Jake P. Taylor-King

My limited experience of blogging has revealed to me that I get more views when: (1) I have short titles; and (2) I'm being negative. I want to do more of (1) and less of (2), so been revamping a few of my posts at wildtypehuman.substack.com

I put a prompt injection into my LinkedIn bio and recruiters are messaging me in Old English and calling me Lord.

@wildtypehuman We'll shift to math conjuring rituals, like the miracleworkers of old. Glory days.

My biggest worry isn't that AI can do math*, but that no one will learn it anymore. In 100 years, will pure math become a dead language like Egyptian hieroglyphs or Latin? However, it will be the opposite to dead languages now: we will be able to say the sounds, but without the understanding. *Alas, I'm sure it can one-shot my PhD chapters.

She's old enough to be your wife's daughter you sick fuck

Starmer is the final boss of cancel culture, and demonstrates that if you just say "no" and lock yourself in your office, all the emails and petitions and dossiers will mean nothing. He's like Bartleby crossed with Stalin crossed with Savonarola. They'll have to dismantle the entire apparatus of "procedure" and bureaucracy itself to unseat him: perhaps he is a secret admirer of Ivan Illich!

@MattyKirsh @endpts The answer I want to know: what’s the money for? No one can run 2bn worth of clinical trials, so I assume data?

Re-surfacing this old @endpts article on Isomorphic with the Demis catch-phrase "You do have enough data - if you are innovative on the algorithm side." Thoughts on the raise aside, Isomorphic is (mostly) unique in the AI x Bio cohort as the group focused on improving the underlying model architecture as the most important lever we have rather than accumulating larger and larger datasets. Incidentally, this reminds me of a conversation with a colleague at one of the frontier labs who ranted to me that "Biotech is not nearly AGI-pilled enough yet". The consensus view in tech-circles seems to be that generalist models will get good enough to solve bio challenges without having to lean on niche bio datasets. . . and this seems to be Isomorphic's approach. It makes sense why they're the AI x Bio shop to place your bet on if you are AGI-pilled investor group like Thrive that has been in the front row for AI's rise outside bio since 2022.

This raise is literally just 1 VC (Thrive) + 3 different google funds, + 3 Sovereign wealth funds.

Here's an idea: if a governmental department is able to give to charity, then it should have its budget cut?

These stochastic parrots, eh: “the model proved a result that in my assessment would have made a perfectly reasonable chapter in a PhD thesis”

BREAKING NEWS “Furious” Starmer was not told local elections had taken place yesterday. The PM claims civil servants kept him in the dark until this morning. He has ordered an enquiry into why the elections had gone ahead without his knowledge. “I thought we’d cancelled them” he told Sky News.

I want to vote for a party that will: - scrap the triple lock (move to GDP/capita lock) - build significant infrastructure and homes (e.g. Forest City) - spend less on healthcare (e.g. move to insurance model) - cancel most student loans (for many, uni is a waste of time and money) - completely stop low-quality immigration - help businesses adjust to higher cost of labour (automation loans etc) - spend more on defence - means test state pensions - commit to experimenting with alternate education models (idk what will work in the future, but not what we have) There’s no option right now that even gets me half.

When I moved to the UK 16 years ago, my first job was at @Arm in Cambridge. This was at a time when the first iPad had just come out, and everyone was getting excited about British-designed chips ending up in smartphones, cars, TVs and... "IoT" devices. Westminster rushed to christen the area "Silicon Fen," and ministers loved the phrase. They visited often, posed for photos, and told us we were building Britain's answer to California. Inside Arm, nobody really thought of the company that way. We saw ourselves doing something recognizably Cambridge: patient, technical, and unflashy. Arm's entire model, licensing chip designs that the world's chip manufacturers and OEMs then built upon, only made sense because of where it sat and how that place worked. I think about that often these days, arriving from my morning commute into the Knowledge Quarter where @synthesiaIO is headquartered. The neighborhood has quietly become a magnet for AI companies. What they are responding to is a particular convergence of frontier research, applied AI and top universities within walking distance of one another. It cannot be conjured by speeches in Parliament or branding exercises. My ask of policymakers, as I mentioned to @TomSBristow in today's The Morning Intelligence newsletter, is small but specific. Resist the urge to call the Knowledge Quarter Silicon anything and stop talking about the next "Google DeepMind." The value of the cluster forming around King's Cross lies in its specificity, the same way Cambridge's value lay in being different. Your job is to understand the ingredients, protect the causes that produce them, and let the place keep developing on its own terms.

@homegymcoop What is this equipment called?

The world is still completely unaware that digital resistance exists and gyms all over the world will soon be full of it (like it or not.) I recently visited YouTuber Will Tennyson’s home gym, and showed him for the first time. The reaction is always the same, total surprise.

This is why British people have lost their faith in government. We are seeing the Civil Service rapidly turning into a 'state within a state' where radical left political agendas trump the democratic will of Parliament, ministers and British voters. The era of the ‘activist official’ must end. We must restore the Civil Service to its proper constitutional place.

@wildtypehuman This is the benchmark I’d love to see. Can the model not only predict a clean result, but identify why a real experiment failed — batch effect, protocol drift, sample quality, wrong assumption, or biology simply not cooperating?

30 years of workarounds. One foundation. Thank you @tom_nhs for finally saying it plainly.

We have a brand new railway that has been sat unused for two years because the unions want to double the number of staff required to run them.

@ArchieHall My god, this is so insulting to intelligent life

The maddest chart in England—the scoring schema the court used to determine that retail and warehouse jobs were comparable. Full context here: notes.archie-hall.com/p/against-equa…

As everyone loves perturb-seq, I’ve got/know at least 3 datasets where the experiment failed for completely different non-obvious reasons. I’d genuinely be impressed if an LLM can one-shot what went wrong.