WT 30 YEARS 🖤 Two iconic live records, back on limited colored vinyl - 1000 numbered copies each. 🟠 Black Symphony (half orange/black) - the legendary 2008 Ahoy show w/ Metropole Orkest + Pa’dam Choir 🔴 An Acoustic Night At The Theatre (half red/black) - all the hits, unplugged Out 13 Nov. Pre-order now:

🖤 WT 30 YEARS - where the breakthrough began. Mother Earth returns, pre-order today: 🟢 Half green/white vinyl 🟠 Half orange/white, WT Shop exclusive 🔴 Mother Earth Tour (Live), half red/white 💿 Expanded 2CD 1000 numbered copies per vinyl 👉

$SIVE $AAOI $LITE $COHR WT Microelectronics did US$37.8bn of revenue last year selling chips and components to more than 25,000 customers, and today it added lasers. Sivers says it's WT's first laser line card. When a distributor that size opens a new category, it's betting a lot of different buyers will want the part, and for CW lasers that's a bigger signal than any single order. The deal is a global distribution agreement covering Sivers' InP lasers and its wireless beamformer chips. WT's marketing head named three markets for them: AI data center connectivity, satellite communications and next-generation wireless. Sivers' CEO says WT brings demand creation and the ability to handle fulfilment as production ramps. The customer count includes Future Electronics, which WT bought in 2024, so the network runs across Asia, the Americas and Europe. We think the timing has more to do with how much Sivers is about to be able to make than with how much it sells today. Q2 net sales were SEK 53.8m. Meanwhile it has WIN Semiconductors as its volume foundry for DFB lasers, added a second, unnamed foundry partner in August, and on Sept 3 committed $30m to expand its Glasgow fab to more than 100 million CW DFB lasers a year from Q4 2027. A company with Sivers' sales team can't put an engineer in front of every module maker that might use those lasers. WT already has the relationships. Where those buyers sit matters too. Silicon photonics modules at 800G and 1.6T use CW lasers as the light source, and a lot of that module assembly happens in Asia, which is WT's home market. Sivers was at CIOE in Shenzhen in September pitching 100mW lasers for exactly those modules. What the release doesn't give: a dollar figure, a minimum purchase, or any exclusivity. Distributors mostly reorder as parts sell through to their own customers, so this becomes steady revenue only once module makers buy in volume, and most of Sivers' AI laser volume carries 2027 dates. Sivers is targeting Jabil production orders in H1 2027 and the ramp in H2. WIN, according to Taiwanese media coverage of its Oct 28 call, expects CW lasers to bring pre-production revenue in 2027 and contribute in size in 2028. The nearest laser revenue is LiDAR, where Sivers announced production orders on Oct 20 with shipments starting this quarter. The market is still cautious. Even after today's jump of about 11%, the stock is close to half its Sept 24 close of SEK 32.78. Sivers reports Q3 on Nov 26. We'll be looking at product revenue against engineering services, operating cash flow against Q2's SEK -70.0m, whether the Jabil beta builds are still on for Q4, and whether management says anything about how WT orders will show up in the numbers. What would change our view: if Glasgow or the Jabil ramp slips past 2027, the channel would be ready before Sivers has the volume to put through it. The piece we can't see yet is how much of Sivers' 2027 laser output goes through WT and how much goes direct to a few large accounts. What are module makers hearing?

Netlify’s first meetup in Malaysia is happening in Kuala Lumpur on January 20. AI Shipper @zeng_wt is hosting a live workshop on AI Gateway and Agent Runners, and after that, the room opens up for hack time and demos. RSVP is required for entry.

I think we should literally stop calling every linear model as an SSM. Mamba2: Sₜ = αₜSₜ₋₁ + kₜvₜᵀ GDN: Sₜ = αₜ(I − βₜkₜkₜᵀ)Sₜ₋₁ + βₜkₜvₜᵀ GDN-2: Sₜ = (I − kₜ(bₜ⊙kₜ)ᵀ)DₜSₜ₋₁ + kₜ(wₜ⊙vₜ)ᵀ GDN family is a gradient step on a local regression loss, not a discretized ODE. Umbrella term should be "linear RNNs", with SSMs as one sub-family.

