Image created with gemini-2.5-flash-image with claude-sonnet-4-5. Image prompt: Cinematic wide shot of silhouetted person at office window holding vintage brass compass or calibration instrument in sharp foreground focus, adjusting it carefully, while through the glass skyscrapers tilt and crack in desaturated background, teal-cyan shadows with warm sodium vapor light, 35mm film grain, Fincher color grade, shallow depth of field, crushed blacks, architectural destruction at epic scale contrasting with intimate human-scale precision work.
2024: AI is the copilot 2025+: humans are the copilot Copilot is the new engineering skill. It’s not easy to leave the driver seat – we must learn to think the AI way and adapt to the alien workflows. Help AI help ourselves.”” https://x.com/DrJimFan/status/2004633997662716397
Given how rapid, ubiquitous & varied AI use is, we are lacking for gradations in AI slang right now. Slop is too large of a category for all bad AI use (obviously bad prompts & models versus good stuff that tricks you into arguing with a bot, etc.). Also no word for good AI work.”” https://x.com/emollick/status/2004709053034090938
I am reminded of this 1897 account of a visit to Nobel’s Scottish dynamite factory, where everything depended on a man sitting on a one-legged stool (so he couldn’t fall asleep) whose job was to watch a thermometer to ensure the chemical never got too hot: “The surroundings are”” https://x.com/emollick/status/2005930710495056363
Some of the smartest engineering ideas come from tradition: The Twisted Chrysanthemum Joint shows how precision alone can hold wood together tightly. No nails, no glue, just skill. A reminder that true craftsmanship often solves problems we still face today. Why it matters ✅”” https://x.com/IlirAliu_/status/2004476296714149902
Terence Tao: “”In recent weeks there have bee…”” – Mathstodon https://mathstodon.xyz/@tao/115788262274999408
One of the weird impulses that comes from working with AI is providing the LLM with unnecessary”closure” – going back to a chat and telling the AI how what happened & how its advice tuned out. Feels bad to leave the AI hanging, even if that doesn’t make a lot of sense.”” https://x.com/emollick/status/2004835779697426536
On net, is our biological hardware a help or hindrance to our intelligence? Are GPUs are to brains what fingers are to robot motors, and there’s a bunch of cognitive dexterity that our brain is uniquely good at implementing? Ala Morevac’s paradox, evolution has spent way more”” https://x.com/dwarkesh_sp/status/2006164582914756882
Where AI is headed in 2026 – Foundation Capital https://foundationcapital.com/where-ai-is-headed-in-2026/
Recently, LLMs were found to encode different languages in similar ways, a sort of Platonic representation of words. It now extends to science:: 60 ML models for molecules, materials & proteins (all with different training) converge toward similar encoding of molecular structure”” https://x.com/emollick/status/2005991284511760421
A big issue in scientific journals around l AI is that our mental & procedural filters aren’t made to detect the difference between good & bad work quickly when AI is involved A flood of papers is fine if we can separate wheat from chaff. But if doing so takes time, that’s bad”” https://x.com/emollick/status/2005806593724809541
Currently, we train LLMs on the text that humans have generated – aka the behavioral output of our cognition. What if we skipped the middleman and instead trained them to directly predict the brain’s hidden states? Would they learn faster and better? We discuss this idea in”” https://x.com/dwarkesh_sp/status/2006382084735340845
Leave a Reply