t_mann

20 ms·

211.
by t_mann 2y ago
> All human systems are not like scientific theories that once you falsify one aspect of them, the whole entire thing is spoiled. Virtually everything that counts as a research accomplishment is predicated on studies that are vetted thro
212.
by t_mann 2y ago
The thing is, results need to be genuine to constitute progress (fake results are arguably worse than none). That's why this undermining of trust in results is fatal. You shouldn't 'offset' fraudulent research with &quot
213.
by t_mann 2y ago
Have you looked into the EU Digital Identity [0]? Would this allow creating such a service? [0] https://commission.europa.eu/strategy-and-policy/priorities-...
214.
by t_mann 2y ago
The article talks about the resignation of the president and antisemitism, but that's actually not the only recent trouble at Harvard: there were multiple independent cases of suspected academic fraud: in the business school [0], the D
215.
by t_mann 2y ago
> "I would have needed a machine-translation tool to be able to communicate with [my mother]" [a CS researcher] said. Now he is beginning to understand her without machine assistance If you need machine assistance to communicat
216.
by t_mann 2y ago
I highly doubt that there'll be any void of artifacts from our age. I'd venture a bet that the vast majority of preservation efforts (of present-day live) undertaken in the cumulative history of humankind were/are being done
217.
by t_mann 2y ago
I really like the general look and feel of masonry/waterfall layout, maybe it's because I grew up reading physical newspapers (and still do), but to me a columnar layout is just an intuitive way to divide up a page. I just wish th
218.
by t_mann 2y ago
I'd think the first application would be along the lines of Github Copilot, perhaps locally hosted - quantitative traders will write a lot of (proprietary) code, too
219.
by t_mann 2y ago
I think the mistake here was using guides. I used guides when I first played them and I remember feeling similar about them. I replayed some games again over the years, without guides, after enough time had passed so that I couldn't re
220.
Ask HN: Tips for Benchmarking LLM Performance?
1 points by t_mann 2y ago|0 comments
221.
by t_mann 2y ago
tbh, I don't even know what it means. A randomly sampled location (of what area?) has a 1:1tn chance of getting hit within any given hour? Day? Year? Over the course of an average human lifetime? Something else? I feel like the charact
222.
by t_mann 3y ago
Yeah, best not to think of any semantics wrt to the words debit and credit. Debit is left-hand side, credit is right-hand side, and just remember the rules how they work :)
223.
by t_mann 3y ago
> If you spend $X on sheep, you credit $X where? Debit it from where? You debit your 'sheep' account (maybe called sth like inventory) and you credit your cash account, or a liabilities towards suppliers account. Your equity st
224.
by t_mann 3y ago
> the point is to double the amount of work in the hopes of catching certain kinds of errors That's also how I like to think about it, as a kind of checksum mechanism. Double-entry bookkeeping originated in medieval European markets
225.
by t_mann 3y ago
That's (somewhat) true for accounts that represent stocks (assets, liabilities, not really for equity though), it's not true for accounts that represent flows (income, expenses). Income is recorded as a credit entry in an income a
226.
by t_mann 3y ago
> Definition 6: Credit An entry that represents money leaving an account. > Definition 7: Debit An entry that represents money entering an account. Not really, the meaning of debit and credit depends on the type of account: https:&#x
227.
by t_mann 3y ago
> You can ... prompt Google’s Gemini AI to make a first draft of the video for you. Can I also ask Gemini to give me a text summary of a Vid?
228.
by t_mann 3y ago
technically, it's demand that is meant here by what's elastic (or not). there's also a price elasticity of supply (eg, if tuition goes too low, some colleges will close). the way the GP used the term was understandable to me,
229.
by t_mann 3y ago
https://en.wikipedia.org/wiki/Price_elasticity_of_demand
230.
by t_mann 3y ago
> it doesn't really matter if you get an addition mostly right Back-of-the-envelope / mental math often works like that, and it's something that humans regularly use, so clearly it has some use.
231.
by t_mann 3y ago
How did you approach this, and how successful do you think it was? Do you have any tips?
232.
by t_mann 3y ago
I've tried to establish simple language codes with close relatives to deal with such scenarios, but I don't think I could rely on them remembering those. It's just such an awfully awkward conversation to be having that I susp
233.
by t_mann 3y ago
Sounds like a service that in itself requires a lot of careful checking and trust, which kind of defeats the purpose. If someone offered me such a service, I'd probably assume that it was a scam. Things like 'Regulated by FINMA&#x
234.
by t_mann 3y ago
The description of the problem seems to be missing the point: you don't need CREATE2 or run any seed search to create new Ethereum addresses for every attack. You can always create new addresses, that's a core feature of Ethereum.
235.
by t_mann 3y ago
The museum-ification of Europe continues unabated. I can't get into the mind of the people who thought that this award was a good idea, much less anyone who'd feel happy about receiving it (it's usually not the 'real&#x2
236.
by t_mann 3y ago
The basic assumption at play here is that data about you that you don't control is likely going to end up being used against you, which I think isn't unreasonable. Flawed risk metrics, even if they are only used to benefit those w
237.
by t_mann 3y ago
Yeah I don't think it will ever make sense to think about Transformer models as 'understanding' something. The approach that I suggested would replace that with rather simple logic like answer_variance > arbitrary_threshol
238.
by t_mann 3y ago
Ok, but LLMs are just tools, and I'm just asking how a tool can be made more useful. It doesn't really matter why an LLM tells you to go look elsewhere, it's simply more useful if it does than if it hallucinates. And usefulne
239.
by t_mann 3y ago
Maybe it requires understanding, maybe there are other ways to get to 'I don't know'. There was a paper posted on HN a few weeks ago that tested LLMs on medical exams, and one interesting thing that they found was that on que
240.
by t_mann 3y ago
I have to admit that I only read the abstract, but I am generally skeptical whether such a highly formal approach can help us answer the practical question of whether we can get LLMs to answer 'I don't know' more often (which

More ›