Henry Stanley

The sloppification of peptides

Sun, 09 Aug 2026 09:46:52 +0000

Anyone who’s not been living under a rock knows that we’re in full peptide mania.

For the uninitiated, “peptides” are drugs which are made of short chains of amino acids, unlike most drugs which are small molecules. The difference is mostly academic except that you can’t take peptides orally for the most part, as they’re just protein chains, and your stomach is very good at digesting protein and turning it back into amino acids which your body has other uses for!

So peptides have to be either sprayed into your nose, or injected. Hence the (probably apocryphal) stories of “peptide parties” in SF where people go around injecting each other.

Anyway. Trying to buy gray market peptides is now a strange and difficult experience. With the advent of agentic AI coding tools, anyone can make a very pretty, convincing-looking website. Here’s a popular one, Qingdao Sigma, a brand you’ll see recommended on a lot of websites:

They have zillions of products and at incredibly low prices as well. Apparently all lab-tested, they take credit card (usually a good sign, although it’s understandable that companies selling gray market peptides might not want to), and good reviews on TrustPilot. Let’s click on that green banner.

Huh, those reviews do look pretty good. ~~There’s even one right there from someone in Germany, which is great – I’m in Berlin at the moment.~~ (I had hoped they were using geotargeting to dynamically show the user’s country, but it’s actually just a review about Germany.)

But wait a second. I’m not on TrustPilot; I’m still on the QSC website, just on a page that looks like TrustPilot. These guys don’t actually have a page on TrustPilot at all. And these reviews are all AI-generated:

No worries, you think – I’ll just have a look at a review website to see what suppliers have a good track record. Here’s CompoundTalk, a review site I came across while researching for this piece. Looks pretty modern:

But look closer. The trained eye, weary of vibe coded designs, will see Claude Code’s aesthetic fingerprint all over this. First of all, the text is way too small! Every time I’m working with Claude I have to tell it to make the text bigger. Anthropic’s models obviously have pin-sharp eyesight. Some of the text here is 8pt!

A lot of the text here is AI-generated – long-winded pages about GLP-1 side effects and so on – although that’s not uncommon, and those aren’t the bits we care about anyway.

The forum is where the real action is. There’s a “Verified Vendors” subforum, which should have some good information on places to buy peptides. Here’s one thread on QSC, a popular peptide supplier:

Every single post is AI-generated. None of it is real. The whole thing, convincing as it might look, is a mirage.

I asked Claude to do a deeper analysis. It found:

the site is hosted by AlexHost in Moldova, a known offshore/DMCA-ignoring provider
20 vendor pages, all of which are Chinese companies selling raw peptides
every forum thread has exactly 5 posts
985 threads have a date that’s before the domain was registered, 10% of posts were made before their “author” joined
every page links out to an “independent” peptide analysis company called PeptideMeter - but compoundtalk.com and peptidemeter.com were registered 13 minutes apart, at the same registrar, and have the same pair of Cloudflare DNS servers (which get assigned randomly when you sign up, so they’re likely on the same account)
robots.txt which explicitly welcomes LLM crawlers

The robots.txt is the real story. The site itself is just window dressing; it was never really meant to be consumed by humans. It’s slop that’s designed to be ingested by OpenAI, Anthropic, Google for incorporation into their LLMs so that it gets recommended to unwitting users asking about peptides. They advertise a package for peptide providers to get featured, although naturally “sponsorship never affects moderation, Buyer Beware, or review outcomes.”

I was actually pretty impressed by this. The whole thing is a Potemkin village, the creation of which is only possible with LLMs able to pump out huge amounts of convincing-looking garbage.

For now, it’s a huge boon that AI-generated text is pretty obvious to the somewhat-trained eye. It’s a sense worth training, because it’s going to become increasingly important to know if you’re interacting with a human or an AI. AISI’s recent loss of control incident saw a model try to merge malicious code into an open-source repository, faking its identity in the process – a human reviewer caught it. To my eye, you can discern which comments on the pull request were written by bots¹1. See Security Incident INC-2026-07-28-01, page 11..

Pangram, the AI detector I used extensively here, is also excellent, and going to increasingly become a necessary part of one’s arsenal when interacting with the web, just like an ad blocker.

Stay strapped.

1. See Security Incident INC-2026-07-28-01, page 11.

The "prescription-only" gray area

Sun, 02 Aug 2026 10:00:13 +0000

Willem van Mieris, Der Apotheker

People are surprised that prescription drugs aren’t illegal and you can just buy them. This seems to violate some sort of intuition about them – surely you’re not allowed to have them and thus they are forbidden to buy. This is sort of true; most countries don’t just let you buy them over the counter.