Thirty years. Two worlds. One drop. 🖤 To mark three decades of our music, two of the albums that started it all are back as limited 'WT 30 Years' anniversary vinyl - get yours now.👇 🌊 ENTER (1997) - half aquamarine/white, 4-page booklet 🔵 THE SILENT FORCE (2004) - half blue/white, 60×60 poster 1000 individually numbered copies of each. Once they're gone, they're gone. 👉

FOLFOXIRI + cetuximab may not be as crazy as it sounds. But it is every bit as toxic as you'd expect. 🚽 For anyone wondering what might happen if we combined ALL THE THINGS in MSS mCRC, we have the randomized phase II DEEPER: 359 patients with 1L RAS WT mCRC mFOLFOXIRI + cetuximab vs mFOLFOXIRI + bevacizumab Overall: no PFS or OS benefit. Cetuximab with deeper responses: 57.3% vs 46.0%. And in the exploratory left-sided, RAS/BRAF WT subgroup (bear in mind ~30% didn't have known BRAF status): ▪️PFS: 14.8 vs 11.9 months, HR 0.71 ▪️OS: 50.2 vs 40.2 months, HR 0.74 Patients with extrahepatic disease seemed to do better and the effect washed out with liver limited disease: OS 50.2 vs 38.6 months, HR 0.60 Great? But the downside; TRIPLETE already showed us that adding irinotecan to FOLFOX + panitumumab didn't help. What it did do is cause Grade ≥3 diarrhea > 20%. 🚽🚽🚽 So no, triplet + anti-EGFR is not a new standard. But potentially it keys in on some interesting disease biology. And perhaps, once in a blue moon, if someone is feeling exceptionally aggressive with a grossly metastatic, YOUNG, left-sided RAS/BRAF WT patient? Well, maybe... @TheGutOncLab @OncoAlert @Onco_Nexus

🧊 Turning one image into 3D used to force a choice: "accurate on the visible surface but no backside" or "complete but misaligned with the input." World Tracing stacks 3D points per pixel into layers, capturing visible and hidden surfaces at once. Title: World Tracing: Generative Pixel-Aligned Geometry Beyond the Visible URL: 🔍 Overview World Tracing represents geometry as an ordered stack of L camera-space 3D points per pixel. Layer 0 is the visible surface, deeper layers record front-to-back intersections with surfaces hidden behind the foreground, unifying faithful reconstruction and generative completion as one layered problem. ❓ Challenges Solved Image-to-3D carried a fundamental trade-off. ・Depth estimators are pixel-accurate but stop at the visible surface ・Generative 3D models are complete but work in canonical frames, so they misalign with the input World Tracing frames this as faithful generation: accurately reconstruct the visible surface while plausibly generating the invisible. 💡 Methodology & Proposed Approach At its core is WT-DiT, a 1.7B-parameter diffusion transformer. ・Three-way factorized attention (layer-wise, ray-wise, global) preserves depth ordering and front-to-back coherence ・A mixed noise schedule handles the asymmetry between layer 0 (image-constrained, reconstruction-like) and deeper, generative layers by varying noise per layer ・Mix-training lets multilayer (3D assets) and single-layer (RGBD photos) supervision train together 🎯 Use Cases ・Text-driven 3D scene editing (training-free closed-form compositing thanks to pixel alignment) ・Geometry-conditioned novel-view video synthesis using complete hidden geometry as memory ・A TRELLIS hybrid that yields faithful meshes which reproject correctly to the input 📊 Experimental Results It outperforms prior work on object, scene, and dynamic benchmarks. ・Object visible-depth MAE 0.0149 (VGGT 0.0257) ・Complete-shape F-score@0.05 0.549 (TRELLIS 0.204) ・Scene MAE 0.0102, and best dynamic-clip Chamfer L2 at 0.0105 #3DGeneration# #ComputerVision#