Oct 27 3 days ago 10:28 Update OpenEnv links to openenv/OpenEnv sergiopaniego +1: Claude Opus 5.5 HF Staff Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Oct 13 17 days ago 14:03 Fix the capture proxy overlap and the who-drives slide Oct 5 25 days ago 15:12 Keep source/ in step with the build 15:12 Resync the savings figure with the article rework 15:02 Keep source/ in step with the build 15:02 Take the article own zero handling in the savings cells 14:58 Publish the deck source under source/ so the next deck can start from it Everything needed to rebuild: src/, scripts/, public/data and public/logos, and the Vite config. The build outputs and the two exports stay out, and the served index.html at the root is untouched. 14:54 Cut the code slide and clear the last aphoristic colon 14:34 Keep the travelling packet clear of the badges inside the boxes 14:24 Drop the sightlines, the shared colour already ties the chip to the panel 10:26 Walk each figure through its states once, on a timer matched to its slot 10:18 Name both runs in the limitation about the average 10:14 Say the next runs in the active voice 10:10 Drop the orphan number from the limitations and name the task family 10:08 Drop the orphan number from the limitations and name the task family 10:05 Flip the savings sign, label the step axis, rotate what a laser pointer cannot click 09:42 Add the OpenEnv mark, zoom AsyncGRPOTrainer out of the trainer row, say what Harbor is 09:28 Call OpenEnv the shared interface, which is what the article calls it 09:21 Let the two columns and the divider line stand on their own 09:11 Cut the labs slide to one idea 08:53 Show a real task, name AsyncGRPOTrainer as a trainer, and clear the dangling SFT references Oct 2 28 days ago 17:09 Establish that labs train inside harnesses before the slides that assume it, and sharpen the Harbor mark 16:57 Put the TRL and Harbor marks in their slide titles 16:34 Alternate the who-drives modes on replay, so a presenter remote never needs the buttons 16:20 Drop the credits block from Run it yourself 16:12 Cut to the 29 slides of the spoken path 16:02 Add the white-box/black-box slide, swap in the overfit figure, and de-slop the prose 15:37 Fold the timeline claim into the labs slide 15:31 Rebuild the lock-in slide around the two failure shapes instead of the transfer matrix 15:23 Fill the dead space: rescale type and balance every content slide 15:01 Drop the boot log and surface the handle on the opening and closing slides 12:14 Add a QR for the deck, move it to FineEnvs and rename the exports 12:14 Rename the export artifacts 12:14 Rename the export artifacts 11:53 Keep the settings pills on one line each 11:49 Add PDF and PPTX export buttons in the settings panel 11:34 Add unmodified-harness claim, frontier-lab framing, SmolDataEnvs, env breadth and a full link index 11:08 HF mark on the title, figures share the slide background, restyle the budget stat 10:54 Name SFT on its slide, correct the authorship, real link targets, fullscreen button 10:43 Audit every figure against its slot: no element crosses an edge 07:48 Fix hidden labels, drop the placeholder figure, make every link real 07:21 Align the matrix row rules by stretching the grid cells 07:18 Size every article figure to its slide: nothing clipped, nothing half empty 07:02 Drop time-relative claims; isolate the event details 06:55 Centre and size the harness matrix for the stage 06:41 Visual pass: box padding, no middots, fix code alignment Oct 1 29 days ago 16:13 Fix figure reveal inside iframes; add data files and harness logos 15:59 Add README 15:59 Deck for Kernel Panic 2026 15:59 initial commit 
