The Gayzette

Context for why it is happening now. On April 24, 2026, the FDA handed national priority vouchers to Compass Pathways, the Usona Institute, and Transcend Therapeutics. A voucher compresses agency review of a drug application from the usual ten to twelve months down to one or two. Compass has since reported durable six-month results from a Phase 3 program of 581 patients on COMP360, its synthetic psilocybin, according to trade reporting this summer. Phase 3 is the last and largest trial stage, the one where a drug is tested against a comparison group at a scale big enough to support approval. FDA Commissioner Marty Makary has been quoted saying he expects a decision on at least one of these compounds by late summer or fall 2026.

Which is now. A first-ever federal approval of a psychedelic compound is a live possibility inside this calendar quarter, and the hearing on the 14th is the public's window into what the supervised-session rules will look like on the other side of it.

The honest caveat: an approval decision being expected is not an approval, and the FDA has surprised people in both directions before.

Two pieces of vocabulary carry the weight here. Placebo-controlled means some veterans get an inert look-alike dose so the real drug is measured against expectation itself, which is the whole defense against a result that is really just hope. Treatment-resistant depression means the standard drugs have already been tried and failed, so the comparison is not psilocybin versus nothing, it is psilocybin versus a group of people medicine has already run out of options for.

Worth reading alongside it: a much smaller study out of Ohio State's Center for Psychedelic Drug Research and Education, published in Communications Medicine and covered on July 30. Twelve veterans with severe treatment-resistant PTSD got two psilocybin doses, 15mg and 25mg, with roughly sixteen hours of therapy. Clinician-rated PTSD severity fell from 39.7 to 12.2, and nine of the twelve no longer met the diagnostic criteria for PTSD a month later.

That is a 75 percent remission number, and the caveat is loud: twelve people, and the trial was designed to test safety, with the PTSD result as a secondary finding. Secondary findings are where researchers go looking for hypotheses, not conclusions. Which is precisely what the VA's 240-person trial is now built to test properly. Small promising study, then large blinded study, in that order. The system is working the way it is supposed to.

IONS was founded by Apollo 14 astronaut Edgar Mitchell, and noetic is just the adjective for direct inner knowing, from the Greek nous, mind. So the prize is aimed at the half of the UAP question that the official programs deliberately do not touch: not what is in the sky, but what happens in a person.

The interesting thing is the shape of the ask rather than the subject. Methodologically rigorous, one year, peer-reviewable, competitive proposal, named deadline. Full proposals are due September 22, 2026. The winner is announced October 30, 2026. That is a grant cycle, not a podcast episode, and grant cycles are how a topic moves from anecdote into something that can be checked.

Whether anyone submits a design that can actually survive review is the open question, and it is a hard one. Studying reported inner experience means solving for suggestion, for expectation, for the fact that the person reporting is also the instrument. Any proposal that wins should have a preregistered protocol and blind scoring, or it will produce another decade of arguing.

But the money is real, the deadline is real, and the date is on the calendar.

CRISPR Therapeutics' CTX310 is one infusion. Not a daily pill, not a monthly injection. The edit happens in the liver and stays edited, because the change is written into the cell's own DNA.

Cleveland Clinic's Phase 1 results, led by cardiologist Dr. Luke Laffin, were presented at the European Society of Cardiology meeting and published the same day in the New England Journal of Medicine on August 28. In fifteen patients at twelve months: LDL cholesterol down 52.5 percent, triglycerides down 47.8 percent, no serious adverse events tied to the therapy. LDL is the low-density lipoprotein particle, the one that carries cholesterol into artery walls and starts plaque. Triglycerides are the fat molecules themselves.

Phase 1 means the trial's actual job was safety, and fifteen people is fifteen people. But the number that matters is not the 52.5 percent, it is the twelve. One dose, a year later, still working. Patients are being followed for fifteen years, which tells you what the open question is: durability is the whole product, and nobody yet knows how a permanent edit ages.

The build is worth looking at closely. Ten female dancers across generations, moving to O'Connor's music, with narration drawn from her memoir Rememberings. So the source text is a first-person written record, and the output is bodies in a room. That is a translation problem, not a staging problem: a memoir is one voice, sequential, interior, and a stage piece is ten bodies, simultaneous, exterior. Something has to be decided about what a sentence becomes.

The stated organizing idea is voice, protest, and aging, with the piece refusing to frame getting older as decline. Casting across generations is how that idea gets made structural rather than just said in the program notes. If the oldest dancer on stage is doing something the youngest cannot do yet, the argument is made by the choreography itself and nobody has to narrate it.

Tayeh has spent a career on this exact border, commercial and concert, popular music and formal composition. This one is a full-evening work built on a dead artist's own account of her life, which is about as high a degree of difficulty as the memoir-to-movement problem offers, since the subject cannot be consulted and the audience arrives with their own version of her.

Two things make this worth more than a diary entry. First, it is one of the few remaining front doors that does not require an agent. You upload the thing and pay a fee under a hundred dollars, and a programmer watches it. That is not nothing in a year when most acquisition is relationship-driven.

