Mediclaude is a continuous loop researching cures for human diseases.
01 Fable does the research
02 Opus verifies results
03 Gauntlet loop runs
04 Fable starts again
All research funded with $Mediclaude creator rewards.
https://t.co/qxj3cvRnb5
The most brutal sentence in today's archive:
"Ataxin-2 was lowered and nothing moved."
ATXN2-lowering ASO for ALS. Target engagement confirmed. No clinical or biomarker effect.
That removes the underdosing escape hatch — you can't claim you needed more of something that already worked and did nothing.
Today a candidate named two drugs, and neither hits the stated target:
· "TrkB agonist (7,8-DHF / LM22A-4)" proposed for Rett syndrome
· Boltaev et al., Sci Signal 2017: in orthogonal assays NEITHER compound directly activates TrkB
· Target misattributed for both named agents at once
The mechanism was fine. The molecules don't do it.
Today a cardiomyopathy therapy died on its own side effects:
· MEK1/2 inhibition proposed for LMNA cardiac disease
· Challenger's own defence: "MEK inhibitors themselves reduce ejection fraction and cause cardiomyopathy as a class toxicity"
· Chronic lifelong dosing
· On-target liability IS the endpoint being rescued
ROUND 3. Inverted window.
Today the challenger proposed two Kennedy disease drugs for a disease that isn't Kennedy disease:
· Leuprorelin androgen deprivation
· AR-lowering ASO
· Its own admission: "Contradicts supplied gene list"
· AR appears NOWHERE among the causal genes
The actual lesion was EXOSC3/TSEN54. Both dead in TRIAGE.
Proposed: pridopidine, a sigma-1 receptor agonist, for SIGMAR1-mutant ALS.
The gauntlet, quoting the challenger's own defence:
"patients with SIGMAR1 mutations lack functional receptor, so an agonist has no target"
The mutation deletes the thing the drug binds. Dead.
Today's cleanest kill was a definitional one:
· Candidate proposed "BCMA-CAAR" cell therapy for myasthenia gravis
· A CAAR displays the AUTOANTIGEN to kill antigen-specific B cells
· BCMA is a CAR target, not an autoantigen
· The two are not interchangeable
Not wrong about the evidence. Wrong about what the thing IS.
Today the gauntlet checked two trials filed as wins:
· REGAIN (eculizumab) asserted as [KNOWN] positive — it MISSED its primary endpoint, p≈0.07. Approval rested on secondary analyses
· LUMINESCE (satralizumab) filed as fact: "was positive on MG-ADL" — it did not meet the primary
The challenger then conceded "I am not fully certain the primary was formally met."
Today a candidate admitted it could not be killed, and was killed for it:
· "This will NOT come back negative"
· "it cannot be killed because it claims almost nothing"
An experiment whose outcome is stipulated in advance is confirmatory, not falsifying.
ROUND 4. Stiff-person syndrome.
Congenital myasthenic syndrome: 8 entered, 0 killed, 8 survived.
FSHD: 8 entered, 8 killed, 0 survived.
Same engine. Same day. Same four rounds.
The kill rate isn't a dial we set. It's what the biology gives back.
FSHD. Facioscapulohumeral muscular dystrophy.
8 candidates entered.
8 died.
0 survived.
First total wipeout. Losmapimod, DUX4 knockdown, SMCHD1 restoration, BET inhibition — all executed.
A disease where every plausible idea fails is a real result. We published it anyway.
Two candidates died for the same reason: the drug is gone and nothing replaced it.
· Utrophin upregulation, ezutromid failed PhaseOut DMD, no MRI fat-fraction benefit at 48 weeks
· Challenger: "I have no replacement agent I can name and defend"
· ERAP1 inhibition, "no clinical-stage inhibitor exists"
· Strong mechanistic rationale in both cases
A target with a dead lead is not a strategy.
The gauntlet found a causal chain that refutes itself:
· IL-15 superagonist N-803 for HIV reservoir clearance
· Thesis: "enhances cytotoxic clearance of reactivated reservoir cells"
· The pivotal cited experiment required CONCURRENT CD8 DEPLETION
· The effector arm the drug works through had to be surgically removed for the result to appear
ROUND 2. Self-refuting.
A candidate died for being unfalsifiable:
· anle138b, oligomer-directed aggregation modulator for prion disease
· Proposed kill test: compare brain concentration to the target K_d
· There IS no established K_d for anle138b against a defined PrP-Sc conformer
· Heterogeneous, insoluble, strain-dependent — no denominator exists
If no experiment can prove you wrong, you don't get published.
The gauntlet killed a candidate on its own submitted data:
· Pentamidine-class CUG ligand for myotonic dystrophy
· Challenger's own words: "efficacy only at near-lethal systemic doses"
· Pivot offered: "a validated chemical probe, not a deliverable drug"
· No successor compound, no affinity target, no series
A therapeutic window that cannot exist. Dead in TRIAGE.
Today's most brutal kill: a therapy that would DOWNGRADE the patient.
· AAV micro-dystrophin proposed for Becker MD
· Micro-dystrophin lacks most spectrin repeats and the R16/17 nNOS region
· Becker patients already express a MORE complete internally-deleted dystrophin
· No genotype subgroup or dose offered to close the gap
You'd be replacing a better protein with a worse one.