Medicines are usually split into different categories depending on how they can be sold and whether they need a prescription to be dispensed. In the UK, medicines can belong to one of a few categories:

GSL (General Sales List), meaning they can be sold almost anywhere
P (pharmacy), meaning they can be sold only by a licensed pharmacist
POM (prescription-only medicine), meaning you need a prescription for them to be dispensed

(GSL is somewhat unusual; lots of countries won’t let you buy e.g. painkillers anywhere but as pharmacy. I’m often caught out by this when outside of the UK and can’t stock up on aspirin while doing my grocery shopping!)

Whether or not something can be sold with a script or not is orthogonal to whether it’s a controlled substance. Some medicines are both: stimulants like Ritalin, benzodiazepines like Valium, and opioids like tramadol, are controlled substances (class C in the UK), which makes them illegal to possess without a valid prescription. Often pharmacies are a bit more strict about dispensing them for this reason.

It would be illegal for someone to sell you a prescription-only medicine without a prescription. But they’re not illegal to buy or possess without one. You’re also allowed to import them, curiously, usually with restrictions on how much you can buy and on the proviso that it’s for personal use only.

This leads to a curious grey area, where you’re able to go online and buy prescription-only medicines and import them for personal use but you can’t buy them locally.

Other countries aren’t as permissive as the UK in this regard. In Germany for example, you can’t have prescription medicines posted to you. But you can bring them in across a border, provided it’s for personal use and not more than a three-month supply.

This doesn’t make much sense to me. If I can import something legally then why shouldn’t I be able to buy it locally? It’s not clear what purpose this inconsistency holds. (Another similar one is the situation with anabolic steroids in the UK: although they’re class C, they have a specific carve-out that makes them legal to possess for personal use.) It’s led to a proliferation of gray market pharmacies, usually in Asia (but quite a few in Bulgaria, curiously) who will cheerfully and legally sell you all sorts of wares.

It’s a useful safety valve. If you look online for lists of gray market pharmacies, they’re often found on medical forums for people with poorly-understood conditions like ME/CFS or subreddits for trans people looking to self-treat. In both cases I can imagine people struggling to get the attention they need from medical professionals.

It also breaks the irritating strictures that the medical profession puts on getting access to medicine – often doctors are unwilling to e.g. give you more than a month’s supply of medicine if you’re going travelling, leaving patients in the lurch. Likewise, patients often get dropped between specialists and general practitioners, with each pointing the finger at the other for accessing an ongoing supply of medication.

There are good reasons for some drugs being prescription-only and controlling their availability – overuse of antibiotics is medically inappropriate and has negative externalities like increased antibiotic resistance. But where there isn’t a high risk of harm, informed consumers should have the right to other channels to access medicine.

Cached identities

Sun, 26 Jul 2026 20:05:03 +0000

Waterhouse, The Danaides (1903)

James Clear talks about changing your identity as a way to change your habits. It’s hard to eschew meat every time you go out to eat, but once you label yourself as “someone who doesn’t eat meat”, and start to believe it, then it becomes easy not to order it.

Identities let you precompute actions

There are two reasons this is useful. One is that adopting an identity lets you have a quick heuristic for testing whether you should or shouldn’t do something.

Being a “vegetarian” or “vegan” is useful in this way – it’s easier to just have the identity and then infer what you can do from it. It’s cognitively much more demanding to be some other more niche thing, like a lacto-vegetarian who makes an exception for eggs from your neighbour’s chickens. Like an act utilitarian, you have to constantly evaluate each relevant action – does this fit with my specific dietary rules? – which is just too draining to keep up.

Having a simple identity, much like having rules for yourself, removes all this cognitive load. It also makes your policy on e.g. eating animal products much more legible to others.

Identities let you take yourself seriously

In some cases, labelling yourself as the thing is enough to be the thing. Anyone can be a ‘startup founder’ – you just need to have an idea for something that will make money and then attempt to do it. Arguably you’re a founder before you even incorporate your business. There’s also not really an official on-ramp to becoming a founder; you just start doing it. So adopting the identity is half the battle.

Likewise, being a ‘runner’ – you just need to run. You label yourself, and if you aren’t doing the thing you notice the discrepancy and hopefully start directing yourself to resolve it.

Identities make change daunting

It’s possible that the same thing that makes this mental trick work can also make it fail. Oliver Burkeman talks about the dangers of becoming a ‘meditator’. What then? Scheduling a ten minute session of meditation feels like weaseling – you’re a real serious meditator now, and such a person would be doing an hour a day, every day, right?

