@dwarkesh_sp I was waiting for a new episode. Looking forward to listening to this @dwarkesh_sp.

Beats Claude Sonnet and 4o on most benchmarks while being fully open source. 🤯

@GaryMarcus The @arcprize benchmark includes training and test splits. Models are always ‘trained’ on part of the ARC data. That’s the official protocol, ‘zero-shot’ was never the requirement.

@JustinLin610 👀

Here’s a link:

Tinkering on @Replit is my new hobby. Can’t wait to turn more random ideas into functional mini-apps.

RT @SmokeAwayyy: o3 scored 75.7% on ARC-AGI Public Benchmark within compute requirements. o3 scored 87.5% in High compute mode. Human per…

@CGlideHurst @aapmHQ @kkbGoBlue @JulieLarkin305 @leigh_conroy This was a great webinar! Thanks for sharing your insights, @CGlideHurst.

@demishassabis Congrats @demishassabis!

RT @alexandrarink2: I am looking to grow my lab! Looking for an enthusiastic postdoctoral fellow with a solid chemistry background. Full po…

@kimmonismus

@Dr_Singularity Incredible science here—mLumiOpto disrupts mitochondria to kill cancer cells, showing big promise in glioblastoma and TNBC. But saying ‘all cancers cured by 2030s’ is a stretch. Preclinical results are exciting; let’s see if it holds up in humans.

@khashifattah @ACRORadOnc Congrats @khashifattah!

@kimmonismus I think the real disconnect is that everyday Twitter users are asking o1 for surface-level tasks, so they don’t see the magic. Meanwhile, scientists and scholars tackle deep, specialized problems.

Cancer isn’t a single disease waiting for a quick ‘cure,’ it’s a complex spectrum with evolving targets. While advances have cut death rates, no credible consensus suggests researchers are about to abandon oncology for better ‘ROI’—they know the path to conquering cancer is still long and demanding.

RT @Tkaraletsos: Our AI for virtual cells article is now published in @CellCellPress with lots of amazing co-authors who gathered with us a…

@OfficialLoganK 👀

I’m not convinced it’s as hard as you think, @joshuapliu. The patterns and reasoning you’re calling ‘mental models’ are often right there in our communications. With the right training, an AI could piece together these subtle cues, closing the gap between simple templates and real decision-making far sooner than we might expect.