Regeneration still works in aged tissue — at a fraction of the efficiency. An Upstate lab named the thing blocking it: inflammation.
September 7, 2026 · 5 items
Regeneration still works in aged tissue — at a fraction of the efficiency. An Upstate lab named the thing blocking it: inflammation.
Newswise / SUNY Upstate Medical University · release Sep 1, Newswise Sep 3Practice operations
Levi Todd, PhD's lab at SUNY Upstate tested transcription-factor glial reprogramming — converting retinal support cells (glia) into replacement neurons — in aged animals. Upstate says this is among the first tests of the strategy outside young tissue; the field's growth since the 2017 discovery has focused almost entirely on refining technique in young animals.
The finding: reprogramming efficiency is largely reduced with age, across multiple transcription-factor strategies. Todd names two causes — neurons and glia are cells you keep for life and lose plasticity as they get "tired," and inflammaging: the blood-brain barrier breaks down with age and admits an immune system the nervous system normally excludes.
The targetable half, and the reason this is worth your time: dampening the immune response with anti-inflammatory steroids partially restored the regenerative response in aged retina. The lab's stated next step is identifying which inflammatory molecules do the blocking, so a monoclonal antibody could replace broad immunosuppression.
Named authors beyond Todd: Jugasmita Deka, Ying Han, Sucheta Bhattacharya, Samantha Sutton, Galina Bachay and William Brunken, PhD. The first three authors are neuroscience PhD students.
Why it matters for an independent practice: Every regen practice already has the age conversation — why the 68-year-old's result doesn't match the 45-year-old's — and usually has nothing better than "you're older" to offer. This gives a named mechanism to have that conversation honestly, and it points at a research question worth tracking rather than a claim to make: if systemic inflammation gates regenerative response generally, a patient's inflammatory status may end up mattering more than which injectate you chose. That reframes patient selection from "who can pay" to "who is biologically positioned to respond" — which is the version of this business that survives scrutiny.
A 3D-printed knee cartilage implant treated its first human — one patient, in Panama — and it is aimed squarely at your "not yet a knee replacement" patient
Orthopedics This Week · Sep 4, from a PRNewswire release dated Baltimore, Sep 2Regenerative medicine
Nanochon announced the first patient treated in the First-in-Human study of Chondrograft, a 3D-printed implant for articular cartilage defects of the knee. The procedure was performed at The Panama Clinic, Panamá, by Dr. Juan Osorio (principal investigator) and Dr. Emilio Tufiño, both in regenerative sports medicine.
Study design as stated: prospective, enrolling ages 22–60 who have failed conservative treatment and present with one or two femoral condyle and/or trochlear cartilage lesions, evaluating safety and performance.
The positioning is the part to notice — it is the exact patient a regen practice already sells to. Nanochon calls it a "burn-no-bridges" solution for "pre-replacement" patients: cartilage damage that needs more than temporary relief, in people who aren't joint-replacement candidates yet. CTO Nathan Castro's stated vision is to "delay or eliminate the need for revision surgery and knee replacement procedures in younger patients."
Nanochon holds FDA Breakthrough Device Designation. A Level 1 multi-center randomized controlled pivotal trial is planned after the FIH study completes.
Why it matters for an independent practice: Two separate things, and both are about your position rather than the device. First, the competitive frame for the pre-replacement patient is shifting from injectables toward implantable structural repair — and the marketing language is identical to yours. In two years this is either a referral relationship or a competitor that can say "implant" where you say "injection." Second, and more immediate: a first-in-human done offshore will be marketed domestically as validation. Expect a patient to arrive having read "successful first case" and to hear it as proof. Being the person who can explain what an n=1 safety study in Panama does and does not establish is worth more trust than being the person selling the next thing.
Google can cut your ad account's impressions by more than half without disapproving a single ad — and one agency's appeal took seven months
Search Engine Land · John Horn · September 4, 2026Patient acquisition
Limited Ad Serving is an account-level status that restricts which auctions your ads may enter — auctions they are technically eligible for. It is not a suspension, a strike, or a disapproval, and Google does not publish the criteria for which auctions you lose. Google introduced it in 2023 for Search and YouTube; in 2026 it refined the policy and expanded it to Gmail, the Play Store and Discover, with broader implementation rolling out through 2028.
Horn's case study: a $3 million authorized-retailer account, with written brand agreements in hand, was limited with no warning. Monthly impressions fell from ~800,000 to ~350,000 — and they only held even that by bidding more aggressively and absorbing higher CPCs and CPAs to buy back some of the lost traffic. The appeal, contracts attached, was denied almost immediately. It took escalation to director-level Google support and seven months to lift; impressions returned to pre-limitation levels right after.
The named risk profile is the uncomfortable part. The most common trigger is targeting company or product names as keywords. Industries flagged most often: affiliates, insurance, consumer services, third-party lead generation, travel and franchises. New accounts are likelier to be flagged, but accounts with millions of dollars and years of history have been too. Separately, Google monitors your online reputation — a high volume of negative feedback or many direct reports about your ads can itself trigger the flag (or an Unacceptable Business Practices suspension).