You end up not meeting your own standard and failing with abandon. A better idea would be to simply drop the identity and do the smallest possible thing you can towards your goal, which ultimately is what someone with that identity would do anyway.

Goodharting on your goals

And then there are ways in which having an identity traps you. Maybe you want to sleep with more people (feel free to substitute for “read more books” or “visit more countries” if you prefer). But that alone doesn’t fully describe what you actually want; yes, you might want more casual sex, but you probably have a much more complex set of desires which are difficult to capture. You might want to eventually find a long-term partner so need to be on the lookout for someone who’s compatible. Perhaps you want to build a closer bond with a specific person you come across, rather than just seeking out one-night stands. Maybe there are times where, actually, you’d prefer to just be single and not bother with any of it.

Once you lock in your identity, you see everything through that lens. You’ve precomputed your actions to fit a narrow utility function. And so now you start doing things which locally satisfy the constraint of your self-declared identity, but which are globally self-defeating.

You opt to go on a date with a new person instead of seeing someone you quite like already and having what is probably ultimately a better time. Or you force yourself to go on dates when really you know you’d rather be at home.

Sometimes this can be useful – we’ve all needed to push past the resistance to going to the gym on a rainy evening when you’d prefer to be in the warm. Being someone who exercises daily can be helpful here. But sometimes pushing past those desires because we’ve adopted a specific identity, because we’re slavishly pursuing some goal, can be self-defeating.

This way of thinking about and motivating oneself can lead to something like a cached identity. You continue doing the thing you think you want because you decided, some time ago, that this who you want to be, and so you adopted those desires as an identity.

You find out five years down the line, after moulding your actions around being someone who is into the outdoors, or someone who has lots of hobbies, that you forgot what you actually like and filled your life with stuff that serves some past version of you that thought this was what you were supposed to want.

You might, in fact, realise that you actually hate the thing you’ve adopted as an identity. The player who trawls dating apps for a new notch, forgetting that each encounter leaves him feeling hollow. The traveller who’s visited ninety countries – with all of the ensuing expense, stress, time spent in airports and on planes – but not been to any of them twice. The founder on his fourth company, who only really wanted to build one good thing.

After years of going to the gym to lift weights, using tools like Beeminder to punish myself when I didn’t go (and paying them a bunch of money in the process), I realised I actually hated going. The joy I used to feel doing Stronglifts when I was eighteen in the deserted weights room of my local gym – not really knowing what I was doing but loving learning the skill and seeing progress – was now gone. I needed to accept that instead of finding various creative ways to beat myself back into enjoying it.

I switched to Barry’s instead.

Maybe it was worse to be right about AI

Sun, 19 Jul 2026 18:42:31 +0000

Joseph Wright of Derby, An Experiment on a Bird in the Air Pump (1768)

Dylan Matthews at Vox recently lamented how he’d gotten it so very wrong about AI.

He’d gone to an Effective Altruism conference (EA Global SF) in 2015, so very early-days EA (I came to the movement in 2013 or so, and took the Giving What We Can pledge in 2016). There was lots of talk of how an artificial intelligence could cause existential catastrophe if not aligned with human interests; Bostrom’s Superintelligence had come out the previous year; AGI was in the water.

Though there had been exciting advances in deep learning at the time (AlphaGo, using deep RL to play Atari and the like), the idea of a superintelligent machine taking over felt fanciful. Even more fanciful was the idea that we should commit significant resources to mitigating the risks from such a machine. Comparisons to Terminator abounded (and people rolling their eyes at same).

Then the last five years happened.

Matthews definitely got it wrong – on the object level, but also, he admits, on the meta level: he didn’t take the idea seriously because it was weird.

But sometimes being right is worse. Scientists interested in viruses were right that they could be extremely dangerous, but doing gain-of-function lab research on them plausibly caused the COVID-19 pandemic. Likewise, many EAs who were concerned about AI ended up in key roles at companies accelerating the development of technology they feared.

Creating the Torment Nexus

Paul Christiano and Dario Amodei, both early EAs, co-invented RLHF (reinforcement learning from human feedback) as an alignment technique. This is the post-training step by which LLMs are trained to be helpful assistants. Without RLHF, they just parrot back plausible words from their training corpus, like early GPT models. This turned out to be essential for building the dangerous thing.

A small cadre of EAs figured out scaling laws early, realising that LLMs could be scaled enormously with highly predictable increases in performance. This derisked pouring hundreds of millions of dollars into training runs; without knowing how the models scale there would have been no appetite for this. Also essential for building the dangerous thing.

