J Sam🌐

> higher licensing act adding earlier non-radiological tasks into radiological classification > Jevons Paradox > Institutional Radiological procedures are overpriced by referral system - 10000 usd vs 500 usd > new fake radiology fellowship created that encroach into non radiology

people don't take into account is that > US Radiologists offshore work to Indian doctors by teleradiology. We don't know if Indians got unemployed and Americans took the money > US is aging population = higher radiological demand > Health care inflation raising salary anyways

Just updated our paper on on-device LLMs for clinical decision support. Paper: arxiv.org/abs/2601.03266 Here's why I think this matters: We've been asking the wrong question. The debate around LLMs in medicine has been "how accurate are they?", but the harder problem is deployment. Patient data can't leave the hospital. Most clinics don't have the bandwidth or budget for cloud inference at scale. The real question is: can a model that runs locally, on modest hardware, actually be trusted for clinical decisions? After benchmarking 188 models across general disease diagnosis, ophthalmology, and clinical judgment simulation — the answer is yes. Gemma 4 31B (@googlegemma) hits 86.5% on general diagnosis, beats GPT-5-mini, scores 100% on uroradiology and breast imaging, and runs at 18 GB. Qwen3.5-27B (@Alibaba_Qwen ) at 16 GB matches DeepSeek-R1 at 671B, that is one-twenty-third the memory, same clinical accuracy. Fine-tune Qwen3.5-35B with domain-specific reasoning traces and it reaches 87.9%, approaching GPT-5.1 (89.4%). No extra memory. No cloud call. No PHI leaving the building. One thing that surprised me: 87.2% of errors across all models were clinically plausible differentials. the model picked a reasonable diagnosis, just not the right one. Above ~31B parameters, hallucination rate drops to zero. Errors start looking like the kind a careful clinician makes on a hard case, not the kind that would make you distrust the system. There's also a pass@3 upper bound of 93.2% for fine-tuned Qwen3.5-35B. The model already "knows" the right answer in most cases. That's a verifier problem, not a model-size problem. Gemma 4 and Qwen3.5 are the first generation where the local deployment story actually holds up under rigorous clinical benchmarking. That's a real milestone. Huge shoutout to the team who made this happen: Alif Munim (@alifmunim ), Omar Ibrahim, Alhusain Abdalla, Jun Ma @JunMa_AI4Health (all equal contributors), Meng Wei, Shuolin Yin, and Leo Chen from @UHN AI hub. Proud of what this group built 🔥🔥

Billion humanoid doctors

10000 kcal is near impossible

I used to eat 5000+ kcal per day while training for competitive swimming . 80+% carbs. 5k+ daily swimming + other cardio Never ever do that unless you are competitive long distance swimmer You will be obese by 2 weeks Now I am hovering between 1600-2000 kcal

@nikillinit > My back office agent is different and it's not an RCM > it does AI agent nurse, clinic front desk voice agent , inventory management and email social marketing automation > looks inside > it's RCM

“We’re a YC company using agents to handle the back office for private practices” Tier 2 VCs:

I tried this and claude called me a bioterrorist

@DeryaTR_ It's useless but a good informational benchmark on long run as the cost of running agentic medical task for claude/chatgpt is going to be 100-150 USD That's why I use Chinese models in my Dr chatbot apps Ideally we should be comparing cheap vs least censuring

It is a good idea to have AI agent clinic benchmarks; however, you can completely skip the conclusions of the paper on whether AI works or not, all you need to see are all the completely obsolete models they tested. This is a major problem in scientific benchmark papers.

We’ve been grading medical AI like it’s taking a multiple-choice exam. But medicine does not actually work that way. A new npj Digital Medicine paper introduces AgentClinic: a benchmark where AI agents must interview patients, gather missing information, interpret multimodal data, use tools, and make decisions step by step. The results show models that score highly on static benchmarks can see performance collapse in real sequential workflows. Some drop to a fraction of their original accuracy. So, what's the bottom line? The future of clinical AI won’t be about who answers trivia best. It’ll be about who can reason, adapt, and work through uncertainty as real clinicians do. nature.com/articles/s4174…

JUST IN: Skin exams are getting automated. SquareMind just raised $18M to build a robotic system that scans your entire body and tracks every mole over time. • Swan robot captures full-body dermoscopic images in minutes • Tracks new and changing spots across visits • Replaces spot-check exams with total skin coverage • Creates a time-series record for earlier melanoma detection • Plugs directly into dermatology clinics Robotics is going to reshape healthcare.

Bekar kadınlar bekar erkeklere göre yaşamdan ve yalnızlıktan çok daha fazla keyif alıyor; ayrıca eş bulma arzuları da erkeklerden daha az.

In my opinion it takes 3-5 year of joblessness to detach schema of past job/education Blue collar doctors might survive but not much anyone else

I think relationship based on "type of educational degree" and "job" is going to be worst form of assortative mating in coming AI Robot unemployment decade Without job or education small talks, most people have no cue to match or vibe with other

Gold: MedKit by Bedirhan Keskin from Türkiye A voice-based clinical simulator where medical students take histories, order labs, diagnose, and get scored on their reasoning against the latest published guidelines. youtube.com/watch?v=6bN6hn…

GPT-5.5 early access results have been impressive. 25% lift in clinical quality. 30% less verbose. We orchestrate hundreds of AI tasks, some powered by our proprietary data flywheels, others by frontier models like GPT-5.5. We test rigorously for clinical accuracy, completeness, reasoning, real-world because benchmarking is a big deal in healthcare. Grateful to the OpenAI team for the early access and partnership.

@MiguelArratia @jakobdiepen Nuclear energy powered hospitals

Professor Arratia no more. Quit academia for good. ALL IN

See a video tour of the first Autonomous Lab built for the Genesis Mission by @Ginkgo at @PNNLab !🇺🇸🤖 US keeping lead in science needs scientists driving AI+robots instead of toiling at the lab bench!

Figure says they are now manufacturing 55 humanoid robots per week, and that they can now build up to one per hour. I guess this means they are doing roughly one 8 hour shift per day 7 days a week, or maybe 12 hours 5 days a week?

Today we’re giving an update on ramping F.03 production at BotQ In the last 120 days, Figure scaled manufacturing 24x - from 1 robot/day to 1 robot/hour We will manufacture 55 humanoid robots this week