30 3 0 1 10 1.9K 1
Got Claude Opus 5.5 to make an original oil painting from scratch!
365 brushstrokes later...
365 brushstrokes later...
0:32 66 9 81 192 779K 102
Finally found a use for the Touch Bar: my Muse lives in it now! @alexandr_wang @natfriedman
It wanders around, chases fireflies, and you can pet it. Tap it, ask out loud, and the answer runs along the bar.
It wanders around, chases fireflies, and you can pet it. Tap it, ask out loud, and the answer runs along the bar.
0:26 20 7 7 111 70.4K 20
This was literally the rating explanation in the last airbnb i was in. 4 stars mean several issues to be fixed 🤡
@levelsio@levelsio · Oct 5Airbnb's entire rating scale is now 4.5 to 5.0
You can't make this up 😂
You can't make this up 😂
Open quoted post →
0 0 0 3 509 0
I put Muse on my smart glasses! @natfriedman @alexandr_wang
That's the real lens, filmed through the glass. Tap the temple, ask out loud, and my Muse answers in front of my eyes!
No ESP32 needed, my Mac plays the gadget. Love how hackable this SDK is!
That's the real lens, filmed through the glass. Tap the temple, ask out loud, and my Muse answers in front of my eyes!
No ESP32 needed, my Mac plays the gadget. Love how hackable this SDK is!
0:08 21 10 3 105 46.7K 47
The pelican riding a bicycle test, but as a Roman floor mosaic. 8,785 tiles, laid one by one by Claude Code. PELECANVS IN BIROTA! @simonw
0:27 3 1 0 50 2.5K 19
The line between filmed and rendered is getting blurry. This mosaic doesn't exist. It's 127,445 tiles, each its own 3D object, placed one by one in code by Claude Code with Opus 5.5. Starts inside her eye, ends on the pearl!
0:32 14 42 4 489 61.1K 217
AI could never
0:24
Pope Leo XIV@Pontifex · Oct 2In this era of artificial intelligence, it is becoming urgent to distinguish human art from what machines produce. There is an ontological difference, even before an aesthetic one, between art and what a machine can generate through statistical calculation based on millions of images created by others. Algorithms lack the spark of humanity. For this reason, the Church wishes to renew an alliance with artists and cultural institutions to safeguard our humanity.
Open quoted post → 1 0 0 8 654 4
Great one shot explainer of quantum teleportation with Claude!
3:26
Andrej Karpathy@karpathy · Oct 2We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
Open quoted post →
0 0 0 4 960 0
Got Claude to lay a Roman mosaic piece by piece, and it looks real! It starts with the eye, 9 tiles for the pupil and a ring of gold, then 7,446 tiles later there's a whole duck. All code in the browser, no image or video model.
0:24 20 38 3 522 32.6K 280
Just got Claude Opus 5.5 to read two intercepted cipher letters that sat unread for 433 years!
Paris, 22 July 1593. A Catholic League insider writes to Rome about the Protestant king: "and on Sunday he is to go to Mass".
Three days later Henri IV did exactly that 🧵
Paris, 22 July 1593. A Catholic League insider writes to Rome about the Protestant king: "and on Sunday he is to go to Mass".
Three days later Henri IV did exactly that 🧵
1 0 2 12 3.4K 2
Saw Opus 5.5 build a LEGO Microduck and wondered what the 2,000 year old version would look like.
Asked Claude Code for a Roman mosaic instead.
It's 7,878 tiles, every one a real 3D stone, laid row by row the way a mosaicist would...
no image model, no video model, just code running in my browser. Zoom in on the eye!
Asked Claude Code for a Roman mosaic instead.
It's 7,878 tiles, every one a real 3D stone, laid row by row the way a mosaicist would...
no image model, no video model, just code running in my browser. Zoom in on the eye!
0:24 30 49 7 708 50.1K 524
Update to this benchmark: I reran @ashxhart’s TensorFold 0.6.0 on the same DGX Spark, Qwen3.8-Flash-Next NVFP4 weights and MTP-6. Versus 0.3.6.3: +5.1% decode and 26% lower cold TTFT across 10 contexts. At 260k (one run): 430s → 208s cold; cached resend 0.52s. Chart ↓
Paolo Rosson@redp314 · Oct 1Tested @ashxhart's TensorFold v0.3.6.3 on one DGX Spark with Qwen3.8-Flash-Next (NVFP4, MTP-6).
Fixed 2,048-token code task:
74.8 tok/s single-stream decode,
313 tok/s across 8 streams.
Context sweep: 56–65 tok/s through 132k input;
50.8 at 260k (one run).
Cold TTFT at 260k: 430s.
Fixed 2,048-token code task:
74.8 tok/s single-stream decode,
313 tok/s across 8 streams.
Context sweep: 56–65 tok/s through 132k input;
50.8 at 260k (one run).
Cold TTFT at 260k: 430s.
Open quoted post →
0 0 0 4 687 0