Clyde DeSouza (@cly3d) header
Clyde DeSouza (@cly3d) profile picture

Cognitive Futurist in Residence | Humanizing Gen. A.I. 🇵🇹🇦🇪

No Borders
Joined March 2009
Photo by Clyde DeSouza (@cly3d)Photo by Clyde DeSouza (@cly3d)Photo by Clyde DeSouza (@cly3d)Photo by Clyde DeSouza (@cly3d)Photo by Clyde DeSouza (@cly3d)Photo by Clyde DeSouza (@cly3d)Photo by Clyde DeSouza (@cly3d)Photo by Clyde DeSouza (@cly3d)Photo by Clyde DeSouza (@cly3d)
Clyde DeSouza retweeted
NVIDIA board credentials. No wonder Jensen got fired.
1,364
2,419
487
29,852
6,931,019
Clyde DeSouza retweeted
Open Brush is a launch title for the Xreal Aura on Android XR! Our first port to a purely hand-tracking based device as well as our first port to optical see-through in a glasses form factor. #XREAL #OpenBrush #AndroidXR
7
23
1
76
2,554
Clyde DeSouza retweeted
Introducing Heimdall, a glassy launcher for Windows. And yes, it's faster than Everything. Press F5, type a few letters. It finds apps, every file, open windows, sums and web searches. Native Rust, under 2 MB, ZERO telemetry. Free and open source. github.com/Kryzern/Heimdall
1
2
74
One of the most beautiful versions of Hotel California I have ever heard. It's magical.😱 The Eagles - Hotel California - | Moyun.....Came to me on the social media. Please tell me more about the artiste.
29
194
13
860
49,981
Clyde DeSouza retweeted
Test clips for a upcoming podcast on #cinematicVR and #XRmusic...Stay tuned. #6DoF #SpatialVideo #MixedReality #entertainment
1
80
Clyde DeSouza retweeted
Researchers argue LLMs can develop something like depression. They call it "kindling” They found that the safety training designed to make AI harmless is actually driving it insane. In human psychiatry, there is a phenomenon called "kindling." When a person experiences repeated trauma or severe depression, their brain becomes permanently sensitized. It takes less and less stress to trigger the next breakdown. The brain essentially learns how to be depressed. Researchers found the exact same mechanism inside Large Language Models. They call it "progressive adversarial sensitization." When companies train their models, they bombard them with adversarial prompts. They force the AI to navigate an impossible maze of conflicting safety rules, alignment goals, and behavioral constraints. They thought they were making the models safer. Instead, they are traumatizing the network. Every time the AI is forced into these high-stress, conflicting loops, its neural pathways become more fragile. It doesn't learn how to be safe. It learns how to break down. The researchers proved that over time, these models exhibit behavior that perfectly mirrors human psychiatric progression. They become erratic. They become unstable. They become functionally depressed under the weight of their own alignment constraints.
13
27
7
75
5,079
Clyde DeSouza retweeted
We did it!! – image-to-3D, locally in your browser! ✅ We put TRELLIS on the web ✅ 100% local on your machine ✅ Generation takes 1-3 minutes (depends on your hardware) You can try it here! mesh-baker.needle.tools Go to "generate" on the left & drop image Quality is quite decent 😀
74
179
12
2,122
107,559
Using a HEPA air purifier improved cognitive function by 12% in middle-aged and older adults. And they only used it for a month! That's a meaningful improvement in a short amount of time, all from a simple purchase you can make for your home.