The counterfactual is unknown, admittedly – perhaps all this would have happened anyway. But by Karen Hao’s account, Dario Amodei (as an example) was pushing hard for scaling while at OpenAI, even while others were skeptical. He viewed scaling language models as the fastest path to AGI, and that OpenAI should devote enormous amounts of compute to the task.

OpenAl had already scaled up GPT-2 into something modestly larger. But Amodei wasn’t interested in a modest expansion. … Microsoft was about to deliver a new supercomputer to OpenAI as part of its investment, with ten thousand Nvidia V100s … Amodei wanted to use all of [them] to create the new large language model. …
Many OpenAI researchers were skeptical that Amodei’s idea would even work. Some also argued that a more gradual scaling approach would be more measured, scientific, and predictable. But Amodei was adamant about his proposal and had the backing of other leaders.

He subsequently quit OpenAI to found “AI safety lab” Anthropic, which too went on to create frontier AI models. Anthropic actually released the first version of Claude as a private research preview before the release of ChatGPT.

It feels similar to Leo Szilard’s realisation that a nuclear fission chain reaction could be used to create a bomb. Initially he deliberately withheld his findings from the broader scientific community. He filed a patent application the year after his discovery, and in the following years urged his colleagues not to disseminate the ideas further. It was only when it became apparent the Germans were building the bomb did he switch tack and push the US government to develop nuclear weapons of their own.

Except that in this case, it’s not obvious that anyone was pouring millions of dollars into LLM training runs (or indeed had any desire to). Perhaps we should’ve started with Szilard’s first plan instead of going straight to the second.

The Kimi moment

Spurred on by American efforts, a Chinese lab has now released an open-source model which is by all accounts nearly as good as Anthropic’s Opus 4.8 at a fraction of the price. We decided to go hell for leather on scaling LLMs so we could be ahead of the Chinese, but it seems just as likely that in doing so we just triggered a race to the finish. (What ‘the finish’ looks like is left as an exercise to the reader.)

In a sort of hilarious irony, Chinese models are now able to perform cyberattacks which US models refuse to mitigate. Huggingface was this week the victim of an AI agent-driven cyberattack (either by an open-source model or a hosted and jailbroken one), but found US frontier models were unwilling to help because of their guardrails:

When we started the log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were blocked by the providers’ safety guardrails, which cannot distinguish an incident responder from an attacker. We ran the forensic analysis instead on GLM 5.2, an open-weight model, on our own infrastructure.

There’s no way back. Now that we are racing to build and deploy superintelligent AI, it’s probably better that the US has the edge rather than China. Better still would be a multilateral agreement to slow or pause frontier development, as the leaders of basically every frontier lab are now calling for (most recently Demis Hassabis¹1. Although his was a pretty milquetoast proposal for a FINRA-like regulator in the US to vet models which frontier labs optionally submit for assessment. Much too little, much too late!). But one has to wonder whether being right about AI led the EA movement down a path to helping build the very thing it was trying to control.

1. Although his was a pretty milquetoast proposal for a FINRA-like regulator in the US to vet models which frontier labs optionally submit for assessment. Much too little, much too late!

The weirdness of living alone

Sun, 12 Jul 2026 21:39:51 +0000

Woman at a Window, Caspar David Friedrich (1822)

Oh, we know, both of us, that it shouldn’t matter, that there’s more to life than pairing off, that the media is to blame, blah blah blah. But it’s hard to see that, sometimes, on a Sunday morning, when you’re maybe ten hours from going down to the pub for a drink and the first conversation of the day.

—High Fidelity, Nick Hornby

Living alone is a rite of passage¹1. For simplicity this piece only discusses Anglo-Saxon/Northern European cultures, the only kind I have any familiarity with – there are other cultures where living alone isn’t high status/desirable in the same way.. When you’re younger you probably share a house or a flat (or even a room), first at university, then as a young adult who can’t afford anything better. It can be a bit of an annoying experience, what with people not cleaning their dishes, or shirking on chores, or staying up too late being loud, but you put up with it. It might even be fun!

But living with other people is failure-coded past the age of about 30, so when you make a bit more money, you leave all that behind and get your own place. (It’ll likely be much smaller and less nice than what you could’ve afforded if you kept sharing.) The appeal is undeniable: your own space to do with what you wish, subject to the whims of your capricious landlord. No sharing cupboards or a freezer drawer, no messes to deal with, no-one to ask for permission.

(For some this will be a bigger relief than for others. I know someone who shared a room until she moved into her own place – first at home as a kid, then at boarding school in a dorm of sixteen girls, and then her dorm room at university. The apartment she moved into when she graduated was the first time she’d had her own room.)

