Hacker News Reader

Comment by littlestymaar | Hacker News Reader

I too, like everyone one else, often hate the tedious and often absurd, effort that bureaucratic procedures sometimes require, but this story doesn't make any sense to me.

We're talking about a guy who's used AI to make personalized ground-breaking medicine for his dog but says he spent three months typing a 100-pages long document for the red tape. In reality, current AI technology isn't particularly designed to help you making radical medical breakthrough on its own (at least yet), but is extremely proficient when comes to writing text that must just check boxes.

I'm sorry but how does that story not smells like complete bullshit to anyone reading this? Given that the guy telling his story is “an AI entrepreneur”, I'm almost 100% sure that the story is almost entirely made up for self-promotion.

View on HN

It's not a radical medical breakthrough, it's applying a technique already documented in the literature and years into human clinical trials. The LLM is just doing literature summary and planning. The most notable AI innovations here are in protein folding and binder prediction.

There's pictures of the dogs tumor progression and scientists from government labs validating his story.

Yeah. It's possible to think that there are cases where lowering the cost of first phase trials or making it relatively simple for people capable of offering informed consent to obtain access to existing safe medications currently undergoing trials for their specific ailments and thinking that a person who claims to have "designed an RNA vaccine" by asking an LLM to teach him about RNA and now wants to inject it into animals under his care might be the exact thing such laws were meant to restrict.

Plus as others have pointed out, LLMs are much better at generating something that looks like compliance documentation than they are at designing drugs (and providing generic background info on compliance processes than they are at relatively little-documented cutting edge research), so if most people aren't instantly sceptical of a yarn spun about how a layman with ChatGPT easily taught himself enough to find an RNA immunotherapy cure for his dog but then was stymied by the amount of typing involved in a 100 page document, it's a good indication of why barriers to even nominally consenting people trying experimental stuff exist.

The LLM did not design the drug. The LLM summarized some papers on how to design similar drugs, and then a dozen specialized tools were used in an established pipeline to design the drug. You people need to read the article and read the background before writing nonsense based on your assumptions.

Here's a previous comment of mine talking about personalized mRNA vaccines with useful citations: https://news.ycombinator.com/item?id=47210284

Funny how you keep skipping the unbelievable part of the story out of your replies: why would he spend 3 month hand typing a document that an LLM can definitely make at least 80% of it in one shot?

Have you ever tried writing a long, complicated document with an LLM? The last 20% takes 99% of the work.

> The last 20% takes 99% of the work.

Of course it does, since the first 80% take literal minutes! But if you compare to doing it entirely manually, it's still x5 more efficient.

Why would you do it all by hand (spending 200hours in the process…) when you're an “AI entrepreneur”…

In fairness, it pains me to see people as gullible as you are just because you like the idea of the story being true.

You don't understand how the technology in question works, and you're just making shit up because you don't want to admit to being wrong.

What are you alleging here anyways? That all the scientists quoted and photographed in the article discussing their part in making the vaccine are in on the game? That the Australian made the story up wholesale? Come on.

> You don't understand how the technology in question work

See my comment history. I do know very well how language models work. Thank you.

What I don't know is why you're claiming you disagree with the story reported here being bullshit (the story being almost literally “ChatGPT did the heavy lifting and the only reason we can't have nice thing is because red tape is blocking humanity”: “he used AI to teach himself about how a personalized vaccine could work, designed much of the process himself. […] The red tape was actually harder than the vaccine creation”), when you know very well it's bullshit because your comments describe what has most likely happened (ChatGPT did nothing much besides telling what could work and pointing towards which scientists to seek help from).

Again, literally no one question the fact that mRNA-based medicine has incredible potential, the bullshit here is not about the medicine: it's about red tape being the only bottleneck in a fantasyland where AI solves all the hard challenges.

True, and even more true in the case that you barely understand what you're doing. That's a feature rather than a bug of this sort of paperwork; the person who's simply pestered ChatGPT until it says "great idea" won't cross that threshold at all, whereas this guy [and the bioinformatics processing chain and experts in the loop he found] crossed it in his spare time. If it was just the "two hours a night typing" as quoted in the article, LLMs can do it in no time.

"ChatGPT better at finding expert advice than filling in compliance forms" and even "getting workable results from latest generation Open Source bioinformatics tools possible for smart laymen with minimal background reading; learning enough to prove they aren't dangerous only takes marginally longer" doesn't sound nearly as bad as "layman asks ChatGPT to cure his dog's cancer, only hard bit is writing enough words to convince gatekeepers" as rendered by news coverage of this (and not really elaborated on more by the TFA). A rendering which really should trigger people's bullshit filters.

Other fields crossed the computers can find potential solutions easily a lot earlier (any idiot can put dimensions into pretty dumb civil engineering tools and get answers that are probably correct; don't as me how I know!) and actually have higher barriers (no, even if you actually learn the relevant physics as well you will still need to pass some elements of your home design via someone with the right professional liability insurance linked to their experience and formal qualifications)

> I'm sorry but how does that story not smells like complete bullshit to anyone reading this?

Because it strokes the anti-establishment anti-intellectual and anti-bureaucracy zones all at once, and a lot of nerds (with love) are contrarians. Us software guys like nothing better than a story about how a smart ass with a computer undid a problem seemingly created by institutions out of whole cloth.

I'm calling absolute 100% bullshit.

Reply on news.ycombinator.com