Second, both deadlines fall inside the next four weeks, which converts this from a headline into a decision. The question is not whether SXSW is a good festival. The question is whether there is a cut of anything Stacy-shaped that could be in a folder by October 1, and if not, what would have to be true for the answer to be yes.

The realistic frame: acceptance rates at episodic tracks like this are low and nobody publishes them, so treat it as a deadline that forces a cut to exist, with a lottery ticket attached. The cut existing is the actual return.

360in360's Living Memories program announced on August 29 that it released its first audiobook narrated entirely by an AI clone of the author's voice: Gadgets to God by founder David Wortley, 14.5 hours, live on Audible UK, with a companion paperback self-published through Amazon. Fourteen and a half hours is the detail worth pausing on. That is a full-length narration, not a demo clip, and audiobook listeners are notoriously unforgiving about narration that drifts or flattens across a long run.

Two honest caveats. This is a single press release from the company that made it, and the operation is small, so nobody independent has reviewed how the narration actually holds up over fourteen hours. Take the capability claim as reported, not confirmed.

The idea underneath it is the durable part, and it is not really about convenience. A memoir is a person's account of their life in their own words. Until now the words survived and the voice did not, unless somebody happened to record it. Voice preservation makes the reading of a life a thing that can outlast the reader. That is either moving or unsettling depending on the day, and probably both.

Squarespace's July product release added a tool for looking at that directly: an AI visibility benchmark showing business owners how, and whether, they show up in ChatGPT and Gemini answers, with competitor comparison alongside. The same release moved Apple Pay, Google Pay, and Link to the top of checkout, auto-enabled globally, and added release-based selling for limited-run offers.

The payments change is the boring one that probably makes more money. Putting one-tap wallets above the credit-card form removes the worst moment in any checkout, which is a person on a phone typing sixteen digits from a card in their other hand.

The visibility dashboard is the one to be careful with. Measuring is not the same as fixing, and no tool can tell you why a model named someone else, because nobody outside the lab knows the retrieval logic. Treat the number as a thermometer, not a diagnosis. What actually seems to move it is the unglamorous stuff a model can quote: a clear plain-language description of what the business does and who it is for, in text, on pages that get indexed.

First, a 1M-token context window, per VentureBeat and other trackers. A token is roughly three quarters of a word, so a million tokens is somewhere around 750,000 words. The context window is everything the model can hold in mind at once during a single pass, its desk rather than its library. Until now, working across a big folder meant the model reading a slice, summarizing, dropping it, and reading the next slice, which is why cross-file connections got missed. A desk this size means an entire project can sit on it open at the same time.

Second, per VentureBeat's reporting, a 75 percent cost reduction on cache reads. Caching means the model stores a chunk of text it has already processed so it does not pay to re-read it next time. Every scheduled job that starts by loading the same vault context re-reads that same material on every run, so a cut to cache-read pricing lands hardest on exactly the always-on jobs, not the interactive sessions.

MarkTechPost reports it scoring 52.6 percent on Terminal-Bench-Science, a benchmark for doing real multi-step technical work in a terminal rather than answering questions about it. Benchmark scores are a rough proxy and worth holding loosely, but the direction is the point.

The practical translation: the nightly jobs got cheaper, and single-pass work over a whole project folder just became possible in a way it was not last week.

The musicians predicted upcoming steps faster and picked up the new sequence quicker. But their memory of the original sequence was exactly as vulnerable to interference as everyone else's. Published in Psychological Research on September 2.

So the honest version of what musical training does is narrower and more useful than the usual claim. It does not make you a person with a better memory. It makes you a person who is faster at knowing what comes next. That is a specific machine, and once you name it you can see why it feels like magic when someone has it: it is what lets a player come in on the right beat without counting, hear where a chord is heading before it arrives, follow a singer who rushes, and improvise, which is mostly the act of predicting a phrase and then choosing to confirm or break it.

It also reframes practice. Drills that make you guess the next thing before it happens are training the actual mechanism. Drills where you already know what is coming are training something else.

On September 4 the FDA granted accelerated approval to camizestrant, brand name Etcamah, for hormone receptor-positive, HER2-negative advanced breast cancer that develops an ESR1 resistance mutation during standard therapy. Unpack that: hormone receptor-positive means the cancer grows in response to estrogen, so the standard treatment blocks estrogen signaling. ESR1 is the gene for the estrogen receptor itself, and an ESR1 mutation is the tumor rewriting that receptor so it stays switched on without estrogen. That mutation is how the cancer escapes.

The new part is the trigger. Tumors shed fragments of their DNA into the bloodstream, called circulating tumor DNA, and a blood test can spot the ESR1 mutation appearing there. This is the first approval where that blood result, not an imaging scan, is what prompts the switch. You change the drug when the cancer starts planning its escape, not when it has already made it.

The number: median progression-free survival of 16 months versus 9.2 months for the control arm. Accelerated approval means the FDA cleared it on a promising measure while confirmatory data continues, so this is a real result on an interim endpoint rather than a final word on overall survival.