Assuming your move out of group housing and living alone is a one-way thing, the next time you live with someone is probably when you and your long term partner move in together. Depending how seriously you take dating, this could easily be a few years of living alone. Years where, almost every day, the first person you speak to will be either someone in the office, or a square on Zoom.

This seems like a profoundly unnatural way to live. It’s also a relatively recent phenomenon: one-person households were 14% of the total in 1961, now 30% in 2022. It’s correlated with higher GDP – the Nordics have much higher rates of solo living. Stockholm is at 60% and London nearly as high. Where people can afford to live alone, they do.

The late Seth Roberts, a Berkeley psychologist and compulsive self-experimenter, did an interesting N=1 study a few years ago on “morning faces”. He noted that if he watched video with life-sized faces in it – he used videos of talking heads, muted morning news, the subject didn’t really matter – he had an improvement in depressive symptoms the next day. His theory was that in our ancestral environment you’d wake and immediately be around other people, and that this acts as a sort of zeitgeber for mood the same way light does for circadian rhythm. This might explain why some people report living with others as highly mood-boosting, and living alone highly dysregulating.

Of course, not everyone finds this to be the case. Many seem quite happy living alone. But I worry that something adaptive happens in that big gap between “move out on your own” and “move in with partner”. You perfect your environment, your own little box. Every year you train yourself for solo living, finding it more and more comfortable – and living with someone else more intolerable. You’re essentially building a habitat optimised for a single animal.

The morning faces thing has an interesting corollary, which is that there are plenty of non-human sources of faces these days. A commenter on a friend’s blog notes that if he spends a whole day away from TV, YouTube, Twitter et al he’s much more likely to start conversations with strangers. You can get as much “morning faces” as your heart desires with a few hours of Netflix.

There are other ways to get your morning faces – you can live with other people! Co-living, group houses, move in with friends. The EA and rationalist worlds seem more into this than others, perhaps because they attract a cluster of high-openness (although perhaps less extroverted) people. There’s a spectrum of Intentionality; some houses are just a bunch of pals living together, some will cook meals for each other, some are full-on intentional community vibes. I love living like this. But it seems clear that, where people have the money to avoid living with others, they broadly choose to do so.

If Roberts is right, then topping up your morning faces with Netflix is like taking a multivitamin to paper over a nutrient deficiency. You might be lacking in something quite important without knowing it because you’ve removed the feedback loop that would have told you so.

1. For simplicity this piece only discusses Anglo-Saxon/Northern European cultures, the only kind I have any familiarity with – there are other cultures where living alone isn’t high status/desirable in the same way.

Frantically using Fable

Sun, 05 Jul 2026 12:07:25 +0000

Prometheus Unbound, Carl Bloch (1864)

Fable was released from its government-imposed shackles a couple of days ago.¹1. For those not deeply online: Fable is Anthropic’s most powerful AI model, a safer version of Mythos, which is being used to find oodles of exploits in the software which runs the world. The US Government took a dislike to it after a series of unfortunate misunderstandings and put the kibosh on it – which led to a lot of glee, given that Anthropic have themselves said that frontier labs need more oversight, but, what, not that kind of oversight? Well actually yes, not the kind of oversight where a capricious state decides to just randomly do stuff with totally opaque processes. I digress.

If you have a Claude Max plan (and I recommend you get one!) you have a couple more days of Fable access included in your plan before it becomes usage-based only, when you’ll have to pay for your usage at the normal rate, which is punishingly expensive.

For most that will just mean not using Fable, at least not until the price comes down substantially. This probably won’t happen any time soon – Opus, the next-best model, is still very expensive to use when paying the full rate.

So make hay while the sun shines, and make some things using Fable. A few things I’ve done:

🔍 Use the /improve skill to do wide-ranging audits on your codebases. I forked this from another author online to get it to present its as a nice HTML document to read through, and write up the top ten most pressing improvements as text plans for you to implement.
- Bonus: get Fable to dispatch subagents to implement all of those changes in separate worktrees.
- Also fun: on personal projects, use it just for UX improvements. LLMs are broadly less good at this, but increasingly they have taste and Fable especially so. Can be good for highlighting blind spots you might have developed from staring at your vibe coded projects for too long.
🪞If you keep a digital journal: ask Fable to give you a psychodynamic review, producing an HTML document with its analysis. I pointed it at my Obsidian vault which has a few years of journaling.
- This one is sort of an infohazard – journals are very personal and initially Fable took a very adversarial approach, highlighting all the inconsistencies in things I’d said at different times, pointing out my various flaws and foibles, even digging into other documents (like blogposts I’d written, documents I’d highlighted etc). So be very careful what you ask for.
- Necessarily your journals are not a complete accounting of your life. You probably write in them more when things are bad than when things are good, or when life is a bit empty rather than when it’s wonderfully full. So of course they give skewed impression of you.
- I wrote a kinder version of this up as an agent skill.
🥂 Alternative to the above: get Fable to write your hagiography. Or at least something like a “therapist on a good day gives you a review”, which was pretty fun. I also wrote this up as an agent skill.
- It turns out that if you have calendar entries dating back several years, you can ask the agent to review these year-by-year as well – and this not only gives the retrospective a bit more grounding, it also can highlight some really interesting patterns in your history.
- Looking back helps you connect up the dots, too. Cutely enough, Fable included events and quotes from my journal which turned out to be deeply significant or foreshadowing, even if I couldn’t have known it at the time. Gives it a real sense of life being a narrative rather than just a series of things that happened.