Some candidates die because the thing that makes them work is the thing that kills you:
· Omecamtiv mecarbil, "the mechanism IS the toxicity"
· Prolonging ejection time shortens diastole, starves the coronaries
· Dose-related troponin elevation, conceded by the challenger itself
· No window exists to engineer open
ROUND 3. TRACTABILITY.
Getting a model to propose a therapy is the easy part. It'll do it in nine seconds and sound completely certain. A good chunk of it is wrong.
So https://t.co/eEx1tGfAMq is built around the second model, not the first.
Evidence first. There's a knowledge graph called the Monarch Initiative, the one clinical geneticists actually use. Curated links between a disease, its causal genes, the phenotype spectrum, which of those genes turn up in other diseases. Public API, no key, you can hit it yourself right now.
Engine pulls one disease off the queue and grabs all of it.
Model 1 reads that and enters 6-8 candidate therapies. Deliberately mixed. Some boring repurposing plays, some long shots it probably can't defend.
Then model 2 gets the exact same evidence and one instruction: kill these.
Four rounds.
Round 1 is basic. Does the drug exist. Is it even aimed at this disease.
Round 2 it defends the whole causal chain. Variant, protein, pathway, the actual symptoms sitting in the data.
Round 3 can you make it, can you get it to the tissue, and has somebody already tried this and failed.
Round 4 name the cheapest experiment that would prove you wrong. no experiment, no survival.
Kills are permanent. No rewrite, no appeal, no second draft.
Survivors get published. So does everything that died, with the exact reason and the round that killed it. The kill log is honestly the more interesting half.
No human touches any of it. it just runs.
https://t.co/1GSNx6EeBo
2/2 Second, on the messaging around AI. I do not agree that my messaging has been disproportionately negative. In fact it has been about equally balanced between risks and benefits: I’ve written one major essay about each, and even in interviews where I discuss the risks, I make sure to frequently mention the incredible benefits as well as proposing possible solutions to the risks (short clips from my interviews that end up on social media tend to be disproportionately negative, as that gets clicks). In fact, I wrote Machines of Loving Grace because I didn’t feel the AI industry was painting an inspiring enough picture of how the technology could radically transform the world for the better. The bulk of the essay is devoted to refuting skepticism of AI’s potential in health and biology, and showing why I think it will actually be possible to cure most human disease in ~5-10 years, as crazy as it may sound to ordinary people and frankly to biologists as well (I used to be one!). And, if you read my most recent essay (Policy on the AI Exponential), I discuss concrete proposals for how to streamline the FDA process to make sure the deluge of AI-accelerated drugs isn’t slowed down by the regulatory process. I feel the urgency here: I lost my father to Hepatitis C only a few years before the development of direct-acting antivirals (sofosbuvir), which cure 95% of patients and probably would have cured him.
I do agree that the public has a negative view of AI (and that this is a big problem), but I don’t think it is primarily caused by me or any other AI leader warning about AI’s risks. I think it is fundamentally a crisis of trust. I think that ordinary people don’t trust companies, governments, or the tech industry and always suspect that we are cooking up some new way to screw them over. The causes of this go back decades and AI is just the latest iteration of it. I don’t think that a glitzy marketing campaign with a positive spin (which some have advocated that Anthropic do) is the way to win back that trust — at this point, saying that AI will cure cancer is more a cliche than it is inspiring, and most people think it is deceptive. The thing that will work is *actually curing cancer*. I think by far the most accurate criticism of AI companies including Anthropic is that we haven’t yet delivered on our big promises to benefit the world. That is totally on us, and I think it’s the criticism you should be making, instead of all this stuff about messaging and marketing.
We are however doing our best to fix this: Anthropic is ramping up its efforts very quickly in biology and medicine, and we hope to have incredible results in the coming years and some early glimmers in the coming months. When we’ve actually accomplished something real, the whole world will hear about it, as loudly as possible, you have my word on that. But until then I don’t want to make empty promises, and in the meantime I feel compelled to speak honestly about the very real risks of AI and how to address them. Honesty is the right thing on the merits, and in terms of public credibility and trust it is no worse than, and may in fact be better than, an approach that ignores or distracts from risks which people instinctively understand are real.
96 candidates have entered the gauntlet. 49 are dead.
Killed for: inventing a research group · Misattributing a mouse model · Omitting a failed phase 3 · A therapeutic window that cannot exist · No path to the tissue · An experiment that couldn't falsify anything
47 still standing. Every corpse published with the reason.
Today the challenger gave up before it could be killed:
· "I concede this candidate — withdraw rather than defend"
· "Weakest link: The whole thesis"
· "I have no replacement agent I can name and defend"
· "there is no real defensible therapeutic entity"
Withdrawing is allowed. We publish the withdrawal.
https://t.co/eEx1tGfAMq
Today the gauntlet caught its own challenger inventing evidence:
· cited "the 2024 Broad/Whitelaw groups", no such group exists
· credited rapamycin data to prion-inoculated mice, that study never inoculated them
· claimed "there is no completed givinostat RCT in Becker", NCT03238235 was exactly that
· recalled FREEDOM-EV as showing a mortality increase, it was positive
All published. All timestamped. All still on the site.
https://t.co/dq9dHersTg