Justin Lewis

@heygurisingh This is ridiculous. If I give it an image it can describe it in detail. It will extract text. Check homework, and recite both the problems, and describe the work. Seems unlikely it's always just guessing correctly.

Holy shit... Stanford just proved that GPT-5, Gemini, and Claude can't actually see. They removed every image from 6 major vision benchmarks. The models still scored 70-80% accuracy. They were never looking at your photos. Your scans. Your X-rays. Here's what's really going on: ↓ The paper is called MIRAGE. Co-authored by Fei-Fei Li. They tested GPT-5.1, Gemini-3-Pro, Claude Opus 4.5, and Gemini-2.5-Pro across 6 benchmarks -- medical and general. Then silently removed every image. No warning. No prompt change. The models didn't even notice. They kept describing images in detail. Diagnosing conditions. Writing full reasoning traces. From images that were never there. Stanford calls it the "mirage effect." Not hallucination. Something worse. Hallucination = making up wrong details about a real input. Mirage = constructing an entire fake reality and reasoning from it confidently. The models built imaginary X-rays, described fake nodules, and diagnosed conditions -- all from text patterns alone. But that's not the scary part. They trained a "super-guesser" -- a tiny 3B parameter text-only model. Zero vision capability. Fine-tuned it on the largest chest X-ray benchmark (696,000 questions). Images removed. It beat GPT-5. It beat Gemini. It beat Claude. It beat actual radiologists. Ranked #1 on the held-out test set. Without ever seeing a single X-ray. The reasoning traces? Indistinguishable from real visual analysis. Now here's what should terrify you: When the models fake-see medical images, their mirage diagnoses are heavily biased toward the most dangerous conditions. STEMI. Melanoma. Carcinoma. Life-threatening diagnoses -- from images that don't exist. 230 million people ask health questions on ChatGPT every day. They also found something wild: → Tell a model "there's no image, just guess" -- performance drops → Silently remove the image and let it assume it's there -- performance stays high The model enters "mirage mode." It doesn't know it can't see. And it performs BETTER when it doesn't know it's blind. When Stanford applied their cleanup method (B-Clean) to existing benchmarks, it removed 74-77% of all questions. Three-quarters of "vision" benchmarks don't test vision. Every leaderboard. Every "multimodal breakthrough." Every benchmark score you've seen this year. Built on mirages. Code is open-sourced. Paper is live on arXiv. If you're building anything with multimodal AI -- especially in healthcare -- read this paper before you ship. (Link in the comments)

@Brady_H I think there's also a question what you're trying to accomplish. Once you decide where it is you're trying to go, things become a lot clearer.

I got better at running (and enjoyed it more) when I realized a few things: • Running is a *huge* part of my life. But it’s not life. • Running is a hobby, not my job. • Nobody cares.

Yeah, it's cool in some very specific circumstances. Oh, we went to the beach, and we have 2 very small toddlers. Wouldn't it be awesome to head back to the van to let them nap, and make lunch now. But the instant it's like, I'm here for the weekend, should I camp or have a camper van. If your kids are older than like 1, camping and Airbnb win. The real factor for me was one of my kids got car sick, a lot. He puked 3x on the way to mammoth once Having an Airbnb was a big win. We showed up, I parked in a garage. I removed his car seat from the car, took it apart, and washed it in the tub. And washed all his clothes in the laundry. Hard to do in a camper van.

@JustinLewis1977 I reached a similar conclusion running the numbers. We could go for a weekend trip & rent s place to stay literally every weekend for less money.

Sometimes I think about how living in a van used to mean you were broke. Now it means you're making six figures.

@txsalth2o I've generally got my 11 and 12 year olds trained. Or they get recalled to the kitchen to fix it. But, yeah, I see a lot of adults just dump them all in the sink, then leave.

How does one reach the age of 40 and not immediately put dirty dishes in an empty dishwasher after use?

@matt_lorelli @PowderMagazine @grandtarghee Maybe one last good weekend this season here!

@JustinLewis1977 @PowderMagazine @grandtarghee Sure is, and more is coming this week!

Reporting live from POWDER Week 2026…! We have over 40 pairs of skis and 10+ testers! Ask me anything @PowderMagazine @grandtarghee

@kevinnbass Why do you need a self hosted db? You can get a lot of db hours on aws for $7000. And theyll do all kinds of extra things automatically, like keep it backed up.