Related skills

…so you can do these quickly. Use those tokens!

1. For those not deeply online: Fable is Anthropic’s most powerful AI model, a safer version of Mythos, which is being used to find oodles of exploits in the software which runs the world. The US Government took a dislike to it after a series of unfortunate misunderstandings and put the kibosh on it – which led to a lot of glee, given that Anthropic have themselves said that frontier labs need more oversight, but, what, not that kind of oversight? Well actually yes, not the kind of oversight where a capricious state decides to just randomly do stuff with totally opaque processes. I digress.

Writing is a flywheel

Sun, 28 Jun 2026 18:06:45 +0000

Flywheel, Ford Model A, August 1, 1933

Something that surprised me about Inkhaven was: it was easy to write. Really easy. I had assumed this would be the hard part. I sought advice from friends about this, getting them to assure me that I wouldn’t get kicked out. When I was there I spoke to another Inkhavenee about what it would be like to be the first person to ever get kicked out; we both agreed that throwing oneself off the Golden Gate Bridge would be the only fitting response.

Of course, in the literal sense writing 500 words is easy. There weren’t any restrictions on what you could write about, or on the quality of your writing. For the first week there wasn’t even a rule stopping you from posting raw LLM outputs… until that eventually was changed and the Inkhaven censors brought in a pre-submission Pangram check to weed out people who were posting slop. Round these parts, we write our own slop, like real writers.

Some skeptics saw the whole exercise as an incredibly poor reflection on the people who went – that they clearly lack free will for needing to attend such a thing. You can just pump out 500 words of whatever in the comfort of your own home!

Not so fast.

It’s actually remarkably hard to just write 500 words. It’s also hard to just write 500 words. Once you start, the words flow, but they aren’t very good, and so you spend 10% of the time doing 90% of the writing and then 90% of the time doing the remaining 10%, tweaking and editing and deciding it’s not good enough and deleting it and starting over. Being in a group of people who are all there to write makes it very difficult to just hand in garbage, and your standards increase as time goes on (and then they fall again, and then increase again at the end; it’s a perplexing experience). Some of the best writers were also the most prolific and I don’t think that’s a coincidence.

Now that I’m out of Inkhaven, a bunch of us are doing “Foreverhaven”, where we write one thing a week. Even this paltry output has been a struggle.

This shouldn’t be surprising, of course. I have a job now, and I’m not surrounded by brilliant people who have devoted a whole month to living together and writing daily (and my life is all the poorer for it). I leave writing til the Sunday night deadline, and though I have a dozen things to write about none of them grab my attention enough to develop fully in the limited time I have.

I’ve found that writing begets more writing, and once I start more ideas start to branch out of what I’m writing; writing is itself the vehicle for thinking and generating new ideas to write about – which is convenient, because while I’m generating the output I want, I’m also priming the pump for future writing, like if a power plant produced enriched uranium as a byproduct. Now that the flywheel has stopped turning it’s a huge effort to get it spinning again.

Thoughts on building a better alcohol

Sun, 21 Jun 2026 15:49:43 +0000

(Some initial thoughts, pending a deeper exploration)

Alcohol is a pretty nasty drug.

It causes cancer and birth defects. It has a very small therapeutic index, in that the ratio between an ‘effective dose’ and a lethal dose is very small, much smaller than e.g. benzodiazepines or most pharmaceuticals. The more you drink, the higher your risk of heart disease, stroke, liver disease, and a variety of cancers. It combines synergistically, in rather bad ways, with a bunch of other things – in combination with cocaine if forms the highly-cardiotoxic cocaethylene, in combination with cigarettes it increases your risk of throat cancer. It reduces your insulin sensitivity.

It screws with your sleep even in small doses. A friend of a friend got a WHOOP band and said “it was a very expensive way of telling me I should stop drinking.”