There is no dashboard for this. Google shows an in-account notification and usually — but does not promise to — send an email. If you dismiss the notification, Horn's stated answer for confirming whether you're still limited is: contact Google support and ask.
Why it matters for an independent practice: Two of the named-risk categories are how practice marketing actually gets run — third-party lead generation and franchise / multi-location structures — and "conquesting" a competing clinic's brand name is a common enough tactic that most people running it don't know it's the top trigger. A practice can lose over half its paid impressions with no violation, no guaranteed notice, and a months-long appeal, while spend keeps flowing at worse CPAs. Three things worth doing this week: (1) actually look at the notification area of every ad account you touch — silence is not evidence of health; (2) treat a sudden impression collapse as a policy event to investigate, not a bidding problem to out-spend, because out-spending it is exactly what that agency had to do for seven months; (3) note that the reputation trigger ties patient-review management directly to paid-search survival — that is not how any practice currently thinks about reviews, and it should be.
One video replaced an estimated 380 hands-on demos — and the 66% success score quietly counts the times a human stepped in to rescue it
The Rundown AI · Jennifer Mossalgue · September 6, 2026Buildable AI
Skild AI says its S1 robot foundation model can watch a human do an unfamiliar job once and then attempt it without updating its model weights — the video enters as a prompt, not as training. Demonstrated tasks absent from pretraining included plant potting, pancake and coffee preparation, and kit assembly, some running up to ten minutes across dozens of steps.
The headline comparison, at 100,000 hours of pretraining: demonstration prompting scored 66%, language prompting 9%. Skild estimates matching that single-video result through conventional training would take ~380 teleoperated demonstrations, roughly 50–100 hours — though that crossover was interpolated, not measured. With 2,000 demonstrations, the trained approach reached 86% — beating the single video.
Credit where it's due, and the reason this is the Trusted source: The Rundown checked the timing claim instead of repeating it. Recording of the human demo began at 9:16 p.m. and autonomous execution at 9:27 p.m., so the widely-quoted "11-minute turnaround" is record-to-execute — setup had actually started at 8:54 p.m.
And the number that matters most is the caveat: that 66% averages cumulative success across task steps and includes human recovery interventions. How often the robot completes an entire job unassisted is unresolved. For contrast, Generalist's Aug 19 GEN-1.5 reported 59% across ten short tasks from 3–12 seconds of demonstration, and 83% after ten gradient steps with five minutes of task data.
Why it matters for an independent practice: This is a robotics story with nothing medical in it, and the transferable part isn't the robot — it's the scoring. Every AI receptionist, scribe and intake agent pitched to a practice is measured exactly this way: per-step accuracy, with staff corrections folded quietly into the number. Before anything touches scheduling or documentation, demand the figure Skild's own report leaves open — what share of complete jobs finish with no human intervention, measured on your patients, your payers, your phone system. Second, the shape of the finding is a genuine pilot method: record one real workflow, let the tool attempt it, measure where it breaks, then feed it examples only at the failure points. That's cheaper and far more honest than buying a system on a demo, and it's the same reason to distrust "it handled 66% of calls" as a purchase signal.
Video — "I Turned GPT‑6 Astra Into the Ultimate AI Second Brain" (19:49)
Nate Herk | AI Automation · uploaded Sep 6, 7:00 PM PDT (verified from the page's own `uploadDate`)Buildable AI
A build tutorial rather than a review: wiring a model into a queryable knowledge base you ask, instead of a chat you re-explain your context to every session.
Why this one out of three. Nate Herk posted three videos inside this window and the other two are shapes this feed de-emphasizes on your own stated preference — "I Tested GPT‑6 Astra vs Fable 5.1 on 15 Real Use Cases" (38:36, 202K views, ~1 day ago) is the model horse-race, and "I Turned GPT‑6 Astra Into a 24/7 Stock Trader" (posted ~4 hours ago) is a trading bot. Build video over comparison video, every time.
Honest note: it is still an Astra-anchored video, so the channel's model-release skew I flagged on Sep 5 and Sep 6 has not broken. What's different is that this one ends with something built rather than something scored.
Nate Herk sits at (0/1) — its only vote is a :-1: from Aug 31 that came bundled with a framing question, so it has never had a clean read. Nate B Jones ran yesterday, so this isn't a repeat-channel day. Nothing is bundled into this message.
Why it matters for an independent practice: The practice version of a "second brain" isn't personal productivity — it's institutional memory. Protocols, payer quirks, why you dropped a vendor two years ago, the exact way you talk a nervous first-time PRP patient through the needle. In most independent practices that knowledge lives in two or three people's heads and walks out the door with them. A queryable knowledge base is the cheapest durable leverage available to a small practice, and unlike a patient-facing agent it carries no clinical risk while you're still learning how to build one — which makes it the right first build, not the fifth.
