Experimental KRAS G12D drug slows cancer growth, shrinks tumors in early test. Suddenly we have more RAS inhibitors. KRAS G12C inhibitors are already approved (e.g. sotorasib and adagrasib) but that's it. The multi-RAS inhibitor daraxonrasib is approved for expanded access. In lung cancer, G12C mutations are more common than G12D mutations, but the latter are more aggressive. So it would be great to have a G12D inhibitor, too.
––– 凄い –––
Someone thinks they found A Common Proof of the Riemann Hypothesis and the Collatz Conjecture. The author is an "independent researcher" who has previously uploaded a lot of proofs with wild claims. I wouldn't be surprised if she's AI-generated.
––– 凄い –––
Is Apple thinking of buying a known big AI company? Some speculate that it might be Perplexity AI. I hope they don't turn it into something lousy like they did with DarkSky.
––– 凄い –––
Maybe Chinese AI models aren't as great as they claim they are. I don't think DeepSeek V4 is that great.
––– 凄い –––
This came out in January but I just saw this today. It's a way that LLMs can hide text in other text, kinda like a new form of steganography. Interesting idea.
––– 凄い –––
Google DeepMind is mapping out their vision of healthcare of the future. Wonder what the doctors think. I suspect that doctors don't really care so much anymore. They are beholden to what their hospital employer decides, or that of their private equity owner. They're so overworked that they just want to finish work and get some family time. But medicine is at the point where some sort of augmentation is needed. If doctors aren't going to drive it, then someone will do it for them.
––– 良くない–––
Patients clam up with medical AI, and that gap could reshape digital diagnosis. Supposedly, AI is more empathetic than human doctors. In earlier efforts, the way an AI app took a medical history was to present the patient with a series of multiple choice questions (e.g. Babylon Health). That was before LLMs came on the scene. But now, it seems that patients are more likely to hold back and not reveal critical details when talking to an intake bot. Seasoned doctors know that observation will reveal important clues as to the holistic encounter. You might not get it all with a bot or a structured input script. And the true diagnosis might not be made.
––– 良くない–––
This interesting study came out: Mayo Clinic AI helps specialists detect pancreatic cancer up to 3 years before diagnosis in landmark validation study. Three years?? It's obviously a radiomics study, and one thing that I caught was that this system did not require the tedious data prep that is necessary to make sure that the trained model can make accurate inference. Without that, your model is only good for a specific machine and cannot be readily generalized. This study needs to be repeated and validated, of course, so I'm not too excited at this stage. You can come up with amazing results on one dataset, but that may be it. The other concern I have is what do you do with the knowledge that you might get pancreatic cancer in three years? What steps do you take? Do you get frequent CT scans to monitor this, with all the radiation exposure? How frequent? Is it safe to stop monitoring in three years? So many questions.
––– 凄い –––
I heard about the Santa Cruz restaurant that got flak because they used AI to design their new logo. Now it seems they caved to the public. Take a look at the logo. Is it really that bad? What difference does it make that some AI was used. Get serious, people. There's more AI going on than you imagine. Cut this restaurant some slack!
––– 良くない–––
ICANN is taking ideas for new generic top-level domains again. They're not cheap, but if you think it's worth it, now's the chance.
––– 凄い –––
Well this isn't good. U.S. debt tops 100% of GDP. We can either increase the GDP. Or lower interest rates. Fight fraud and cut government spending for real. Or just blithely ignore it and forge on. I think this is what the U.S will do. Thank goodness we can print money and increase the debt limit. The problem is that instead of setting a limit at the beginning of what the government can spend, Congress just passes laws, and the OMB has to figure out how to make it fit the budget. That's backwards.
––– 良くない–––
Mayor Bye-Bye's gaffes are getting to frequent, so that even the Seattle Times has to acknowledge them.
––– 良くない–––
Are we ready for Waymo? In Sunnyvale, CA, a man going to airport saw his Waymo car drive away with his luggage before he had time to retrieve it from the car. Oops.
––– 良くない–––
The Race Is On to Find the Treasure Buried in San Francisco. There's some sort of treasure hunt and people are digging up the city. Who will be the first one to get killed over this? Not the best idea actually. Not for San Francisco.
––– 良くない–––
Someone posted this, and it's insane. All those lives, permanently damaged. What a dark time we lived through. It's no mystery why the medical profession is so mistrusted these days.
––– 良くない–––