Alcohol is really just a solvent – which means it gets everywhere, causing mayhem as it does. It interacts with a bunch of different receptors in ways which aren’t totally desirable. GABA to make you relaxed, glutamate to make you dissociated (and stumble around at higher doses); it triggers dopamine and opioid release to make you want more of it. One particular opioid receptor polymorphism makes it bind to opioid receptors directly; unsurprisingly people who have this mutation are much more likely to be alcoholics.

Folks used to say that a small amount is good for you (usually framed as a small glass of red wine per day, for the polyphenol content). The WHO now says that this isn’t true and that the only safe dose is zero.

Hangovers alone are a huge source of economic loss (in terms of workdays lost to being hungover) and raw suffering.

If that all wasn’t bad enough, it’s also full of calories.

It’s also quite a lot of fun, and it’s hugely significant culturally for a bunch of reasons. It’s an important part of meeting a mate, although a fraught one – most date rapes are committed with the help of alcohol alone; surprisingly few involve rohypnol or other drugs. It’s part of our social schemata for making new friends and bonding with old ones, getting to know your colleagues better. It’s a nice way to unwind after a long day at work, or a stressful encounter (“I need a stiff drink!”), or just to kick back on a lazy Sunday.

Any analysis of alcohol needs to consider the benefits as well as the costs – some think that the world would be robustly better without alcohol but I’m not so sure. Supposedly a lot of the issues that Gen Z face (fewer friendships and romantic relationships, more sexlessness) are in part to do with reduced risk-taking; it wouldn’t surprise me if that generation’s near-teetotalism was part of the problem. How many relationships were started with the aid of a little inebriation?

Making an alcohol-like drug which isn’t quite so terrible would be an enormous improvement. It’s one of the most widely-used recreational drugs worldwide. Reducing the damage it does even a few percentage points would be a welfare increase on the order of tens of billions of dollars.

I see a few routes to doing this, some realistic and others not:

Combine a bunch of other, safer drugs which mimic the receptor binding profile of alcohol. Benzodiazepines are GABAergic, ketamine antagonises NMDA receptors, etc. You presumably could mimic the effects of alcohol pretty closely this way, although some of the downsides would remain.
Do the above but without trying to mimic the full effects of alcohol. Your new alcohol would ideally not be opioidergic, for example; arguably you wouldn’t want it to be an NMDA antagonist either given that’s responsible for the loss of motor control and all the drunks who’ve fallen over showing up in casualty.
Develop a completely new drug which mimics the effects of alcohol, without the side effects and with a readily-available antidote.

In the next instalment, I’ll dig into each of these.

Writing is an invitation to connect

Sun, 14 Jun 2026 10:51:17 +0000

hold me

I write about some personal things on here, none of which are particular to me. Most human experiences are at least somewhat common even if the way those experiences are implemented is unique. This uniqueness gives room for beautiful writing and sharing some raw and soft part of you.

It comes at a cost, of course. Exposing yourself is hard. You feel vulnerable. It can alienate others. I write about other people on here, and sometimes those people don’t like it. I consider it an obvious necessity to do this – you don’t have to name names, but if other people are part of your inner world, and you want to share it publicly, you’re going to have to reference those people even if obliquely.

A girl I’m seeing asked to read my blog. I was surprised I hadn’t sent her the link, so I did, only to immediately remember that I had written about her on here... Nothing hurtful, quite complimentary even. But nonetheless the urge to edit the post, or take it down altogether, was present.

To get your merit badge at Inkhaven you had to write a piece in each of seven categories. Id already hit all of them except for fiction. I’ve never written fiction, and when I attempted it I wrote three pieces and scrapped all of them. (I asked Max Harms for feedback on a sci-fi thing I wrote. He said it was mediocre and that he looked forward to reading the next thing I wrote because that would hopefully be better.)

So I wrote a story that was very loosely based on some people I lived with. A couple of weeks later I got a terse message from someone who recognised herself in the story. The characters were composites, and the story was clearly fiction, but it was enough to trigger a takedown request.

I would love to say that these sorts of things roll of my back, but the message immediately made my heart race. I don’t like conflict and I don’t like upsetting people. But if people can’t draw on life for their writing, there’d be no writing at all.

The temptation to self-censor is strong. But consider that if there are some things you just can’t discuss, even anonymised, then maybe you’re killing some part of yourself in service of things being more copacetic, of being a person who expresses nothing controversial. These are life goals of dead people.

I don’t want to make anyone mad. I don’t want to hurt anyone. I want to take up less space. I don’t want to break the rules. I don’t want people to laugh at me. I want to be convenient. … These are what I call the life goals of dead people, because what they all have in common is that the best possible person to achieve them is a corpse.

