Hacker News new | past | comments | ask | show | jobs | submit login

> “When was the last time a ruling of the chair was overturned on appeal in the House?”

> Less than a minute later, the mysterious account responded with an answer — 1938 — and a decades-old edition of the Congressional Record to prove it.

It's not so much that he knows procedures well, it appears that he has some sort of didactic memory that he has focused on this topic. He would have had to basically already memorized this fact obscure enough that a congressional scholar was tweeting for an answer.




Or he knows how to use the resources available for finding the answer to this question, which is a more useful skill than memorizing facts anyway. The embarrassing part is that congressional aides don't have the same skill. I wonder if they'll stop tagging him in questions now that they know he's just a college student from the UK. Hopefully someone recognizes his talent and hires him as a staffer (there are plenty of Brits working in US government).


I agree, how to use the resources available is an underrated skill. I find an increasing part of the value I add is in being able to find answers in our vendor’s less-than-optimal documentation and search.

I do think the subject of the article has something much deeper going on.


> the subject of the article has something much deeper going on

Yes, he's got an obsession. That's the differentiator that can't be taught and that nobody can compete with, unless they've got it too.


> It's not so much that he knows procedures well, it appears that he has some sort of didactic memory that he has focused on this topic.

It's both. Source: I follow him on Twitter, and I'm a professor of American politics.


It is the former. The latter is not necessary: lots of those of us with ASD don't need anything close to an idetic memory to hold insane amounts of minutiae about some random topic or other if it happens to be the topic we're deeply interested in


Or perhaps really well trained AI? Until they named him I was expecting that to be the answer. That’s exactly the type of question a well trained LLM could answer, the proof is what gives it away as not AI, as most AIs I know are unable to offer up any sort of evidence to back up any claims they make bec they don’t actually know what they’re saying or why they “believe” something.


And fwiw both GPT-4 and Claude 3 Opus failed to give me the correct answer.


>Or perhaps really well trained AI?

We are really not there yet.

Not to mention the fact that he's been pouring through video archives to do his research, and finding the information needed for research is a big part of the work he's been doing (it's not available as readily as one would hope).


Nitpick: 'didactic' should be 'eidetic'.




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: