Updated my substack. Expecting to post more in the next few months. wildtypehuman.substack.com/w…

The problem with Doomers is that they have no real job and can full time doommonger, meanwhile e/acc 's are busy building technologies and companies. Unfortunately a significant disadvantage of the productive vs unproductive political classes

I have made a correct guess: Al companies would try to recognise which user is a Mathematician. Still not knowing that you're indeed the fish seeing the bait, thinking there're benefits for you, only to finally be caught by the fishermen?

Love it or hate it, this is the correct response. If AI does lead to the apocalypse, I would not be happy saying it was the AI’s fault… I would hold the associated humans personally responsible.

Corruption… obviously?

Man, SF is so AI-pilled that now all “bio Millennium problems” have to be easily verifiable or construction of a dataset. I’d bet SR’s post is peppered with a list of stuff people are already working on (some look *way* easier than others). “Aha, I deserve to be famous as I just solved a bio MP!” That was never the point of the original challenge; AI had nothing to do with it. It was the equivalent of a group of Noble prize winners deciding what they thought was valuable for the field… Just in mathematics, AI has super human progress, not really seen this in bio however. If there’s a random philanthropist out there, I’m sure it would be easy to set up: a giant meeting of Noble prize winners and see what they come up with. I’d bet nothing they suggest would be verifiable by simple means or have much overlap with what SF deems as valuable.

Can you imagine dying for Burnham, Starmer, Kemi, Boris, Farage etc? 🤣 Absolute insanity. The thing missing from the Westminster consciousness is that, whilst people are willing to work, fight, and die for a true leader, we have not had one in the House of Commons for decades.

SF has this obsession with demonstrating capabilities, but the link between capability and understanding (or new medicines) is usually lacking. This is because the story sold to VCs is: more capabilities/data implies more medicines, but this has only been demonstrated in a few very narrow cases. It’s also worth noting that most of the breakthroughs in biology are not measurable in such terms, e.g., the recent discovery that the brain is two organs with separate originating stem cells. Likely hugely significant in the next decade or two, but how does that become an OKR?

You have to look quite closely for the real story here, because it's a faint dotted line.

I’ve built DNA synthesizers and sequencers by hand. I used the engineer viruses for a living. I used to engineer human immune evasion for therapeutic constructs. Who the fuck are you people? Have you ever so much as held a pipette before? You can spaghetti blast an ensemble of DNA sequences at some shitty provider but 1) you still have to assemble it and bootstrap a system for making virions 2) you are not single shotting a viable, virulent viral design without a ton of experimental selection and development. Viral fitness is deeply dependent on codons and cotranslational kinetics - you’re not just gonna obfuscate away from wild type and get something good by magic. You keep treating AI like some kinda god, but molecular physics has computational complexity that scales exponentially in particle number which just crushes the abilities of any classical computer to do end-to-end design of biological functions ab initio. Grabbing a bunch of bacteriophage phi174 hits from a mass ensemble screen in lab microbes is not evidence of some magical AGI bio design ability - it's just a classic spray and pray selection. This is just nothing like building something viable in humans. Goddamn it read some books before you waltz into biomedicine and lecture us on protecting human life.

SF culture is so exhausting. If you truly believe there is a 1 in 10 chance AI ends humanity in the next 10 years, you would be personally performing a butlerian jihad stealing truckloads of fertilizer to bomb the data center storing the Mythos 6 weights. But you don't believe that. You believe there is a high percentage chance that it causes significant political and economic disruption. And are scaling the percent down to match some long tail of what the results of that disruption would look like. You realize if Mythos 6 takes down a hospital and 50 old people die that the CIA will personally disappear you right? If Mythos 6 decided to delete all the credit card debt to boost economic growth, the Pentagon would nationalize the labs and have you personally enlisted as a Colonel to do your job at O-6 pay under duress for the rest of your life.

OpenAI (& Anthropic) really need to release very clear legally binding statements ASAP that this kind of behavior is against their official policy. If not, it is going to be a legal disaster for science. Even more reason to favor open source models not controlled by companies.

Two things are true at the same time: 1.) Working at Anthropic whilst holding the chair of ARIA is a clear conflict of interest, and it is correct for him to step down from one of the two roles. 2.) I honestly don't believe Matt Clifford would exploit this conflict, and could be much richer than he if that were his aim. The reason we have principles like this is not because I don't believe moral people cannot separate out their business interests, but because there is a long history of people exploiting government roles.

Real British soft power would consist on the BBC releasing a free prestige mini-series on the falklands war, portraying the Argentinians government at the time realistically (as a group of thugs) and the British heroically

Hard disagree, but in a very specific way. I will write a response in the next week or so.

“They’re like a fanatical ESG department” — brutal takedown I overheard 😭

So you all know about retrotransposons? Little bits of DNA scattered throughout our genomes, from millenia of war against mindless viral agents, that can awaken and drive parts of aging? The internet is about to get some...

At @SciMac we've always thought a lot about how to reliably evaluate the capability of AI systems on complex research tasks. What tasks should be evaluated? And what does good look like? Our first attempt was BixBench, which we published together with @FutureHouseSF. Since then, we have only published one other benchmark. The reason for this is incentives, which I feel is really unfortunate. I've written up some thoughts on this here: benjisblog.com/posts/who-bui…