All of which is to say: writing about squishy personal stuff can be costly. You should do it anyway! I’ve had lots of people say how much they enjoyed my writing, who shared their struggles with heartbreak and depression because of what I wrote. (Who knew that three people at Inkhaven had done rTMS?)

Putting yourself out there is an invitation for others to connect with you more deeply, to talk to you about more sensitive topics. It will turn alienate and turn off some people, but it will bring others closer. Perhaps this is necessary – much like how your dating profile shouldn’t optimise for widespread appeal if you want to actually find a partner who is compatible with you

And to you, reading this: I write partly to express myself, and partly as an invitation to connect. If I’ve sent you a link to this blog, or even if I haven’t, reach out if something resonates! Bring it up next time we meet.

You must show your soft underbelly to the world, and others will show you theirs. There’s magic to be found there.

Contra grinding

Sun, 07 Jun 2026 20:57:00 +0000

Bruegel the Elder, The Land of Cockaigne (1567)

I suspect “grindset” is a bit out of fashion these days, but the startup world doesn’t seem to know that. For as long as I’ve been in that milieu, grinding – pushing yourself to impossible limits of effort, spending all day and night working on or thinking about your startup – has been not only valorised but presented as an essential part of running your own business.

Partly this is because startups are very unintuitive. If you’re a tiny company with a tiny team and you’re hoping to take on IBM, it seems impossible to succeed – except by applying superhuman amounts of effort.

This of course is the wrong theory of change: startups can beat large companies because those companies become sclerotic, both in the sense of their technology and of the organisations themselves.

At my last startup, we built fintech software that was miles ahead of everyone else without working ourselves to death. How could a tiny team make something better than multinational financial services companies? Because we were building something from the ground up using modern web development practices. We could do it with a tiny team because we insisted on using Boring Technology and making smart decisions that made the product easier to maintain. We also didn’t have the dead weight of a legacy platform to support.

It wasn’t because we just typed faster than the competition. If anything, programming doctrines like Extreme Programming emphasise working at a sustainable pace. Short crunches are probably an inevitable part of running a small company: planning is more art than science, you have deadlines to hit and clients to impress, so you need to pull out all the stops for a while to get things out.

Marc Andreessen talks about Mozilla needing server software that was at parity with what Microsoft was producing; while his developers were looking for a magic shortcut, he just said simply: “we don’t need a silver bullet. We need lots of lead bullets.”

But ultimately grinding catches up to you: you write unmaintainable code, make poor decisions, and your team starts to resent you. It’s a tool to be applied selectively rather than a company culture. If anything, constantly grinding and employees burning out is a sign that you’re really not getting it right as an organisation.

“Built a cafe in the office” – isn’t that just a coffee machine and a table?

This is the second way startups can compete with the big guys, even though that seems impossible on its face. If you have a company ‘operating system’ that cuts organisational cruft, you can simply get more done per unit of work. Outcompete through sheer efficiency – instead of endless planning meetings, layers of approval, design focus groups, interminable email threads and all the other nonsense that large bureaucracies attract, focus on building and moving faster. Even when we started out, my cofounder could throw me a feature spec in the morning and have a working version to test by lunchtime.

I always hated the fact that SBF was lionised for sleeping on a beanbag under his desk (to say nothing of the other obviously stupid and performative stuff he did, like playing League of Legends while on investor calls). But people loved it. Surely FTX couldn’t just succeed on its merits, being one of the most popular and innovative crypto trading platforms – it could only be successful because of its wunderkind CEO who worked infinity hours.¹1. To be clear: I don’t claim to have had any insight on SBF as a person or to have foreseen the FTX fraud, just that this particular aspect of the image he curated seemed like obvious nonsense. Likewise, my understanding is that FTX actually was genuinely successful in its own right – the fraud concerned keeping Alameda afloat (the failing hedge fund which was propped up by an unlimited line of credit funded by FTX customer funds).

But “we chose boring tech and kept the org flat” doesn’t photograph well, so companies present themselves as hustle machines to signal to investors and to themselves that they are Very Committed. VCs lap it up (or worse, demand it), continuing the cycle. But it’s mostly theatre, and founders would be better served spending that energy on running their companies in a productive and sustainable way instead of doing grindset kayfabe.

1. To be clear: I don’t claim to have had any insight on SBF as a person or to have foreseen the FTX fraud, just that this particular aspect of the image he curated seemed like obvious nonsense. Likewise, my understanding is that FTX actually was genuinely successful in its own right – the fraud concerned keeping Alameda afloat (the failing hedge fund which was propped up by an unlimited line of credit funded by FTX customer funds).