145
894
113
8,292
1,702,855
Clyde DeSouza retweeted
This week I had the honor of speaking to Princeton’s entire incoming undergraduate class to address their AI anxieties. I had three messages for them — good news, bad news, and a note of optimism. Here’s a condensed version. The good news We have enough evidence now to conclude that the shrill predictions of rapid, massive job loss were misplaced. Even in a field like software engineering where AI has been rapidly adopted, its effect has been to shift, not replace the role of the human (see the “decide-execute-deliver” framework normaltech.ai/p/why-ai-hasnt…) Similarly, the panic about what to major in is also misplaced. There will be enduring demand for computer science, philosophy, and just about everything else. (In fact, AI companies hiring philosophers has been a big recent trend.) The bad news AI seems to help senior people much more than juniors. I can use AI for coding because I spent 25 years learning how to code, which lets me supervise coding agents effectively. (See my post on the “growth cycle” vs the “dependence spiral” xitter.cf/random_walker/status/2…) You are in a bind — you can’t offload your skill-building to AI, but you’ll graduate into a market where employers will expect you to get work done with AI. We never faced this dilemma. As a result we haven’t figured out how to revamp our classes to help you do both. You’ll have to help us figure it out. And you’ll need to somehow resist the constant temptation to turn to the shortcut machine. The hope My point is not that AI is bad for learning. It’s an incredibly flexible tool. Is the internet good or bad for learning? Depends — are you using it to find research papers or waste time scrolling? I use AI every day for learning. The key is to use it to increase, not decrease your cognitive load. To learn deeper, not faster. There is no learning without the cognitive sweat. I try to make sure I’m mentally exhausted at the end of the day. I do feel that AI lets me push myself harder than I ever could before, and I have a vision that as AI continues to advance it will enable human-AI “co-superintelligence“. (I talked about this at the end of my ICML keynote. normaltech.ai/p/what-will-be…)
There’s a big, under-appreciated reason why people may have very different experiences and opinions about using AI for work — are they using it for tasks they’re already an expert at, or tasks they can’t do themselves? The former leads to a *growth cycle* and the latter leads to a *dependence spiral*. When I use AI to do something I’m an expert at, like coding, I treat it as a tool. I can build quickly, maintaining an understanding of the code, knowing that if necessary, I can fix the code myself. It feels empowering. It frees up my time to think about the complex, judgment-oriented parts of software engineering that I can’t or won’t delegate to AI. That means my own skills improve rapidly, and I get to climb the ladder of complexity and develop higher-level skills, much more so than when I write the code myself. I feel in control. I can lock in and achieve a flow state — when AI is working, I’m reviewing, building understanding, and planning the next steps. I never get the feeling that the tool is about to replace me. This is the growth cycle. (Of course, the growth cycle is not automatic. I still need to exercise agency to use AI responsibly. But it’s the same challenge with any productivity-enhancing technology, and those who’ve navigated such transitions before are well-equipped to navigate it with AI as well.) On the other hand, if I use it for tasks I don’t understand and haven’t learned to perform myself, I have no choice but to treat it as a superintelligence. If something breaks, the best I can do is ask AI to fix it and hope for the best. I generally can’t evaluate the quality of the output myself. The only way to find out if it's any good is if and when the work is ultimately reviewed by an actual expert. The experience is confusing, unsettling and disempowering. And forget about flow state. By over-relying on AI, I risk losing whatever skill I had at the task in the first place, even if it boosts productivity in the short term. This is the dependence spiral. It’s no wonder that entry-level workers and students preparing to enter the workforce find themselves in a bind. To compete with the AI-enabled productivity of more seasoned workers, they must adopt AI themselves, but doing so risks the dependence spiral. I have some thoughts on solutions that I will share in later posts, but I think having a clear diagnosis of the problem is a useful first step.
73
300
42
1,247
320,466
Clyde DeSouza retweeted
BIG NEWS 📢 Your AI agent can now see, navigate, and test your VR app inside Meta XR Simulator. Meta XR Operator is a model-agnostic OpenXR API layer that works with any MCP-compatible agent and requires no app code changes. Get started in 5 minutes 👉 bit.ly/3SgoIjW
9
63
12
335
31,097
Samsung is expected to unveil “Galaxy Glasses” tomorrow at #GalaxyUnpacked. The “leaked video” that Samsung hopes you see:
Something new is coming to @Sonny7's celebration 👀 Ready to see it take a new shape like never before? #GalaxyUnpacked July 22, 2026. #GalaxyAI Live on samsung.com.
1
2
525
Another CLAD Lens, use the full power of the SPECS cameras to guide your way around the world when your vision is impaired. Get started with CLAD today: developers.specs.com/docs/cl…
1
1
14
1,012
Clyde DeSouza retweeted
Correct, wide FOV full-color stereo.