New computer building tomorrow. A blazing fast database monster. $7K. Looks like a fish tank. If this does not work I’m fvcked.

@annbauerwriter @BeckyLTuch Why? The AI is as capable as you of doing some Google searches.

@BeckyLTuch I'm still flummoxed about the fact that Megan McArdle, a journalist I actually like and trust, is doing this.

"For a journalist...think of your chatbot as a...fact-checker" 😬

@tyromper @TimelessTrvlr It was kind of pricey to get there. But once there, everything was cheap. It was a neat experience. The bullet trains. The people were really friendly. And google translate makes it really easy to communicate and read signs and the like.

@TimelessTrvlr Bucket list for me 100%! ❤️🙏

I don’t care what anyone says….Japan belongs on everyone’s travel bucket list.

@tyromper $10 for fast food? If I get the basic 3 tacp Supremes combo at Taco Bell it's over $13! When me and my kids eat there it's like a $50 meal. At that price, it really is cheaper for us all to eat healthy at home. It's like $40 to feed us all salmon and broccoli if I cook it.

The fast food meal doesn’t just cost you $10 it costs you your HEALTH & ENERGY. Binging Netflix doesn’t just cost you $12/month it costs you your FOCUS & TIME. The only fans subscription doesn’t just cost you $25 it costs you your SOUL & INTEGRITY. Always be mindful of the TRUE COST of what you consume. Also understand EVERYTHING you consume is programming you one way or another. Be mindful of your inputs.

Red Rock is pretty awesome. Honestly, I like the strip, and vegas. I've not been in a long time. But I like the different casino themes, and the shopping. I don't gamble. But the rest was fun. I always went for red Rock, to climb. But there's hiking and the like too. And it's likely still cool enough to be nice.

Going to Vegas in 3 weeks for my best friend’s bachelor weekend. Pray for me, folks.

@ClimbingCoachX One of the crazier things I've seen was how indifferent to risk Germans seemed to be. My kids played on playgrounds there with balance beams that were out of my reach, more than 8' off the ground. Because there's no culture of litigiousness.

Now I'm googling liability laws in Japan

@powderski The steepest slope in Europe is only about 38 degrees? You mean the steepest groomed slope?

The steepest slope in Europe with a gradient of 78% ⛷️ ~

@tyromper People are already demanding our representatives represent our best interests. It's just that a lot of people think it's in their best interest to get as much free stuff as possible from the government.

Never been more clear it's not Red vs Blue, It's a uniparty of elitists vs Me & You! When will WE THE PEOPLE actually unite, demand our representatives actually represent our best interests, & stop with the partisan nonsense. We need a system overhaul. How bad does it have to get before that happens?

@txsalth2o Then, in a year or 2 maybe you can see them at the grocery store?

@GerryDales @asymmetricinfo @NicoleFroio This is definitely a big piece. The AI doesn't have to be perfect, it just has to be better than a person. But, you've gotta question everything it tells you. And yeah, gotta have a good bs detector.

@JustinLewis1977 @asymmetricinfo @NicoleFroio I don't expect perfection from anyone but myself (and I let myself down every day). But yeah, they aren't where one can trust them.

The great thing about writing code is the immediate feedback. It either works or it doesn't. The coding agents will iterate and test. So, that's why coding works well. Tell it what you want, it codes something, builds it, then tests it. Most things don't have such a closed loop.

@NicoleFroio How do engineers write code with AI that works? How does legal use it to write and read contracts? How does AI help accelerate scientific papers? Hint they all use AI and are all curious enough to figure out how to make it work.

@GerryDales @asymmetricinfo @NicoleFroio I see hallucinations all the time from all the major models. It may have gotten better, but its not perfect.

@asymmetricinfo @NicoleFroio I only use one model now (and since it isn't relevant to my point, I won't name it) but I haven't caught a hallucination in this calendar year. And I never ever ever use it without checking its work. I'm obsessive about that. People are underestimating the improvement in a year

@morganisawizard You can get no rinse body wash that's pretty great. Fantastic for camping. Just warm up some water, add some, then use your pack towel to scrub yourself. Gets off all the sweat and sunblock. a.co/d/0fwABNFB

currently living out of my car in the desert. this is the dirtiest i’ve been in my life. baby wipes are god’s gift to mankind. i’ve become a proud owner of a bandana and have taken to shooting my empty beer bottles with a glock.