This is an email I sent to our leadership team a few minutes ago. We may be witnessing the third major leap in computing productivity, after high-level languages in the 1960s and spreadsheets in the 1980s In the last few weeks, AI coding really took off. Cursor, Cody, Replit Agents are FAR better than GitHub Copilot. Research on ~5,000 devs in Fortune 100 shows that even GitHub Copilot makes them ~25% more productive. ...
This week, I learned: When running a Hello world app: FastAPI takes ~26K RAM, 3% CPU NodeJS + Express takes ~62K RAM, 2% CPU Deno + Express takes ~62K RAM, 1% CPU Deno + Fresh takes ~54K RAM, 0.4% CPU I was testing out different video LLMs: Luma Labs lets you create videos from text Runwal ML lets you create video from an image + text Viggle lets you add images to a video or move a character in a certain way Veed.io is a video editor that offers AI video editing features Deepmotion generates 3D animations from video Wonder Dynamics may be similar to DeepMotion I tested out a few audio LLMs: Suno is fast, has a better UI, lots of examples Udio is slow, poor UI, creates richer music Reflection 70b is one of the top models now, and is open source!. It works by making the LLM reflect on its answer inside <reflection>...</reflection> tags. The best diarization model today is whisperX. Run on Colab T4 GPU with: Scale’s SEAL Leaderboards seem fairly good. coedit-xxl is Grammarly’s fine-tuned google/flan-t5-xxl model run on CoEdit - text editing dataset. It’s mainly for single-line editing, though, and far from a full-document or full-email zero-shot editor.