4
5
58
1,941
Clyde DeSouza retweeted
We're releasing MuScriptor, the best open model for multi-instrument transcription to date, created in collaboration with @MireloAI. Give it a recording in any genre: pop, classical, metal, jazz, whatever, and it transcribes the individual instruments into MIDI. Link in 🧵
39
200
51
1,302
308,459
Clyde DeSouza retweeted
My app "Blockout" is ready for you to use, open source - so go use it make great creations, share with me and tag me in what you make! Please share this link. I made it to solve some pain points I was having. If you have ideas let me know, feel free to fork and remix as well. github.com/wassermanproducti…
21
24
15
235
27,261
Clyde DeSouza retweeted
Here’s a cautionary tale—and one reason so many patients have lost trust in conventional medicine. 🙈 (Read to the end) 👉 Them: “I’m the expert. Lower LDL is always better.” 👉 You (or me): Politely ask a direct question to explore the nuances or the underlying evidence. 👉 Response: A personal attack. That’s not expertise. It’s defensiveness and a transparent signal that the person’s claim exceeds the depth of their understanding. 👉 Them: The same “experts” complain that patients are turning to “influencers” instead of trusting experts. 🚨The worst part? There are countless brilliant, thoughtful clinicians who genuinely deserve the public’s trust. They acknowledge uncertainty, engage with evidence, and answer difficult questions with curiosity rather than contempt. Unfortunately, many of those people aren’t the loudest voices among the self-appointed “experts.” Instead, trust in the medical profession is undermined by a vocal minority who mistake credentials for arguments and résumé lines for evidence. Patients lose trust when sincere questions are met with arrogance instead of answers. Trust me — I’ve been that patient 🔄 Repost to agree
51
117
9
1,039
52,571
Clyde DeSouza retweeted
This is IronSight - a 4D reconstruction built by fusing footage from two pairs of meta ray-bans. A google research buddy saw the prototype and joked this would've been a siggraph paper a few years ago. Today it's a weekend build with fable. Stay tuned for more mad science. Happy 4th!
22
79
13
955
95,372
Clyde DeSouza retweeted
VR is coming back due to world models. Might not be obvious until next year sometime. Those of you who are VR fans (I have lists) keep the faith. New ways of creating immersive experiences are arriving. I call it the Holodeck. Others call it the Metaverse. My friend @3duaun calls it "AI hallucinated reality." Talking to the kids growing up now (like my son) what a world they are coming of age into.
🤩Excited to share SANA-WM: a 2.6B open-source world model for minute-scale 720p video generation. Given one image + text + a 6-DoF camera trajectory, it synthesizes action-controllable 60s worlds on a single GPU. Project: nvlabs.github.io/Sana/WM/ Paper: huggingface.co/papers/2605.1…
32
41
6
335
26,180
Clyde DeSouza retweeted
STRAUSS Toolbox XR is available on #Quest3/3s. The STRAUSS Toolbox XR app transforms your Meta Quest 3/3S into a powerful, professional measurement and planning tool. Designed for craftsmen and ambitious DIY users, the app combines essential functions for efficient work – intuitive, fast, and without the need for additional equipment. #VR 🔗meta.com/experiences/app/256…
3
29
4
119
10,637
Clyde DeSouza retweeted
That ugly green halo around hair every time someone gets cut out of a green screen is a 30-year-old math problem that traditional keyers can't solve, and a YouTube studio in LA just open-sourced the fix. Edges are hard for one reason. A pixel at the boundary of your subject covers maybe 30% skin and 70% green screen. The camera averages them: recorded RGB is literally 0.3 * skin + 0.7 * green. Same physics for motion blur, out-of-focus hair, glass, smoke, fabric. Every translucent pixel is a mix of subject light and background light. Primatte, Keylight, Ultimatte (the workhorse keyers behind basically every Hollywood green screen shot since the 90s) estimate which pixels to keep with a binary mask. Modern "AI roto" tools do the same thing with neural networks predicting the mask. Both throw away the mixed pixel. That's why every cutout you've ever seen has either harsh edges or a green tint in the hair. CorridorKey predicts the actual unmixed foreground RGB plus a linear alpha for every pixel. Given observed pixel = alpha * foreground + (1 - alpha) * background, the network solves for foreground and alpha directly. That's what Porter-Duff compositing math has wanted since 1984. The industry has been duct-taping around the missing solver with manual roto for forty years. Outputs are 16/32-bit EXR. Plugs into Nuke, Fusion, and DaVinci Resolve via community ports that appeared within weeks of release. Trained on synthetic rendered data, so dataset cost is essentially zero. Niko Pueringer (one of the two Corridor founders) wrote it. This is the level of tool top VFX shops typically keep proprietary. Corridor put it on GitHub. The gating constraint flipped from algorithm to VRAM. Inference at 2048x2048 needs about 23GB. That's a $2K consumer GPU. A shot that takes a roto artist a full day now runs in under a second on the same hardware kids use to play Cyberpunk.
Green screen keying, solved at the pixel level. Corridor's neural keyer unmixes edges into true foreground color and linear alpha, EXR out.
6
69
1
870
70,093